Training and Consultancy - Personal Blogs

Welcome to my blog pages.

This is where I write about whatever is on my mind. Most of the time it connects in some way to statistics, analysis, research, artificial intelligence, organisations, healthcare, communication or the way people behave.

The subjects may vary, but underneath them are three questions that I keep coming back to:

What do we think we know?
Why do organisations behave the way they do?
What happens when technology gets involved?

Some posts draw on my experience of working in statistics, analysis, research, healthcare and information systems. Others are simply observations about things I've encountered, read or thought about. Some may connect with the courses I teach or the work I am involved in, while others have nothing to do with work at all. Each of these blogs has a short summary so you can decide if you want to read the whole article.

If you have an idea for a future blog, or simply disagree with something I've written, I'd be interested to hear from you.

A note on AI and authorship:

The thinking in these articles is mine. The ideas, themes, examples, and conclusions come from my own experience, expertise, and reflection. AI has been used in some of these articles only to help tidy language, improve flow, and structure the content before careful human review. Used in this way, AI is no different in principle from a spell checker, a grammar checker, or a good editor. It can help communicate ideas more clearly, but it cannot replace the human knowledge, judgement, responsibility, and accountability that should sit behind every published article.

A Bit of Feedback

I recently bought a replacement ring light and nearly talked myself out of it because of several one star reviews. The strange thing was that many of the complaints turned out to be about features the product had never claimed to include in the first place. The reviewers weren't necessarily wrong to be disappointed, but they were judging the product against an expectation rather than what had actually been promised.

It got me thinking about how much trust we place in other people's opinions, and how easily feedback can influence our decisions. We use online reviews when deciding what to buy, course evaluations when deciding whether training has been worthwhile, and customer feedback when judging services. Organisations do much the same thing when deciding which suppliers, trainers or consultants they want to use again. Good feedback can open doors. Bad feedback can close them. No feedback can leave someone with very little evidence to show that they did a good job.

That last point is particularly interesting. We tend to think of the absence of feedback as neutral. It isn't necessarily. If an organisation is deciding whether to use a trainer again and nobody completes the evaluation, the trainer doesn't get the benefit of those people who may have thought the course was excellent. At the same time, negative feedback needs to be fair. A trainer shouldn't be blamed because the wrong course was booked, the room was unsuitable or a delegate expected something that was never part of the course.

If you've ever completed a course evaluation, left an online review or simply decided not to leave any feedback at all, you may find yourself looking at the whole process rather differently after reading this. Because feedback isn't just about telling someone what you thought. It can affect what happens to them next.

The Princess and the P-Value

There are some fairy tales that seem perfectly sensible when you’re seven and become considerably more questionable when you grow up. The Princess and the Pea is a good example. A young woman sleeps on twenty mattresses with a tiny pea underneath them and wakes up covered in bruises. The queen decides that this proves she must be a real princess. As a test of royal sensitivity, it’s wonderfully absurd. As an introduction to the p value, it turns out to be surprisingly useful.

The important thing is that the queen already knows there’s a pea. That isn’t what she’s trying to establish. What she wants to know is whether the young woman’s reaction provides evidence that she really is a princess. To do that, she needs a starting assumption and some idea of how unusual the result would be if that assumption were true. That’s remarkably close to the basic logic behind a statistical hypothesis test.

The trouble begins when we start asking what the resulting p value actually tells us. A p value of 0.01 doesn’t mean there’s a 1% chance that the null hypothesis is true, nor does it mean there’s a 99% chance that the alternative explanation is correct. It tells us how unusual a result at least as extreme as the one observed would be if the null hypothesis were true. That distinction sounds small, but it changes the way we should interpret the result.

The article also looks at what happens when we turn a statistical convention into a rule of nature. The familiar 0.05 threshold can make a result with a p value of 0.049 seem fundamentally different from one with a p value of 0.051, even though there’s no magical boundary between the two. Things become even more complicated when we ask lots of questions, because the more opportunities we give chance to produce an unusual result, the more likely we are to find one.

And even a statistically significant result may not be important in practice. A tiny difference can produce a very small p value when the sample is large enough, without making the difference useful or meaningful.

Obsession with the How

After more than 30 years of working with statistics, analysis and information, I’ve probably seen most of the ways data can go wrong. Sometimes the calculation is wrong. Sometimes the data’s unsuitable. And sometimes the analysis is perfectly correct but answers a question nobody actually needed answering. That last problem is perhaps the most interesting because technical competence can make an analysis look impressive without making it useful.

Much of the discussion around becoming a data analyst now focuses on learning the tools. SQL, Python, R, Power BI, Tableau and, increasingly, AI are all valuable, but knowing how to use them isn’t the same as knowing how to analyse data. Before we start manipulating datasets or building dashboards, we need to understand what the data actually represents, how it was collected and coded, what level of detail it contains and whether it’s appropriate for the question we’re trying to answer.

That distinction matters because the computer is interested in the format while the analyst needs to be interested in the meaning. A variable can be stored as a number without being a numerical measurement, a dashboard can be beautifully designed using inappropriate data and a statistical calculation can be mathematically flawless while telling us very little. Even the idea of data quality needs some qualification. Rather than simply asking whether data is good quality, we should be asking whether it’s good enough for the purpose for which we intend to use it.

The article then moves beyond choosing and understanding data to the part of analysis that is often overlooked: interpreting the result. Producing a p value, confidence interval or correlation coefficient is now almost effortless. Understanding what that number actually tells us, what it doesn’t tell us and how much weight we should give it is much harder. As AI becomes increasingly capable of handling the mechanics, that distinction is likely to become even more important.

If It Walks Like a Duck

A slice of chocolate cake that turns out to contain rather less cake than it appeared to is a surprisingly good way of thinking about artificial intelligence. The problem isn’t necessarily that what we’re looking at is false. It’s that we’ve become accustomed to assuming that appearance tells us something reliable about what lies underneath.

That’s where the old saying about ducks comes in. If something walks like a duck, swims like a duck and quacks like a duck, we naturally conclude that it’s a duck. For most of human history, that has been a pretty sensible shortcut. The difficulty is that technology can now create something with all the characteristics of a real thing without the real thing ever having existed.

The article explores what that means for the way we judge information. A photograph can show an event that never happened, a voice can say something the person never said and an AI system can produce an answer that sounds knowledgeable without actually being supported by knowledge or evidence. The problem isn’t always that the answer is wrong. Sometimes it’s that we have mistaken a convincing appearance for proof.

That takes us well beyond AI. We do much the same thing with statistics, business claims, social media posts and measures that look as though they tell us something important. Followers become a proxy for expertise, revenue for success and engagement for influence. A number can be completely genuine while still telling us very little about the thing we think it represents.

The article isn’t an argument for believing nothing. It’s about becoming more thoughtful about what we choose to believe, particularly when the consequences of getting it wrong matter. We need to distinguish between something being possible, plausible and established, and become more comfortable saying that we simply don’t know.

What Was the Question Again?

There’s an old Two Ronnies Mastermind sketch in which Ronnie Corbett’s specialist subject is answering questions before they’re asked. His answers are perfectly sensible, but they’re always answers to the question Ronnie Barker is about to ask rather than the one he has actually asked. It’s funny because we can see immediately what has gone wrong. In organisations, the same thing can happen without anyone noticing.

A dashboard gets built, a report is produced, datasets are linked or a number is requested, and only afterwards does anyone ask what question the work was supposed to answer. The calculation may be flawless and the technology may work perfectly, but none of that makes much difference if we haven’t agreed what we’re trying to find out. I’ve encountered this myself, including a dashboard that I was presented with despite nobody having first asked me what I needed it to do.

AI makes this problem rather more interesting because it has made producing answers extraordinarily cheap. Reports, presentations, analysis and recommendations that once took hours can now be created in minutes. That’s useful when the work needs doing, but it also removes some of the time and effort that used to give us an opportunity to stop and ask whether it was worth doing at all.

The same issue runs through data linkage, automation and even basic statistics. More data isn’t automatically more useful. A dashboard containing hundreds of measures isn’t necessarily better than one containing five. Linking datasets isn’t a purpose in itself. The starting point should still be the question, followed by the information needed to answer it and only then the choice of tool.

The article looks at why asking that awkward question can sometimes make you appear difficult, particularly when everyone else is excited about what the technology can do. But knowing how to use a dashboard, statistical method or AI system isn’t the same as knowing when it should be used.

Because the more capable our tools become, the easier it is to produce an impressive answer. Which makes it even more important to ask what the question was in the first place.

The Kitchen Drawer

In this article, I use the rather unfortunate contents of my kitchen drawer as a way of thinking about information management. It contains cables, chargers, adapters and various unidentified pieces of technology that might, under circumstances nobody can quite describe, prove useful one day. The problem isn’t that everything in the drawer is useless. It’s that nobody can remember which bits are useful, what they belong to, or why we kept them.

Organisations can end up with exactly the same problem. Reports, spreadsheets, dashboards, presentations and datasets are created for perfectly good reasons, but rarely acquire an equally clear decision about when they should disappear. They accumulate across shared drives, databases and inboxes until preserving everything makes finding anything worthwhile increasingly difficult.

The article explores why this happens. Creating information usually involves an explicit decision. Getting rid of it often requires another decision that nobody feels particularly keen to make. Uncertainty tends to favour keeping things. Perhaps someone will need it. Perhaps the project will restart. Perhaps that spreadsheet contains something important. So the information stays, rather like a cable that nobody can prove is useless.

AI adds another dimension because it makes producing information dramatically easier. Reports, summaries, presentations, meeting notes and emails that once took hours can now be produced in minutes. That may be useful, but it also means organisations can create information faster than people can evaluate, manage and retire it. A perfectly accurate report can still contribute to information overload if nobody needs to read it.

The article argues that the answer isn’t simply better technology for searching through everything we have accumulated. It is becoming better at deciding what deserves to be kept in the first place.

And just as I know there’s probably a useful cable somewhere in that tangled mess, organisations may eventually discover that the information they desperately need is in there somewhere too. The difficult bit is knowing which cable to pick.

The Dress that Nobody Needed

In this article, I use my wife’s habit of buying discounted clothes as an unlikely way of thinking about AI productivity. A £300 dress reduced to £120 hasn’t actually saved £180 if nobody was ever going to spend £300 on the dress. The £180 saving only exists because we compare reality with an imaginary purchase.

I suggest that some claims about AI productivity risk making exactly the same mistake. If AI reduces a three hour task to ten minutes, that is useful if the task actually needed doing. But if the task was unnecessary in the first place, we haven’t saved two hours and fifty minutes. We’ve simply spent ten minutes doing something we didn’t need.

The article explores how organisations can fall into this trap with reports, dashboards, emails, meetings and other information. Tasks that once took enough time to make people question their value can become almost effortless with AI. That can remove the very friction that encouraged us to ask whether the work was worthwhile.

The problem isn’t really new. Technology has repeatedly made individual activities more efficient while encouraging us to do more of them. Email saved enormous amounts of time, for example, but also made it possible to communicate about almost everything. AI could do something similar with organisational information, creating more reports, more analysis and more content simply because producing them is now so easy.

The article argues that productivity shouldn’t be measured simply by how quickly something can be produced or how much output is generated. The more important question comes first: why are we doing it at all?

And, just as my wife’s wardrobe is full of bargains that seemed like good ideas at the time, organisations could end up with a wardrobe full of reports that nobody really needed.

Get By With a Little Help from AI

In this article, I explore the irony of asking artificial intelligence to decide whether my own writing sounds human. What began as an attempt to understand how my writing had changed over eight years became a broader reflection on authorship, confidence and how we judge whether something was written by a person.

Looking back at my older articles, I realised that many ideas I had treated as separate had gradually become connected. Concerns about evidence, assumptions, measurement, truth and analytical judgement had been appearing repeatedly, even before AI became such a prominent part of my writing.

The irony came when AI suggested that some of the improvements in my later writing, including clearer structure, stronger connections and greater confidence, could themselves be signs of AI involvement. Some of the articles being assessed had received no AI input at all. Years of developing as a writer had somehow become evidence against me.

The article questions the growing pressure to make human writing deliberately less polished or less confident so that it doesn’t look artificial. If we change the way we naturally write because we’re worried about being mistaken for AI, then AI is already influencing our communication, even when it hasn’t written a word.

I ultimately acknowledge that this article has been AI assisted, but argue that assistance is not the same as authorship. The ideas and concerns come from years of writing and thinking, while AI has helped me examine and develop them.

You can read the full article, where I explore why we may now be asking machines to decide whether humans sound human.

Looking at the Right Thing

We often assume that good analysis is about getting the numbers right. But it’s possible to have accurate data, correct calculations and increasingly sophisticated analytical tools and still reach the wrong conclusion because we’re looking at the wrong thing.

This article brings together several of my earlier pieces, from questioning what statistics actually measure to the problems created by assumptions, confirmation bias and our growing reliance on AI. The technology may have changed, but the underlying problem hasn’t. A number can be accurate without telling us what we think it tells us. A statistically significant result may not be important. A prediction may change the very thing it was intended to predict. And AI can produce a convincing answer without necessarily understanding the question we actually wanted to answer.

The problem is that sophisticated tools can make poor analysis look very impressive. A beautifully presented graph, a detailed report or an AI generated explanation can all give us confidence that we’re on the right track. But presentation isn’t evidence, and a technically correct answer isn’t necessarily a useful one. We still need to understand the data, question the assumptions and decide whether the analysis is actually addressing the problem we care about.

That’s really what links these earlier articles together. The tools have changed, but the need to ask awkward questions hasn’t. Sometimes the problem isn’t that the answer is wrong. It’s that we’ve answered a different question from the one we should have been asking. I explore that idea in more detail in this article, bringing together examples from statistics, data analysis and AI to ask a deceptively simple question: are we actually looking at the right thing?

A Duty of Care

We often judge people by what we can see. A missed target, a change in behaviour, increased absence or someone who suddenly seems disengaged. But there may be a great deal happening that we know nothing about.

A genuine duty of care means recognising that context matters. Health, family problems, bereavement, caring responsibilities and other pressures can all affect how someone behaves at work, even when none of those things are visible.

The same applies when people become the subject of public criticism. Social media and the news can strip away context remarkably quickly, turning complicated human situations into headlines and opinions. Before judging, perhaps we should remember that we rarely know the full story.

None of this means ignoring poor performance or removing accountability. It means understanding what might be behind what we see before deciding what to do about it.

I explore this in A Duty of Care, including why organisations need to look beyond performance measures and why our responsibility to other people doesn't stop when they leave the workplace.

It Doesn't Add Up

We often assume that getting an answer means we've solved the problem. But that's not always the case.

Several of my earlier articles have approached this from different angles, looking at statistics, AI, human judgement and critical thinking. The common thread is simple: an answer can be correct and still lead us to the wrong conclusion if the question, assumptions or data are wrong.

AI makes this particularly interesting because it can produce remarkably convincing answers, even when something doesn't quite add up. That makes the old habits of good analysis more important than ever: question the assumptions, check the evidence and make sure we're actually answering the right question.

Sometimes the problem isn't the answer. It's the question we asked.

I've brought these ideas together in It Doesn't Add Up, which looks at why choosing the right question and the right tool matters just as much as producing an impressive answer.

The Responsibility to Think

AI has a rather strange way of making us feel as though we know more than we actually do. Give it a question and it can produce a fluent, confident answer in seconds. The problem is that confidence and accuracy aren't the same thing.

A polished answer can still be wrong. A convincing argument can still rest on poor evidence. And because AI makes producing plausible information so easy, being able to question what we're given is becoming more important, not less.

That means critical thinking isn't something we can hand over to the machine. We still need to ask where an answer came from, whether the evidence supports it, what assumptions are being made and what might have been left out.

I use AI because it can be genuinely useful. It can help organise ideas, explore possibilities and get a first draft moving. But usefulness doesn't make it reliable, and fluency certainly doesn't make it right. The responsibility for deciding whether something deserves to be trusted remains with us.

This article looks at why scepticism is different from cynicism, how AI can reinforce the opinions we already hold, and why asking better questions may be more valuable than simply getting faster answers.

The machine can produce the answer. We still have to decide whether it's a good one.

Rhetoric to Reality

We’ve become very good at finding positive ways to describe things that aren’t going particularly well.

A problem becomes a challenge. A challenge becomes an opportunity. A staffing shortage becomes an opportunity to develop resilience. A waiting list becomes a carefully managed patient journey. Even patients being treated in hospital corridors can apparently be described in ways that make it sound rather less concerning.

There’s nothing wrong with positive language. The problem starts when changing the language becomes easier than changing the reality.

I’ve often found myself in the uncomfortable position of being the realist in the room. Not because I’m against change, but because I think someone needs to ask whether the proposed change is actually achievable, whether the risks have been properly considered and whether it will make things better for the people who are supposed to benefit from it.

And sometimes, when things go wrong, the person who asked those questions at the beginning is still there trying to sort out the consequences.

This article looks at the growing gap between rhetoric and reality, particularly in the NHS, and asks whether we're sometimes spending too much time finding better ways to describe problems and not enough time fixing them. If the language changes but the patient experience doesn't, what exactly have we achieved?

Because ultimately, a waiting list with a different name is still a waiting list. A patient in a corridor with a more professional description is still a patient in a corridor. And calling something an opportunity doesn't make the underlying problem disappear.

When Perception Gets Ahead of Reality

History is full of decisions that seemed perfectly reasonable at the time but look astonishingly misguided in hindsight. Bloodletting, lobotomies and witch trials are very different examples, yet each illustrates how intelligent and well intentioned people can become more confident in an idea than the evidence justifies. The article asks whether the rapid adoption of artificial intelligence could be creating a similar problem, not because the technology is inherently dangerous, but because our confidence in what it can do may be moving faster than our understanding of its limitations.

A central concern is accountability. Describing AI as though it “decides”, “thinks” or “makes mistakes” can subtly shift responsibility away from the people and organisations that created the circumstances in which those actions were possible. Someone chooses the objectives, provides the data, grants the permissions, sets the boundaries and decides that the risks are acceptable. The machine may perform the action, but responsibility remains with people.

The discussion then turns to governance and ownership. AI may involve data teams, IT departments, security specialists and external suppliers, but none of those automatically owns the business outcome. Whether AI supports recruitment, healthcare, finance or another function, someone within that business area needs to understand what the system is intended to do, what could go wrong and what happens when it does.

The article argues that creating an impressive AI system is considerably easier than creating one that can be trusted. History provides a warning about what happens when enthusiasm gets ahead of scrutiny and assumptions become stronger than evidence. This article asks whether the greatest risk lies not in AI becoming too intelligent, but in humans becoming too confident about technology they haven't yet fully understood.

Sacrifice to Moloch

Sometimes perfectly sensible decisions can produce outcomes that nobody actually wanted. That is the idea behind the Moloch principle, where competition and individual incentives push people towards choices that make sense on their own but create a worse result for everyone. The ancient image of Moloch provides a striking metaphor for sacrificing something valuable because the pressure to keep up becomes impossible to resist.

The idea has particular relevance to the current race around artificial intelligence. Technology companies may privately prefer more testing, stronger safeguards and a slower pace of development, but when competitors move quickly, caution can start to look like a competitive disadvantage. The same pressures affect governments, organisations and individuals. Nobody necessarily wants to accelerate without knowing where they are going, yet nobody wants to be the one left behind either.

Drawing on the Prisoner's Dilemma, the Cold War arms race and examples from the author's own observations of AI and LinkedIn, the piece considers how these dynamics develop. Fear of missing out, investor expectations, environmental pressures and the constant message that organisations must adopt AI before their competitors can all contribute to an escalating cycle. Each decision may be rational, while the combined result becomes increasingly difficult to justify.

The question is not whether artificial intelligence should be stopped. It is whether we are allowing competition to decide its direction for us. Sacrifice to Moloch looks at what happens when everyone feels compelled to keep racing, even when nobody is quite sure where the race is taking them, and asks what we might end up giving away simply because slowing down no longer feels like an option.

Talking Garbage

Artificial intelligence is developing rapidly, and the NHS is understandably interested in what it might offer. AI can process information, recognise patterns, support research and perform tasks at a scale that would be difficult for people to match. But before asking what AI can do for healthcare, this article asks a more basic question: what information are we actually giving it to work with?

Drawing on more than thirty years working with data, statistics and information standards, including experience with the NHS Information Standards Board, the article argues that the quality of the information underneath an AI system matters just as much as the sophistication of the technology itself. Inconsistent definitions, missing information, changes in recording practices and systems that cannot properly share meaning can all affect the conclusions produced from data.

The article uses the familiar idea of “garbage in, garbage out”, while pointing out that AI creates an additional problem. Poor information can produce an answer that looks convincing. Unlike an obvious numerical error, an AI generated response can be confidently written and entirely plausible while still being based on flawed or misunderstood information. The discussion also considers interoperability, shared definitions, data standards, public trust and the particular complexity of information systems across the NHS.

Better data isn't simply an AI requirement. It would improve clinical care, research, planning and decision making whether AI succeeds or not. The most advanced system in the world will struggle if the foundations underneath it are weak. Read the full article to see why the future of AI in healthcare may depend less on how clever the machines become and more on whether we have been wise enough to make the information feeding them trustworthy.

Mind Your Language

Artificial intelligence has advanced at extraordinary speed, with systems now able to write, analyse information, create images and videos, generate software and carry out increasingly complex tasks. At the same time, the language we use to describe these systems has become increasingly human. We talk about AI thinking, understanding, learning, reasoning and showing empathy, but the article asks whether those descriptions are actually justified.

The article explores why people naturally anthropomorphise AI, particularly when systems communicate through language that sounds remarkably human. A machine can produce a moving response about grief, offer apparently compassionate advice or provide an explanation that feels thoughtful. But producing convincing language is not the same as experiencing emotion, understanding a situation or genuinely caring about the person asking the question.

The distinction matters when AI is used in areas such as healthcare, education, recruitment, lending, policing and criminal justice. These systems can identify patterns, process information and reproduce language associated with empathy, but they don't have lived experience or human judgement. The article also considers the emergence of AI agents that can take actions rather than simply answer questions, raising further questions about responsibility when those actions go wrong.

The argument is not against AI. Its capabilities are remarkable and it can be extremely useful when used appropriately. The concern is that the technology doesn't need us to exaggerate what it can do. The full article explores why describing AI accurately matters, because the language we use to talk about machines can influence how much we trust them, where we use them and who we hold responsible when something goes wrong.

Welcome to the LinkedInverse

LinkedIn was created to help professionals build connections, share knowledge and develop relationships. Yet the way the platform rewards engagement can sometimes produce almost the opposite result. This article introduces the idea of the “LinkedInverse”, a version of LinkedIn where appearance, reach and easy engagement can matter more than genuine connection or useful insight.

Drawing on my own experience of using LinkedIn as a trainer, consultant and writer, the article looks at the difference between the professional life people actually experience and the highly polished version presented online. It questions whether comments, reactions and connection numbers really tell us anything about the value of what is being shared.

The article explores some of the familiar features of the LinkedInverse: generic motivational statements, transactional networking, people presenting themselves as experts in almost everything, perfectly packaged stories about lessons supposedly learned from everyday encounters, “proud and humbled” announcements and the recurring airport photograph accompanied by profound thoughts about leadership. The humour is used to make a more serious point about how platform incentives influence behaviour.

The argument is not that LinkedIn is inherently bad, or that people should stop using it. There are plenty of thoughtful people having worthwhile conversations. The concern is that what generates engagement is not necessarily what creates learning or meaningful relationships. The full article asks whether we are creating genuine professional connections, or simply learning how to create the appearance of them, and welcomes readers to the LinkedInverse.

Babies with Power Tools

Artificial Intelligence is advancing at a pace that is challenging our ability to understand, regulate and prepare for it. Across LinkedIn, many people are saying that organisations which don't adopt AI quickly risk being left behind. While there is a strong argument for exploring the opportunities AI provides, there is also a need to think carefully about what we are adopting and how we use it.

This article explores the idea that AI is not simply another tool. Unlike a hammer, chainsaw or power drill, we are creating systems that can analyse information, generate content and support decisions in ways that increasingly resemble tasks requiring human judgement. The challenge is that even the people who build these systems don't always fully understand why they produce particular outputs.

Using the analogy of "babies with power tools", the article considers the risks of developing powerful technology faster than our ability to understand its consequences. It looks at issues around healthcare, jobs, trust, regulation and the importance of maintaining human responsibility when using AI.

AI offers enormous possibilities, but capability without understanding creates risks. The question isn't whether we should use AI, but whether we are ready to use it responsibly.

Read the full article to explore why the most important part of AI adoption may not be the technology itself, but the judgement we apply when using it.

Guardrails Before Gadgets

Artificial intelligence is becoming increasingly embedded within the NHS, from patient triage through the NHS App to software that drafts clinical notes. The potential benefits are obvious: better access, less administrative burden and more time for clinicians to spend with patients.

This article looks beyond the headlines to ask a different set of questions. What happens when AI gets things wrong? How should systems that influence patient care be regulated? Where does accountability sit when recommendations come from software rather than people?

Drawing on current NHS proposals, existing medical device regulation and wider research into AI in healthcare, the article explores the balance between innovation and patient safety. It considers issues such as clinical validation, governance, digital exclusion, automation bias and the safeguards needed as these technologies become part of routine care.

The argument is not for or against AI. It is that healthcare has always demanded a higher standard because mistakes can have life changing consequences. The full article examines why the real challenge is not simply adopting AI, but ensuring the guardrails are in place before patients are asked to trust it.

Seeing the Forest and Missing the Trees

Bryan Dodds continues his unintended journey through successive layers of healthcare information systems, each one shaped by a different attempt to solve the same underlying problem. Earlier encounters showed him records that travelled with patients in a single, continuous form, whether in paper envelopes, maternity notes, or health passports. Those systems worked because they assumed care and information moved together.

The next shift drops him back into a recognisably modern NHS, but one where that unity has been replaced by fragmentation. Information is now distributed across multiple highly capable systems, each holding its own version of clinical truth. GP records, hospital systems, pharmacy data, and specialist services all operate with internal consistency, yet none reliably contains the full, current picture. The result is not obvious failure, but a quieter misalignment that only becomes visible when information is needed at the point of care.

Bryan observes how this arrangement emerged from a genuine need. Population level reporting, performance monitoring, and system planning required structured, comparable data. Early mechanisms such as Korner returns reduced complex clinical reality into standardised summaries that could be aggregated nationally. Later developments, including MIQUEST and incentives like the Quality and Outcomes Framework, reinforced the need for consistency, coding, and extractable data.

This shift made it possible to understand healthcare at scale. Trends, outcomes, and variation across populations became visible in ways that were previously impossible. But the same process also distanced data from the lived reality of individual patients. Clinical narratives were translated into codes and fields optimised for analysis rather than consultation, creating parallel versions of the same information: one for care, another for measurement.

Bryan begins to see the consequence of this dual system. At scale, healthcare sees the forest with increasing clarity. Up close, it depends on fragmented records, patient recollection, and interpretation to understand individual trees. The system has become highly effective at population insight while remaining dependent on imperfect continuity at the bedside.

The full article follows Bryan as he recognises the tension between visibility and proximity, and what is lost when the ability to see everything comes at the cost of seeing any one thing clearly.

The Patient That Carried The System

Bryan Dodds finds himself pulled into another shift in time without ceremony or warning, still carrying the uneasy conclusion from his previous experience: information behaves less like a static asset and more like something that moves with purpose. The idea that information is not furniture but luggage has begun to take shape in his mind, and this new jump tests it further.

Across different settings and eras, he encounters the same pattern in different forms. In the UK, the Personal Child Health Record, known as the Red Book, sits quietly in a parent’s hands as a portable account of a child’s health. Vaccinations, growth, and milestones travel with the family rather than being trapped inside institutions. In other settings, particularly parts of East Africa, similar health passport systems serve the same function, ensuring continuity where formal system integration is limited or absent.

What links these examples is not technology or policy design but movement. Patients move between homes, clinics, and communities, and the record moves with them. The system does not rely on perfect communication between organisations. It relies on the patient carrying continuity themselves.

Bryan notices how strikingly consistent this principle is across very different healthcare environments. No shared blueprint appears to exist. No global agreement has been reached. Yet the same solution emerges repeatedly: if care moves, the information must move too. Anything else eventually breaks under the weight of fragmentation, assumptions, and administrative distance.

The contrast with modern digitised systems becomes more uncomfortable the longer he observes it. Contemporary healthcare invests heavily in storing, securing, and structuring information across multiple platforms. Yet in doing so, it often creates conditions where information exists everywhere and arrives nowhere reliably. The patient, intentionally or not, becomes the thread that reconnects these fragments.

What begins to stand out is not a technological gap but a structural misunderstanding. Systems have been designed around places, while care happens through movement. The older portable models feel almost disarmingly direct in comparison. They do not attempt to solve communication between institutions. They sidestep it entirely by ensuring the record travels with the person it describes.

The full article follows Bryan as this realisation deepens, revealing what happens when healthcare systems forget that continuity is carried by people long before it is stored in systems.

When Luggage Becomes Furniture

Bryan Dodds continues his unintended journey through time after triggering a system labelled “Synchronise Historical Patient Streams”, a decision that drops him into the 1960s rather than any kind of digital utopia. What he finds is not a primitive healthcare system, but one that operates on a radically different assumption: information follows care, not the other way around.

In this world, medical records are physical objects carried alongside patients. Lloyd George envelopes move between surgeries, hospitals, and home visits as clinicians add handwritten notes, referrals, and observations in real time. It is messy, inconsistent, and occasionally baffling, but it holds together because the patient remains the central point of continuity. Nothing is “integrated” in the modern sense, yet nothing is lost in the idea of where care is happening.

Bryan begins to contrast this with his own era, where healthcare information is split across multiple systems, each highly sophisticated and independently convinced of its completeness. Hospitals, GP practices, mental health services, and social care all maintain separate digital realities. The information exists, but rarely in the same place at the same time, and retrieving it often requires navigating layers of systems, permissions, and assumptions about ownership.

The result is a paradox. Modern healthcare has become extremely good at storing and classifying data, yet less reliable at ensuring it appears when and where it is needed. Patients, knowingly or not, have become the carriers of their own fragmented records, reconstructing their histories across appointments and organisations. What was once physically transported now has to be mentally reassembled.

The older system carries its own flaws, from lost files to illegible handwriting, but it retains a simple logic: records move because people move. Bryan realises that this idea has been replaced by something more complex but less reliable, where information is treated like static furniture rather than something that travels.

The full article follows Bryan as this realisation deepens, uncovering what happens when systems prioritise storage over movement, and why the simplest ideas about care may still hold the most uncomfortable truths for modern healthcare design.

The Strange Case of the Time Traveller and the Medical Record

Bryan Dodds begins this series not as a heroic adventurer, but as someone trying to complete a routine NHS data migration with the quiet confidence of a man who has clicked “final version” more times than he cares to remember. One ill judged press of a button labelled “Synchronise Historical Patient Streams” pulls him out of a world of dashboards, colour coded certainty, and systems that behave well only when observed by senior management.

He finds himself in a GP surgery from an earlier era, surrounded by filing cabinets that feel less like storage and more like guardians of human memory. Here, information is not abstract or remote. It is physical, carried in Lloyd George envelopes, handwritten, annotated, and passed from clinician to clinician with an almost unsettling intimacy. Every folder is a person’s story, held together by ink, judgement, and an assumption that care travels with the patient, not the system.

What starts as a bewildering encounter becomes something more uncomfortable. Bryan sees a system that looks chaotic but functions on a simple principle modern technology often forgets: records follow people. The GP reads, writes, and thinks in the same moment. Midwives, consultants, and nurses all add their own layers to the same evolving document. There is no illusion of perfect structure, yet continuity of care remains intact.

For someone steeped in digital transformation, the revelation is unsettling. The “primitive” system is not primitive at all. It is built around movement, context, and shared understanding rather than fragmented datasets and disconnected platforms. The moment lands when Bryan realises that care itself moves, and the information has the decency to move with it.

The full article continues Bryan’s journey into a world where assumptions about progress begin to unravel, and where every system he thought he understood starts to look suspiciously fragile.

When Everything Else Falls away

Some periods strip life back to what actually matters. Work, plans, visibility, and momentum all become secondary when illness and family crisis take over. That is the reality described here, where a sudden loss of capacity coincides with a partner’s serious health complications and a severe ME relapse. The result is not a chosen pause, but a forced narrowing of life to immediate needs.

Days become structured around hospital visits, domestic responsibilities, and managing energy that is no longer reliable. Even basic tasks carry a cost that has to be weighed in advance. Work does not disappear, but it loses its dominance. It is fitted in where possible, often imperfectly, around everything else that cannot be postponed.

Alongside the physical strain sits a quieter pressure. Responsibility does not ease simply because capacity has fallen. Supporting someone through illness while managing one’s own health limits creates a constant background tension. There is no clean separation between roles, only shifting priorities from one hour to the next.

In that environment, professional identity starts to feel less stable. Output slows, visibility drops, and usual markers of success such as recognition or awards lose relevance. What remains more visible is the gap between expectation and reality, and the effort required just to keep going.

There is also the less discussed emotional weight that comes with chronic illness and reduced capacity. Guilt about letting people down. Frustration at cancelled work. A sense of falling behind in systems that reward consistency and output, even when circumstances make both difficult.

What comes through most clearly is a reordering of priorities that is not theoretical but lived. Health sets the boundaries for everything else. Family becomes the fixed point. Work continues, but in a reduced and reshaped form.

The full article carries the detail, nuance, and honesty behind this experience, and it is worth reading in full to understand how quickly life can change shape, and what it takes simply to get through it when everything else falls away.

Disruption to Destruction

This piece looks at how organisations that embrace “disruption” can drift into instability and long term damage when change is driven more by narrative and authority than by grounded understanding.

It describes a familiar pattern. An external hire or consultant arrives with a mandate to shake things up. Early momentum is strong, old habits are challenged, and leadership often celebrates the energy and decisiveness. But as the new narrative takes hold, something more fragile begins to form underneath. Established processes are dismissed, experienced staff are reframed as obstacles, and informal knowledge built over years is quietly pushed aside.

As influence consolidates, perception starts to matter more than reality. Selective use of data and carefully framed interpretations create a picture of crisis that only the disruptor appears able to fix. Those who question this framing risk being labelled as resistant or outdated, while others withdraw or adapt their behaviour to avoid conflict. Over time, organisations lose the people who understand how things actually work.

The result is a gradual weakening of trust, shared understanding, and ethical stability. Decision making becomes more centralised around a single narrative, while alternative perspectives fade. What looked like renewal begins to resemble erosion, with capability and memory slowly draining away.

The article draws on real world parallels from corporate change, public services, and leadership culture, showing how the promise of transformation can become self defeating when accountability and institutional knowledge are undervalued.

It closes by arguing that disruption only has value when it is anchored in humility, transparency, and respect for existing expertise. Without that balance, the shift from disruption to destruction can be subtle but severe.

My Way or the AI Way

This article explores a growing tension in modern organisations: what happens when human judgment collides with algorithmic certainty. AI systems are increasingly embedded in high stakes decision making across healthcare, law, employment, finance, and the voluntary sector. They appear confident, data driven, and objective, yet they carry no responsibility for outcomes. The result is a subtle but powerful shift in authority, where professionals are asked to balance their own judgment against machine generated recommendations that feel increasingly difficult to ignore.

Across sectors, similar dilemmas emerge. In healthcare, clinicians may be influenced by AI diagnostics even when they disagree based on experience. In mental health settings, automated risk flags can override professional intuition, creating uncertainty about liability when outcomes are poor. In recruitment, legal cases and research have shown that algorithmic systems can reproduce bias, raising questions about fairness and accountability. Even in law, reliance on AI generated material has led to serious professional consequences when errors were not verified.

The core issue is not whether AI is useful, but where responsibility sits when it is wrong. In almost every scenario, accountability remains with the human, even when the system shaped the decision. This creates a paradox: following the AI may feel safer, but ignoring it can be just as risky.

The article argues that organisations must resist passive compliance with automated outputs and instead build cultures where judgment, transparency, and documentation are central. Oversight cannot be symbolic. It requires understanding, challenge, and the confidence to disagree with systems when necessary.

The full article goes deeper into real world cases and regulatory responses, showing how this tension is already shaping professional life.

Breaking the Cycle

This final article in the series brings together the themes explored so far: decisions made before evidence, uncertain figures becoming accepted fact, and the absence of meaningful evaluation. Taken together, they describe a familiar organisational loop where action leads, evidence follows, and reflection is often missing altogether.

Across government, healthcare, charities, and business, decisions are frequently made under pressure and then supported by data assembled afterwards. Targets appear early, uncertainty is quietly reduced, and early assumptions gradually harden into accepted truth. Over time, repeated statements and internal reporting give those assumptions the appearance of certainty, even when their origins are weak or unclear.

The same cycle repeats in evaluation. Reviews are often delayed, limited, or shaped to confirm what has already been decided. By the time learning could meaningfully influence direction, attention has usually shifted elsewhere. This creates organisations that are active but not necessarily informed by what has actually worked.

There are examples that point in a different direction. Some public bodies, research programmes, and charities now publish uncertainty, share negative findings, and build evaluation into design rather than treating it as an afterthought. Smaller scale experiments, when properly measured and reported, offer a more honest way of building knowledge than large irreversible transformations.

The article argues that breaking this cycle is less about new tools and more about culture. It requires permission to express uncertainty, willingness to publish uncomfortable findings, and incentives that reward learning rather than just delivery. Evaluation becomes not a final step, but part of how decisions are made from the beginning.

If you are following this series, this final piece pulls the threads together and shows what change might actually look like in practice. The full article explores the patterns in more depth, with real examples and practical ways organisations could shift from repetition to genuine learning.

The Missing Mirror

This reflects on a recurring problem across government, healthcare, charities, and business: organisations often talk about learning and improvement, yet rarely build systems that genuinely test whether their decisions worked. Instead, projects are launched with confidence, described as success stories early on, and then quietly moved past without rigorous follow up. What should be a cycle of learning becomes a cycle of announcement and amnesia.

In the NHS, repeated reorganisations illustrate the pattern. Structures are changed, services reshaped, and new systems introduced, but evaluation is often incomplete or inconsistent. Large technology programmes have followed a similar path, where ambitions were clear but evidence of impact arrived late, fragmented, or not at all. Local government outsourcing, charity initiatives, and corporate transformation programmes show the same rhythm: delivery is recorded, but outcomes are rarely examined with equal care.

Once a decision is made, organisations tend to search for confirmation rather than challenge. Evaluation becomes something that justifies choices rather than tests them. Over time, this leads to a culture where activity is mistaken for impact, and where “lessons learned” rarely alter future behaviour. Staff become cautious about raising doubts, and analysts are drawn into reinforcing rather than questioning the prevailing narrative.

There are exceptions, particularly where independent evaluation is built in from the start and results are published even when they are uncomfortable. These cases show that genuine learning is possible when scrutiny is treated as part of the process rather than an afterthought.

The central message is simple: without structured, honest evaluation, organisations lose sight of what is actually working. They move, but do not necessarily improve.

The full article goes much further into real world examples, including health, government, and private sector cases, showing how this pattern repeats and what it would take to break it.

How Guesses Become Gospel

This article continues When Decisions Come First, looking at what happens after policy choices are made: how early assumptions harden into accepted truth.

It often starts with a cautious figure in a report or slide deck. A tentative estimate becomes a confident statement, then a repeated “fact”. Over time, the uncertainty that surrounded it fades. What remains is something that sounds established simply because it has been repeated often enough.

This pattern shows up across sectors. In the NHS, early claims about savings from digital transformation have persisted long after the evidence behind them was questioned. Targets such as A&E waiting times shifted attention from patient experience to reported performance, with organisations adapting to the measure rather than the reality it was meant to capture.

Local government and charities follow similar patterns. Recycling targets are quietly reshaped through changing definitions. Digital programmes are described as successful on partial or selective data. Early positive signals are amplified, while inconvenient findings are sidelined.

Private companies behave in much the same way. Environmental and performance claims often rest on narrow datasets or convenient baselines, yet once repeated in reports and marketing, they become accepted fact.

Repetition is the engine. Once a figure appears in multiple documents, it gains authority through familiarity rather than verification. Reports begin citing earlier summaries of themselves, forming a loop where origins are lost.

The result is a shift from evidence-led decisions to assumption-led justification, where narratives survive longer than the data that created them.

The full article explores how this happens in practice, why organisations fall into it, and what it would take to break the cycle before “guesses become gospel”.

When Decisions Come First

Most institutions claim to follow “evidence-based decision making”, yet the sequence is often reversed. Decisions are made first, and evidence is then shaped to support them. What looks like analysis is frequently justification.

In healthcare, this pattern is familiar. Organisational restructures are introduced with confident language about efficiency or modernisation. In one NHS setting, multidisciplinary teams were replaced with a segmented “factory model”. Data collection, management, and analysis were split into separate units. The expectation was improved productivity. No independent evaluation was commissioned. When delays emerged, blame settled on downstream analytical teams rather than the design of the system itself.

A similar dynamic appeared in hospital capacity planning, where bed numbers were fixed in advance and modelling work was used to validate the decision rather than test it. When assumptions produced higher estimates, they were adjusted until the model aligned with the original figure.

Across public services, charities, and private organisations, the pattern repeats: restructures are announced, success is declared early, and evaluation is either absent or selective. Reports highlight efficiency gains while ignoring unintended consequences.

The drivers are familiar. Speed, political pressure, reputational risk, and the discomfort of revisiting announced decisions all discourage genuine testing of assumptions. Once a narrative is set, data tends to serve it.

The result is a quiet inversion: evidence no longer shapes decisions, it reinforces them. Analysts are left navigating between what the data shows and what leadership has already decided.

Organisations that resist this pattern tend to do one thing differently: they build evaluation into the design phase, before decisions harden. Read the full article here.

Beyond the Buzzword

This article is part of the From Code to Consequence series, drawing on the July 2025 panel discussion where experts in data governance, healthcare analytics, and public sector transformation asked a deceptively simple question: what do we really mean by “AI”?

The term has become so stretched by marketing that it now covers everything from basic automation to complex generative systems. One panellist captured the absurdity with a simple observation about an “AI-enabled toothbrush”, highlighting how easily the label is applied without clarity on what the system actually does.

This lack of precision matters. A rule-based scheduling tool and a large language model are fundamentally different, yet both are often grouped under the same heading. That confusion leads to weak governance, misplaced trust, and poorly judged risk.

The panel described AI as a spectrum. At one end are deterministic systems that follow explicit rules. At the other are probabilistic models that generate outputs based on patterns in data. These systems do not understand meaning or context; they calculate likely continuations. As one speaker put it, “it’s not about certainties.”

That distinction is critical. Probabilistic systems can produce different answers to the same question, and occasionally fabricate information. Their outputs may appear confident, but confidence is not reliability. As another panellist warned, “you only have to be wrong once.”

Human judgment remains essential because it brings context, scepticism, and accountability. AI does none of these things. Without a clear definition of what AI is in each use case, organisations cannot govern it, test it, or take responsibility for it in any meaningful way.

The starting point for governance is not regulation or tooling, but understanding what system is actually in front of you.

Epistles, Ethics and Algorithms

This article revisits themes first explored in Epistles, Enquiries and Ethics (2021), updated in the context of rapid advances in data science and artificial intelligence, and informed by insights from the From Code to Consequence panel discussion on AI governance and ethics.

In the early NHS of the 1990s, meaningful data was scarce. Information lived in paper records, isolated systems, and professional experience. Decisions relied heavily on judgement because evidence was limited and slow to arrive.

Today, the situation is reversed. Data flows continuously from clinical systems, consumer devices, and wider digital platforms. The challenge is no longer availability but interpretation: separating signal from noise, recognising bias introduced through linkage, and avoiding false precision from convenient metrics.

This abundance is complicated by declining data quality. Online content is often optimised for reach rather than accuracy, while AI systems trained on mixed datasets absorb those weaknesses. The growing use of synthetic data adds another risk, as models begin learning from their own outputs, reducing diversity and reliability over time.

The ethical question has therefore shifted. The danger is no longer just missing information, but over trusting what is available. AI systems amplify this risk because they produce fluent, confident outputs that can mask uncertainty. They assist judgement, but do not replace it.

For many years, legality defined acceptable use. Experience now shows that lawful is not always ethical. Responsible practice requires clarity of purpose, awareness of impact, and honesty about uncertainty.

As St Paul observed, “not everything is beneficial.” The article argues that permission does not equal wisdom.

The From Code to Consequence panel reinforced this point: responsibility sits with those who build and deploy systems, not the systems themselves. Governance, AI literacy, and human oversight are therefore not optional additions, but core requirements for safe and meaningful use.

Avoiding the Light Touch

This article, drawn from the From Code to Consequence panel hosted by In the Know Limited, explores a growing global tension in AI governance: what happens when some countries regulate tightly while others take a lighter approach.

At the heart of the discussion was a simple concern. AI systems don’t respect borders, yet regulation does. That mismatch raises an uncomfortable question: if companies can shift development or deployment to less regulated environments, does stricter governance still matter?

The panel’s view was consistent. It does, and arguably more than ever.

One analogy compared regulation to vaccination: partial uptake weakens the overall effect. Another speaker noted that AI doesn’t stay local. Systems, data, and decisions move across jurisdictions, meaning weak regulation in one place can create risk everywhere else.

The EU AI Act was central to the discussion. It follows a risk based model, banning some uses outright and placing strict obligations on high risk applications in areas such as employment, education, and healthcare. Importantly, it applies not just to EU based organisations but to any system affecting EU citizens. That extraterritorial reach changes the equation for global businesses.

Other frameworks sit alongside it. GDPR and UK data protection law govern personal data use, while sector rules such as medical device regulation apply where AI influences diagnosis or treatment. In those cases, oversight, traceability, and human review are not optional extras but legal requirements.

A recurring theme was that compliance is not geography dependent, it is impact dependent. AI is judged by what it does, not where it is built.

The panel also highlighted practical risks: performance drift, over confidence in polished outputs, and failures that emerge only after deployment. These issues reinforce the need for continuous oversight rather than one off checks.

Ultimately, the conclusion of the article is clear. Light touch regimes elsewhere do not weaken stricter ones. They make them more relevant. As AI spreads across borders and industries, governance based on risk, accountability, and human oversight becomes the only workable model for managing real world impact.

AI Compliance Frameworks: Fact or Fiction

This article builds on the From Code to Consequence panel discussion (July 2025), where experts from healthcare, public policy, and data governance explored a simple but uncomfortable question: are AI compliance frameworks already in place?

The short answer is yes, but not in one place, and not in a way that makes life simple.

The discussion opened with a reminder that AI is not a single thing. Some systems follow fixed rules, others generate outputs based on probability. That difference matters, because the law does not treat all systems the same. What it does insist on is accountability: someone, not something, is responsible for outcomes. As one panellist put it, “You only have to be wrong once.”

From there, the conversation mapped out the current regulatory landscape. In the EU, the AI Act sets risk based obligations, bans certain uses outright, and adds strict requirements for high risk areas such as employment, education, and healthcare. It is already active in stages, with penalties reaching millions or a percentage of global turnover. In the UK, GDPR and the Data Protection Act still govern personal data use, including rights around automated decision making. Cookie rules add another layer for digital services. Other regimes, such as medical device regulation, apply when AI influences clinical decisions.

A key theme was that compliance is not a document or a final sign off. It is continuous oversight. Systems drift. Outputs degrade. Users over trust polished results. Without monitoring, risk accumulates quietly until it becomes visible through failure.

The panel kept returning to two practical requirements: clear decision ownership and traceable records. Not code transparency, but clarity about how a system is used, where human judgement sits, and how decisions can be challenged or reviewed.

Compliance, in this sense, is already here. The gap is not legal absence, but organisational readiness.

When Oversight is an Afterthought

A panel discussion, From Code to Consequence (15 July 2025), brought together experts in AI, data governance, healthcare, and public sector transformation to examine a simple but uncomfortable question: what happens when oversight is treated as an afterthought?

The answer ran through every theme in the discussion. When governance is bolted on at the end, rather than built in from the start, failure doesn’t appear as a single event. It builds quietly. Small issues compound until they become operational disruption, legal exposure, and loss of trust.

The panel broke oversight into two linked layers: the data layer, covering where data comes from and whether it is valid and consented; and the system layer, covering how AI is built, deployed, and monitored. Treating these separately, or delaying governance until deployment, leaves gaps where risk accumulates unnoticed.

Several failure patterns were highlighted. AI systems drift as data changes, slowly degrading performance without obvious warning. Outputs can look authoritative even when they are wrong, creating false confidence. Poor governance also introduces security risks, regulatory breaches, and reputational damage that can take years to repair. Perhaps most importantly, it creates uncertainty about accountability when things go wrong.

A recurring point was that AI is not accountable in itself. Responsibility must sit with people, clearly defined before deployment, not after failure.

Oversight is often framed as friction, but the panel argued the opposite. Strong governance enables safer, faster, more confident adoption of AI. It is not a brake on innovation but the condition that makes it viable.

The full article explores these ideas in more depth, including practical implications for organisations already deploying AI at scale.

When Machines Speak Medicine

“When Machines Speak Medicine” is a reflective piece exploring how large language models behave when they are tested with deliberately misleading or “adversarial” clinical information.

The article describes a structured experiment using 300 physician-validated medical scenarios, each containing a single hidden error such as a false lab result or incorrect diagnosis. Multiple AI systems were then tested under different conditions: standard use, carefully engineered prompts designed to reduce mistakes, and settings where randomness was removed to make outputs more deterministic.

Across these setups, the focus was not just whether the models got answers right or wrong, but how they handled uncertainty and contradiction in clinical reasoning. The work highlights a key concern: even when models appear confident, they can still be influenced by subtle inaccuracies in the input data, producing fluent but unreliable medical interpretations.

The broader message is about trust in AI within healthcare. Medicine depends on precision, context, and careful validation, while language models operate by predicting plausible sequences of words rather than verifying truth. This creates a tension between how convincing outputs can appear and how safe they actually are in practice.

The article ultimately raises questions about where AI fits in clinical decision making. It suggests these systems may be useful as supportive tools, but only within tightly controlled environments where human oversight remains central.

For the full argument, including the implications for medical safety, AI reliability, and how clinicians should interpret machine-generated advice, it is worth reading the complete piece.

Leading with Integrity

Integrity at work is often spoken about in abstract terms, but in practice it shows up in how people treat each other when there is pressure, power, or conflict involved. Most of us recognise, whether through faith, conscience, or simple experience, that dignity, fairness, and respect are not optional extras. They are what hold teams and relationships together.

In many workplaces, however, those principles weaken. Bullying, exclusion, and subtle undermining are often dismissed as personality clashes, humour, or “just how things are here.” Over time, these behaviours shape culture far more than formal policies do. People become quieter, more cautious, and less willing to speak up. Trust erodes in ways that are not always visible until performance and morale begin to fall away.

Silence plays a role in this. Watching harmful behaviour without challenge sends its own signal. Inaction becomes part of the system that allows it to continue. Leadership decisions then matter even more, because they define what is tolerated and what is not.

There is also a recurring organisational pattern where those who achieve short term results through force or intimidation are rewarded, while those who prioritise care, fairness, and stability are overlooked. Over time, this creates fragile cultures that depend on fear rather than commitment.

Ethical leadership offers a different path. Often described as servant leadership, it focuses on enabling others to do their best work, removing obstacles, and protecting wellbeing. It requires attention, courage, and consistency, not just authority.

This matters because workplace culture is never neutral. It either protects dignity or erodes it.

The full article explores these tensions in more depth, including personal experience, leadership responsibility, and what it takes to build environments where people can actually thrive.

Beat the System

Leadership is being reshaped by artificial intelligence, but not always in ways that serve people well. Systems now produce forecasts, summaries, and performance metrics at speed, and the pressure on leaders is to match that same mechanical efficiency. The risk is that leadership itself becomes reduced to output management, where people are treated as data points and performance is judged only through dashboards.

AI excels at processing information, but it cannot understand fear, uncertainty, or motivation. Yet organisations increasingly lean on it for decisions that affect roles, workloads, and restructuring. When that happens, the human side of work quietly erodes. Conversations become shorter, empathy becomes optional, and staff begin to feel like replaceable components in a system optimised for speed.

“Beating the system” in this context is not about gaming rules or finding shortcuts. It is about resisting the idea that leadership should be modelled on machines. The qualities that matter most now are the ones AI cannot replicate: patience, judgement, empathy, and the ability to notice what is not being said.

Servant leadership offers a counterpoint. It shifts focus away from authority and towards support. The leader’s role becomes clearing obstacles, enabling growth, and asking what people need to do their work well. That includes training, reassurance, and honest communication during change, not just targets and performance reviews.

Examples from organisations like Microsoft and HCL Technologies show that cultures built on empathy and trust can also deliver strong results. Even in healthcare, compassionate leadership has been linked to better staff wellbeing and patient outcomes.

AI can handle tasks. It cannot carry people through uncertainty. That remains a human responsibility.

Mind the Gap

Data has become central to how organisations operate, with dashboards, models, and AI systems now shaping decisions at every level. On the surface, this looks like progress: faster analysis, cleaner visuals, and constant streams of insight that appear to remove uncertainty from decision making. But the reality is more complicated. These systems do not understand the data they process. They identify patterns and produce outputs that look precise, even when the underlying interpretation is weak or incomplete.

This creates a growing gap between appearance and understanding. Many teams now depend on automated tools, copied scripts, and AI generated analysis without fully engaging with the logic behind them. When outputs are presented with confidence and visual clarity, they are often accepted without challenge. The habit of questioning results is slowly being replaced by trust in the tool itself.

Across sectors such as healthcare, finance, and public services, this shift has already contributed to costly errors. Statistical mistakes have shaped medical decisions, biased datasets have influenced financial outcomes, and governance failures have gone unnoticed until after significant damage has been done. Artificial intelligence adds further complexity, producing fluent and persuasive responses that can be wrong in subtle but hard to detect ways.

The core issue is not the technology, but the erosion of analytical judgement. Knowing how to operate tools is not the same as knowing how to interpret their outputs, test assumptions, or recognise when something does not make sense. As organisations prioritise technical fluency over reasoning, that gap becomes more significant.

The full article explores this tension in detail, showing why confidence in data tools is rising faster than understanding, and why judgement remains the most important safeguard in a data driven world.

Paying the Price

Most organisations try to trim training down to something quick and cheap. A checklist, a short guide, a cheat sheet that looks neat in a folder. It feels efficient in the moment, especially when budgets are tight and deadlines are looming. The problem is that it creates a sense of readiness without the substance that actually keeps people safe or effective.

A UK outsourcing firm learned this the hard way after a cyber-attack exposed data on over 100,000 employees. The breach began with a simple phishing email, but what made it worse was the lack of proper training. Only one of the finance staff involved had ever received security awareness training. That gap turned a routine scam into a major incident.

Similar patterns show up across data work, governance, and AI systems. In healthcare, poorly coded patient data has masked critical side effects in drug trials, delaying the recognition of serious risks. In another case, a flawed statistical approach in a hospital setting led to the rollout of an ineffective treatment, costing millions and offering no real benefit. In finance, an automated loan system was later found to be biased against certain groups, despite passing internal checks that no one deeply understood. Each case had documentation, but not understanding.

The attraction of shortcuts is easy to see. A cheat sheet or summary gives the impression that people are prepared. But it rarely helps when reality stops matching the script. Most serious failures happen in those grey areas where judgement matters more than rules.

The article highlights that behind these issues sits a bigger truth about training itself. The visible session is only a small part of the work. Effective training depends on deep preparation: understanding the sector, tailoring examples, testing material, and anticipating where people are likely to misunderstand. That hidden effort is what turns information into usable judgement.

Curse of the Cheat Sheet

Everyday social media is filled with neatly packaged “cheat sheets” promising to compress complex subjects into a single, attractive page. Curse of the Cheat Sheet takes aim at this trend, arguing that these visual summaries often create a false sense of understanding rather than genuine knowledge. What looks helpful on the surface can quietly encourage people to believe they have mastered ideas that actually require sustained study, practice, and reflection.

The piece unpacks how these one-page guides spread across platforms like LinkedIn, where they are designed to be shared quickly and consumed even faster. Subjects like leadership, AI, data governance, and organisational change are reduced to bullet points and frameworks that strip away context, difficulty, and the human realities behind them. The danger is not just oversimplification, but the confidence it can produce in people who have only skimmed the surface. Decisions then get made on the back of incomplete understanding, with predictable consequences when reality refuses to behave like a diagram.

At its heart, the argument is about how knowledge is formed. Real expertise comes from time, failure, discussion, and repeated exposure to complexity. Cheat sheets bypass that process, offering reassurance without depth. The article closes on the idea that convenience in learning often carries a hidden cost, especially when it replaces the slower work of actually understanding a subject rather than just recognising it.

Splitting the Bill

Splitting the Bill explores the familiar but often awkward social problem of dividing shared expenses at the end of a group meal. It starts from the simple moment when the bill arrives and quickly turns into a negotiation about fairness, expectation, and etiquette. What looks like a small practical decision often carries emotional weight, especially when people have ordered very different amounts or have different views on what “fair” actually means.

The piece highlights how uneven ordering, alcohol consumption, and dietary differences can distort an equal split. Someone who had a light meal can end up subsidising a heavier spender, while attempts to calculate precise shares can create their own friction at the table. The article also points out that cultural norms vary widely, with some groups preferring strict itemised splitting and others relying on informal rotation or even one person covering the full bill. None of these systems is perfect; each carries trade-offs between convenience, fairness, and social comfort.

At its core, the article is less about arithmetic and more about relationships. The way a group handles money at the table often reflects trust, communication, and how comfortable people are discussing fairness openly. It suggests that many of the tensions arise not from the bill itself, but from the lack of prior agreement about how it should be handled.

If you want a light but revealing look at why something so ordinary can become socially loaded, this one opens up more than it first appears to.

The Truth is Out There

This article reflects on the enduring idea that truth in data, evidence, and analysis is both essential and increasingly contested. It revisits earlier writing from 2018, a period when I was transitioning out of national leadership in NHS data analysis and beginning independent consultancy work. At that time, the focus was on setting out a personal and professional commitment to truth as the foundation of good analytical practice. Seven years on, the piece returns to those ideas in a world that has changed significantly.

The updated reflection considers how the environment around data and truth has shifted. The experience of a global pandemic, fluctuations in public trust, and the rapid emergence of AI systems have all altered how information is produced, interpreted, and believed. In this context, the article explores the growing difficulty of distinguishing reliable evidence from persuasive but potentially misleading outputs. It also touches on the challenge of maintaining integrity in analysis when tools are increasingly capable of generating plausible but not necessarily accurate narratives.

The central thread is that truth has not become less important, but harder to hold onto. Data can still inform better decisions, but only when handled with discipline, scepticism, and clarity about its limits. The article closes with a reflective tone, inviting readers to consider how their own working practices respond to this changing landscape and whether they are still anchored in evidence or drifting toward convenience and assumption.

Human vs AI

A simple anagram puzzle becomes a way to probe how human reasoning differs from large language models. The task is constrained: rearrange the letters in “artificial intelligence” into valid phrases using every letter exactly once. While the human author finds multiple correct solutions through mental tracking, pattern sensing, and iterative adjustment, the AI repeatedly fails to maintain the constraints, often producing fluent but invalid answers.

The contrast centres on how each handles rules and feedback. The human approach relies on internal tracking of letter counts, rejection of near-misses, and a sense of when a combination “feels” correct. The model, by contrast, generates likely word sequences without persistent awareness of prior errors or strict enforcement of constraints, requiring repeated correction and still frequently repeating mistakes.

Even when explicitly guided or asked to verify outputs, the model struggles with consistency, sometimes contradicting itself or accepting and rejecting the same solution across turns. The narrative uses this to illustrate a broader limitation: large language models do not retain structured memory of attempts or perform reliable self-checking, instead optimising for plausible language rather than logical validity.

The piece argues that this makes them unreliable for tightly constrained or high-stakes tasks, where precision, verification, and error tracking matter more than fluency. While they can be useful for drafting, summarising, and exploratory work, they require external systems or human oversight to ensure correctness.

The central point is not that AI lacks usefulness, but that it lacks disciplined reasoning and self-correction. The human advantage lies in maintaining constraints, recognising inconsistency, and adjusting strategy across attempts in a way the model cannot reliably replicate.

Utopia and Economics

AI is often presented as a route to a more humane and productive world of work, where automation removes repetitive tasks and frees people for higher value activity. The piece accepts that the technology can already support much of this vision, but argues that there is little evidence it is currently unfolding in that direction.

The central constraint is not technical capability but the economic system into which AI is being introduced. Existing incentives already reward cost reduction, headcount reduction, and efficiency gains over reinvestment in people. As a result, AI is shaped by those pressures rather than reshaping them, reinforcing patterns of job displacement, wage pressure, and concentration of value among large firms.

While AI is often described as a neutral tool, the argument stresses that outcomes depend on ownership, governance, and institutional context. Without active intervention, it is likely to amplify existing inequalities rather than correct them, particularly as productivity gains flow upward while risk and disruption fall on workers.

The piece also challenges assumptions about labour market adjustment. It questions whether large scale retraining is realistic for many displaced workers and highlights the limits of current policy responses, which are often fragmented or insufficiently funded. Those in routine or administrative roles are identified as particularly exposed, with limited evidence that equivalent new roles will emerge at scale.

Policy responses proposed include better labour market data, updated employment protections for non-traditional work, long-term and funded retraining tied to real demand, and mechanisms to redistribute some of the gains from automation. It also suggests that governments should play a more active role in shaping AI deployment through procurement, regulation, and investment in public-interest systems.

The underlying message in this article is that AI will not naturally produce broadly shared benefits. Without deliberate design and intervention, efficiency gains are likely to deepen existing inequalities. The outcome depends less on the technology itself and more on the structures that govern how it is used.

AI As An Excuse

AI is increasingly being used inside organisations as a justification for restructuring rather than a clear driver of capability change. The piece argues that phrases like “AI-enabled workforce optimisation” often mask conventional cost cutting, with automation used as a narrative to legitimise workforce reductions.

It distinguishes between Artificial Narrow Intelligence, which includes today’s widely used tools such as generative AI, search systems, and rule-based hybrids, and Artificial General Intelligence, which would require genuine reasoning and adaptability. Current systems are described as fluent but non-understanding, producing outputs through pattern prediction and still requiring human oversight due to errors and hallucinations.

Despite this, organisations are making significant decisions based on assumptions that these tools can safely replace human roles. The risk highlighted is that removing people also removes the checking and judgement layer needed for these systems to function safely.

The piece also addresses the assumption that displaced workers can easily retrain, presenting this as unrealistic for many due to financial, personal, and structural constraints. Effective transition, it argues, requires funded support and meaningful pathways rather than informal expectations of self-retraining.

The central point in this article is that AI is being treated as a substitute for people when it is still a tool that depends on them. Without investment in human capability alongside automation, organisations risk gaining short-term efficiency while losing the resilience needed to manage systems when they fail.

In the In-Between

The piece describes a growing “in-between” period in the labour market, where traditional jobs are disappearing faster than new roles are appearing. This gap is not theoretical but lived, affecting people as income becomes unstable and confidence in future work erodes. It challenges the repeated reassurance that new jobs will naturally replace those lost to automation, arguing that this transition is rarely smooth, predictable, or evenly distributed.

Much of the focus is on the human reality of this gap. People face financial strain, uncertainty, and a slow erosion of identity while waiting for new opportunities that may not arrive in time or in suitable form. Retraining is often presented as the solution, but in practice it can be slow, costly, and poorly matched to individual capabilities or circumstances. The idea that workers can simply be reshaped into new roles is questioned, with emphasis on differences in aptitude, experience, and learning style.

The text also highlights systemic issues in how policy, education, and industry respond. Education systems struggle to keep pace with changing job requirements, while policy tends to focus on end states rather than the difficult transition period in between. Support mechanisms exist but are often fragmented, time-limited, or difficult to access when needed most.

A key argument is that the future of work will depend less on technical skills alone and more on human capabilities such as judgement, adaptability, empathy, and the ability to navigate uncertainty. These traits become more important as automation expands, yet they are often underemphasised in training and hiring.

The central message in this article is that the transition to AI-driven work is not just a technological shift but a deeply human one. Without attention to the lived experience of displacement and retraining, there is a risk of creating a workforce that appears prepared on paper but feels unsupported and disconnected in reality. The “in-between” therefore needs to be actively designed for, not ignored, if the transition is to be workable and humane.

Are You a Luddite?

The word “Luddite” is often used as an insult for anyone who questions new technology. Are You a Luddite? challenges that assumption and argues that skepticism is not the same as opposition. Raising concerns about AI, asking difficult questions, or wanting evidence before adoption does not make someone anti technology. In many cases, it is exactly the kind of critical thinking that organisations need if they are going to use technology responsibly. Blind enthusiasm can be just as dangerous as blind resistance.

The article explores the history and meaning of the term “Luddite”, separating the popular caricature from the reality. The original Luddites were not simply afraid of machines. They were responding to the social, economic, and human consequences of technological change. Drawing parallels with modern AI, the article examines how legitimate concerns about governance, accountability, fairness, jobs, and decision making are often dismissed as technophobia. It argues that questioning whether technology is appropriate, safe, or beneficial should be viewed as a sign of responsibility rather than ignorance.

The central message is that progress and scrutiny are not enemies. The people asking awkward questions about AI are often helping organisations avoid costly mistakes, reputational damage, and unintended consequences. Technology should be challenged, tested, and understood before it is trusted. If you have ever been labelled a Luddite simply for asking sensible questions about AI, data, or digital transformation, this article offers a thoughtful perspective on why healthy skepticism remains one of the most valuable tools we have. The full article is well worth a read.

The Real Price of AI

Artificial intelligence is often presented as a shortcut to greater productivity, lower costs, and effortless efficiency. The Real Price of AI challenges that narrative by arguing that the true cost of AI goes far beyond software licences and cloud subscriptions. Every AI system carries hidden costs in governance, oversight, training, compliance, risk management, and human accountability. Organisations that focus only on the purchase price may discover that the biggest bills arrive later.

The article explores the many forms these costs can take. AI systems need people to monitor them, validate their outputs, investigate mistakes, and manage the consequences when things go wrong. There are costs associated with data quality, security, ethics reviews, legal compliance, staff training, and maintaining public trust. Even seemingly successful AI projects can generate unexpected expenses if biased decisions, inaccurate outputs, or poor governance create operational, financial, or reputational damage. The promise of automation can sometimes disguise the reality that organisations are taking on new responsibilities rather than eliminating old ones.

The central message is that AI should be treated as an ongoing commitment rather than a one off purchase. The question is not simply whether AI saves money, but whether organisations have properly accounted for the full cost of using it responsibly. If you are involved in AI procurement, governance, strategy, or implementation, this article offers a valuable reminder that the cheapest AI solution is not always the least expensive in the long run. The full article is well worth reading for anyone trying to understand the economics and responsibilities that sit behind the AI hype.

The AI-Way Code

The AI Way Code argues that successful AI adoption has far less to do with technology and far more to do with governance, oversight, and understanding what different AI systems actually do. Too many organisations treat AI as a single tool when, in reality, it is a collection of very different technologies with different strengths, weaknesses, and risks. Assuming they all behave the same way is a recipe for confusion, poor decisions, and unexpected consequences.

Using the metaphor of a chaotic family gathering, the article introduces different types of AI as colourful relatives. Generative AI becomes the enthusiastic cousin who is creative but prone to making things up. Predictive AI is represented by parents constantly trying to guess what happens next based on past experience. Large language models are the confident uncle who always has an answer, whether it is correct or not. Agentic AI is the overzealous older sibling who takes initiative and acts independently, sometimes creating more problems than it solves. Symbolic AI appears as the rule bound grandad who is reliable in familiar situations but struggles when life becomes messy and unpredictable. Through these characters, the article highlights the planning, controls, and clean up required when deploying each type of AI.

The central message is that AI is not something you simply switch on and trust. Different AI systems require different safeguards, different levels of supervision, and clear accountability. Organisations that fail to understand these differences risk poor outcomes, legal issues, loss of trust, and operational chaos. If you are trying to make sense of the growing AI landscape, this article provides an accessible and memorable guide to the personalities, risks, and realities behind modern AI. The full article is well worth a read

What Are You Talking About?

Most people assume that large language models (LLMs) like ChatGPT understand what they’re saying. They don’t. What looks like insight is often just fluency, confident phrasing stitched together from training data. These systems don’t track people, events, or facts across time. They don’t build a mental map of what’s happening in a conversation. Unlike humans (or even ants), LLMs can’t form or update internal representations of the world. That’s why they contradict themselves, forget details, and confidently offer falsehoods with no awareness that anything has gone wrong.

These limitations show up in all kinds of bizarre ways. Legal documents filled with fictional citations. AI shopkeepers who give away stock and claim to wear business suits. Systems that describe a video without noticing that a theft has taken place. The underlying issue isn’t a lack of data. It’s that these models aren’t designed to reason, track change, or understand. They predict text, but they don’t know what they’re talking about. Plug-ins and workarounds such as external databases or tool-based agents help a little, but the models still can’t maintain stable knowledge or beliefs.

Until systems are built with proper world models that are structured, updatable, and internally consistent, they can’t be trusted for tasks that demand consistency, memory, or reasoning. No amount of fine-tuning can change that. If you rely on LLMs without understanding their limits, you’ll be caught off guard when they break. If you’re using, designing, or evaluating AI tools, it’s worth understanding why this limitation matters. The full article explores it in detail. Give it a read if you want to understand where current AI is falling short and why that matters.

What the Dickens?

This article takes aim at the obsession with identifying AI-generated writing based on shallow markers like punctuation and phrasing. Starting with Dickens' famous opening to A Tale of Two Cities, the author points out the absurdity of using elements like an em dash to discredit a piece of writing. Instead of reading for meaning, too many people scan for signs of machine involvement and dismiss the content without engaging with the argument at all.

The writer expresses growing frustration with this surface-level culture, especially on platforms like LinkedIn. Thoughtful posts are ignored while tired criticisms of AI writing get amplified. People who put genuine time and care into their work often have to change how they write just to avoid being labelled as artificial. It’s exhausting, and it’s silencing the voices that actually have something to say.

The real issue isn’t how something was written, but whether the ideas are worth engaging with. This piece calls for a shift in attention; towards substance, nuance, and reflection. If we keep rewarding noise and ignoring care, we’ll be left with nothing but shallow content. If this resonates you will ant to read the full article. It is a far, far better thing you'll read, than you have read before.

AI Think, Therefore AI Am

Artificial intelligence is often described as “thinking,” but that label may be more confusing than helpful. The real issue is not whether AI is getting smarter. It is that we have never clearly defined what “thinking” actually means. The words we reach for come from a time before AI existed. They carry assumptions that do not map well to how these systems work.

Take Descartes and his famous line, Cogito ergo sum, which means "I think, therefore I am." He was not just making a philosophical point. He was laying a foundation for certainty and using it to support his argument for the existence of God. Thought, for Descartes, was proof of being. It was tied to awareness, doubt, and a divine framework. That kind of thinking has little to do with what machines are doing when they predict the next word in a sentence.

And yet, the outputs from these systems can seem intelligent. They answer questions, solve problems, and mimic human reasoning. But calling that “thinking” risks misleading people. The danger is not the AI itself. It is that we are using the wrong language to describe what it does. If you are curious about how our old definitions are falling short, and why that matters, read the full article.

Method to the Madness

We often trust our perceptions, models and past experiences to make decisions, in life, in business, and in science. But this article challenges that instinct, showing how our thinking can be shaped by hidden assumptions and biases. Drawing on the principles of the scientific method, it explores how habits like scepticism, empiricism, reductionism and determinism can help us test what we believe rather than just accept it.

The parallels with business are clear: questioning accepted strategies, being honest about weak evidence, and resisting the urge to force everything into neat narratives are just as relevant to running a company as they are to running an experiment. The article draws comparisons between scientific theories such as the Big Bang or evolution and business frameworks, not to discredit them, but to show how both need constant testing and revision.

It is a call for intellectual humility. A reminder that good decisions often come from not being too sure. Whether you're leading a team or launching a product, adopting the mindset of a scientist, curious, methodical, ready to revise, can help you avoid costly errors. If that sounds useful, the full article is well worth a read.

When Harry Met AI

AI can seem slick and capable, but trusting it blindly is a bit like handing the keys to a business over to a teenager called Harry. Confident, unpredictable, and not always honest about what it's done or why, AI systems can behave just like adolescents, especially when left unsupervised. The blog explores this comparison in a light, relatable way that’s grounded in real risks and responsibilities.

By walking through a day in the life of Harry, a poorly teen making bad choices thanks to a fever, the article draws a parallel with AI models that go off track. The parent, like a responsible business, notices the signs, steps in, seeks expert help, and keeps Harry away from others to prevent more harm. The piece reminds us that you can’t expect good decisions from either a sick teenager or a malfunctioning AI.

It also reinforces the need for oversight, structure, clear boundaries, and the ability to pause or escalate when things go wrong. Someone has to be accountable, someone has to be watching. If any of this sounds familiar or raises a few questions about your own use of AI, it’s worth taking five minutes to read the full article.

AI Know What You Did Last Night

AI systems don’t all work the same way. Some generate content, others make predictions, some act independently, and a few stick rigidly to rules. Each model has its own logic, strengths, and blind spots. When combined, they often work at cross purposes, misunderstanding each other’s output or turning small errors into bigger ones. That’s the risk when different types of AI are thrown together without planning. Things start to behave unpredictably.

Adaptive AI is slightly different. It doesn’t invent or forecast. It learns from immediate context, just what happened a moment ago, and reshapes its behaviour in real time. That can be helpful, even charming, when it gets something right. But if it misreads the room or takes a single action as a preference, it adapts in ways that feel personal but are actually based on guesswork. It’s not malicious, just confident with too little to go on.

This article introduces Jamie, a human version of Adaptive AI, who tries to help based only on what he saw at a party the night before. Through that lens, we explore how short-term learning can quietly cause longer-term confusion. If you’ve ever wondered why your tech suddenly acts like it knows you too well, or keeps adjusting just when you’d figured it out, Jamie might be the reason.

Read the full article to meet Jamie and the rest of the AI family, and see what happens when one of them tries to run breakfast.

When Seeing is No Longer Believing

We’ve long leaned on the idea that what we see and hear is real. A photograph, a video clip, a voice on tape—these were once enough to settle arguments and confirm events. But with AI now able to create realistic faces, voices, and footage from scratch, the old rules no longer apply. What looks and sounds convincing might be entirely synthetic, and most of us aren’t equipped to tell the difference.

It’s not just politics or crime scenes. AI is writing music, generating news stories, producing deepfake interviews, even mimicking dead celebrities. Some of it feels creative or clever, but it also clouds authorship, truth, and accountability. As more of this content circulates, we need better tools for spotting fakes, new ways of confirming what’s genuine, and a stronger instinct to pause before we believe what we’re shown.

This article looks at real-world examples, from political deepfakes to AI-generated Spotify tracks, and asks what we can do next. If you’ve ever wondered whether reality is starting to slip, or if the phrase “perception is reality” still holds up, this piece is worth your time. Have a read and see where you land.

What's in a Name?

Shakespeare said a rose would smell just as sweet by any other name. But in artificial intelligence, names matter more than ever. The term "AI" is now used so broadly that it often misrepresents what a system can actually do. From simple automation to complex machine learning, calling something "AI-powered" can create false expectations.

Many products marketed as AI are really just basic software following preset rules. When companies use terms like "smart," "autonomous," or "agentic," people expect systems that understand, adapt, and make independent decisions. But most AI tools simply analyze data patterns without real awareness or judgment.

This mismatch leads to trust issues, legal risks, and disappointment. Examples include tools labeled as "robot judges" that only assist human decision-making, or hiring algorithms that replicate existing biases but are called "intelligent." The language used shapes how people interact with these tools and who is held responsible when things go wrong.

If you work with AI or use it, it is important to use clear and honest language. Describe what your system actually does instead of what you hope it does. And as a user, be cautious of overhyped claims.

Names are not neutral. In AI, they influence perception, trust, and accountability.

To learn more about why naming matters so much in AI, read the full article.

The Power of the Repost

Lately, I’ve noticed something odd on LinkedIn. Thoughtful, well-crafted posts about AI, data, leadership, and culture often get little engagement, even when people are reading them. Meanwhile, quick personal updates like job announcements or award photos get lots of likes, comments, and shares.

This isn’t just about what’s popular. LinkedIn’s algorithm favours early reactions, so many valuable posts quietly disappear. That’s why reposting matters more than most realize. Sharing someone else’s content keeps important ideas alive and helps them reach new audiences, especially for those of us without huge followings.

If you find something meaningful or thought-provoking, don’t just read and move on. Give it a second life by reposting. It’s a small action that can make a big difference in how ideas spread and how conversations develop.

Want to understand why reposting matters so much, and how it shapes the quality of content we see? Read the full article for deeper insights.

It's my party and I'll AI if I want to

AI is often talked about as a single clever tool, but it’s more like a family—full of different personalities that don’t always get along or behave as expected. Imagine hosting a chaotic family gathering, where each relative represents a type of AI with its own quirks and challenges.

Cousin Steve (Generative AI): Creative and enthusiastic but prone to making things up. Great for ideas but needs supervision and fact-checking.
Mum and Dad (Predictive AI): Always guessing what comes next based on past data. Useful, but their predictions aren’t guarantees and can reflect bias.
Uncle Rob (Large Language Models): Fluent and confident talker who sounds convincing but doesn’t always check facts. Best used with boundaries and oversight.
Older Sibling (Agentic AI): Takes initiative and acts independently. Powerful but can cause chaos if unchecked, so limits and checkpoints are essential.
Grandad (Symbolic AI): Rule-based and predictable, but inflexible when things don’t fit neat categories. Best in well-understood, stable environments.

Most AI setups today are hybrids, mixing these “family members,” which creates complexity and potential conflicts that require careful management.

For a full breakdown of each AI type, what to expect, and how to prepare your “party” for success, check out the complete article. It’s a must-read for anyone working with AI today.

Rise of the Doomsayer

AI sparks all kinds of reactions: excitement, fear, or outright denial. Among these voices are the doomsayers, those warning that AI could bring catastrophic collapse. Like ancient prophets sounding the alarm, they paint vivid pictures of AI taking over work, society, or even humanity itself.

But their predictions deserve a closer look. While some claims are extreme, like AI wiping out all jobs or suddenly becoming sentient, they do not align with current realities. Instead, AI is reshaping how we work, automating some tasks but also creating new roles. And today’s AI systems, no matter how clever, are not self-aware; they are just pattern-matching machines.

Where the doomsayers gain traction is in the real, present-day risks: biased algorithms locking people out of opportunities, flawed medical tools missing diagnoses, deepfakes undermining trust. These are not sci-fi fears; they are happening now, quietly and sometimes dangerously.

The challenge is clear. AI is not just another tech upgrade. It changes decision-making and impacts real lives, so organizations must slow down, ask tough questions, and build trust from the start. Oversight, transparency, and accountability cannot be afterthoughts.

Doomsayers may exaggerate, but they also push us to be vigilant, thoughtful, and responsible. Ignoring them risks repeating mistakes. Listening and acting wisely might just be how we turn their warnings into progress.

For a deeper dive into the fears, facts, and future of AI, the full article is well worth your time.

When Belief Get's in the Way

This article looks at what happens when belief is mistaken for truth, and how relying too much on perception can lead to serious mistakes. From healthcare to the workplace, the piece shows how decisions often rest on what feels true rather than what is actually supported by evidence. Misdiagnoses happen, people are promoted for their visibility rather than their value, and entire systems are shaped around assumptions that go unchecked for too long.

The article uses powerful examples, including the long-standing misconceptions about Chronic Fatigue Syndrome and the careless overuse of antibiotics, to show how these patterns play out in real life. It also explores how performance is measured in many organisations, often in ways that reinforce outdated ideas and reward style over substance. These stories expose the quiet damage caused when belief is given more weight than reality.

If you have ever wondered why poor decisions persist even when the evidence says otherwise, this article will give you a clear and thoughtful perspective. It challenges comfortable assumptions and invites the reader to think more critically about what they accept as true. The full piece is well worth reading for anyone who wants to better understand the hidden influence of belief in everyday decisions.

The Dreamer and the Oracle

Artificial intelligence is being used to paint portraits, write stories, generate music, and simulate entire conversations. That’s the side most people see. But there's another kind—quieter, less visible—that claims to know what you're likely to do next. It predicts whether you’ll succeed, whether you’re a risk, whether you deserve a loan, a job, or a second chance.

This article explores the difference between creative, generative AI and its more quietly influential cousin, predictive AI. Through examples from Blade Runner, Gattaca, Minority Report, The Matrix, and more, it shows how fiction has been warning us for years about what happens when machines aren’t just tools, but judges of human potential.

The systems we trust to be neutral often aren’t. They don’t just make predictions. They shape the outcomes. They create feedback loops. They define futures people haven’t had the chance to live.

If you’re worried about rogue machines, you’re looking in the wrong direction. The real threat may be the systems that seem calm, logical, and fair—yet gradually decide who you are allowed to be.

Read the full article to see how the most powerful AI isn’t the one that acts, but the one that assumes.

Bias Remorse

Bias can easily slip into AI systems through the data used for training or during development choices. Several common types of bias are explained, such as selection bias caused by limited or unrepresentative data, and labelling bias that comes from inconsistent human categorisation. Recognising these helps understand why AI sometimes produces unfair or flawed results.

Real-world examples show how bias impacts important decisions. A hiring algorithm unfairly downgraded applications referencing women’s colleges, while a legal risk assessment tool gave inconsistent scores due to biased historical data. Cases in finance and healthcare reveal how bias can affect credit limits and patient prioritisation, illustrating the real consequences of unchecked bias in AI.

Practical steps are outlined for tackling bias, including scrutinising data carefully, testing models with unusual scenarios, and involving diverse perspectives in development. Keeping thorough records and proactively searching for issues can prevent problems before systems go live. Reading the full article will provide a clearer picture of bias in AI and useful approaches to manage it effectively.

A Good Data Die

Data collection and analysis often take all the attention, but the moment when data becomes outdated or irrelevant is just as important. Holding on to old information can create risks that go beyond just wasted storage. Forgotten files and duplicated records can become a liability, especially if they contain sensitive personal or business details. Over time, the value of data fades while the risks remain or even grow.

The article explores what happens when obsolete data isn’t properly deleted. It uses the example of a spy who keeps every piece of intelligence after a mission, only to have that data become a vulnerability. In reality, old data left unsecured can be exploited, misused, or accidentally exposed. When AI systems consume outdated or biased data, they make confidently wrong decisions that cause real damage. Regulatory requirements like GDPR make clear that data should not be kept longer than necessary, and failure to comply carries serious consequences.

Practical advice is offered on how to handle data lifecycle responsibly: setting clear expiry rules, automating deletion, respecting consent, conducting audits, and building data literacy across teams. Treating data deletion with the same care as data collection helps organisations reduce risk and build trust. There is real value in knowing when to let data go. To understand the risks and learn how to manage data more effectively, the full article is worth a read.

The Only Way is Ethics

Data often looks simple at first, but it can be full of hidden biases and flaws. Legal compliance sets the minimum standards, but it does not guarantee fairness or transparency. Without asking the tough questions about where data comes from and how it is used, decisions based on that data risk causing real harm.

AI systems now play a major role in decisions about jobs, healthcare, and finances. Yet many operate as “black boxes,” making it hard to understand or challenge their outcomes. If the data behind them is biased or incomplete, those systems only amplify existing inequalities.

A healthy data culture means being aware of context, questioning assumptions, and treating ethics as a central part of every data process. Transparency and responsibility are key to building trust and avoiding unintended consequences.

If you want to understand why ethics must come first in data and AI, this article is well worth a read.

The Dangers of Personal Experience

We all rely on personal experience and gut feeling to make decisions, but when those choices affect many people, mistaking perception for reality becomes risky. Without solid data and the skills to interpret it, decisions are prone to bias and assumptions, which can lead to misguided policies and misallocated resources.

Data literacy is crucial across all levels of an organisation, not just for analysts. When people don’t understand the data’s context or limitations, they fill the gaps with their own views, which may be narrow or biased. Building a workforce comfortable asking tough questions about data helps prevent decisions based on incomplete or misleading information.

The gap between perception and reality can be wide, especially in complex issues like social mobility or homelessness. Data doesn’t replace experience but challenges it, offering a broader, more accurate picture. When more people understand data and can question it effectively, decisions become fairer and more effective.

If you want to explore why data literacy matters so much and how it can help organisations avoid costly mistakes, this article is worth a read.

When Perception and Reality Clash in Data Analysis

We often hear that "perception is reality," but in data analysis, this idea can be misleading. Our personal views are shaped by biases and experiences, while reality is independent and fixed. When interpreting data, it’s easy to mistake statistical results for absolute truths, yet measures like p values, confidence intervals, and study power all come with uncertainties that challenge our initial impressions.

Statistics like p-values can indicate if a result is unlikely to be random, but they don’t tell us how meaningful the effect actually is. Tools such as sensitivity and specificity help assess tests’ reliability but still allow for errors: both false positives and false negatives. These limitations show why numbers should be understood as clues rather than certainties, and why careful interpretation beyond surface-level results is essential.

The gap between what data seems to say and what it truly means matters a great deal, especially in fields like health research. Overconfidence in statistical significance without considering context can lead to overstated conclusions or missed insights. Accepting uncertainty and approaching analysis with curiosity and scepticism helps reveal a clearer, more nuanced picture.

If you want to dive deeper into how perception and reality clash in statistics, and why understanding this gap matters, this article offers useful examples and practical reflections.

Going Beyond the Tools to Technique

The data environment has rapidly evolved with an explosion of data and analytical tools. Many organisations now use programming languages like Python and visualisation platforms such as Power BI to explore data. However, there is a growing problem where businesses value tool skills more than a deep understanding of statistical and analytical techniques. This focus on tools over technique can lead to shallow analysis and flawed decisions.

In other fields, expertise is judged by skill rather than the tools used, yet data analysis often reverses this. Just as a surgeon is chosen for their knowledge rather than their scalpel, and a musician for their skill rather than their guitar, businesses must prioritise strong analytical techniques over simply knowing how to operate software. Without this foundation, organisations risk misinterpreting data and missing chances to innovate or compete effectively.

If you want to learn why focusing on analytical skill instead of just tools can protect your business and improve decision-making, take a moment to read the full article. It explains how investing in data literacy and expertise can make all the difference in today’s data-driven world.

The Power of Data Storytelling

Data storytelling is an often overlooked skill that can transform how businesses present and use their data. Rather than simply showing numbers and charts, data storytelling combines analysis with narrative and visuals to create engaging, clear, and meaningful stories. This approach helps decision-makers understand insights better and motivates action, turning complex data into something relatable and memorable.

Many organisations invest heavily in training analysts to extract data but fail to support them in communicating those findings effectively. Without a compelling story to frame the data, important insights risk being misunderstood or ignored. By adopting data storytelling practices, companies can bridge the gap between raw data and strategic decisions, encouraging collaboration and fostering a culture where data truly informs business direction.

If you want to see how data storytelling can boost your organisation’s impact, improve communication, and drive positive outcomes, take a moment to read the full article. It explains why storytelling is not just a presentation tool but a vital skill for modern businesses aiming to thrive in a competitive world.

The Imperative of Data Literacy in a Digital Age

Data literacy is essential for organisations to fully benefit from the data they collect. While many invest heavily in technology and technical skills, the real value comes from employees who understand the full story behind the data: where it comes from, its limitations, and what it actually means. Without this deeper understanding, organisations risk misinterpreting their data, making poor decisions, and missing important insights.

A common problem in many companies is the siloed approach to data management, where different teams only see part of the process and fail to grasp the bigger picture. This fragmentation can lead to confusion, conflicting interpretations, and even misuse of data. Developing data literacy across the entire workforce helps to break down these barriers, enabling clearer communication, stronger data governance, and better alignment between departments.

If you want to learn why data literacy is more than just technical know-how and how it underpins ethical, accurate, and effective use of data in organisations, the full article is worth a read. It explains how a data-literate workforce supports better decision-making and protects a company’s reputation in today’s digital world.

Someone Else's Shoes

This piece reflects on the idea of seeing situations from other people’s perspectives, starting with a childhood story about hand-me-down clothes and shoes. It explores how early feelings of unfairness shaped the author’s understanding of empathy and the importance of considering others’ circumstances. The writer connects this personal insight to broader issues like workplace stress, management challenges, and discrimination.

The article stresses that empathy is essential in leadership and daily interactions. Treating everyone equally isn’t enough because people have different needs, experiences, and backgrounds. Understanding these differences helps prevent inappropriate discrimination and improves how we relate to others. This approach also applies to data analysis, where knowing the context behind the numbers and the people involved can make the results more useful and accurate.

Finally, the blog encourages readers to pause before judging others or reacting negatively. It suggests practicing empathy regularly, not just in difficult moments, and appreciating that people’s actions often reflect unseen pressures. The message is that we all benefit from stepping into someone else’s shoes and seeing the world from their view.

If this resonates with you, I recommend reading the full article to explore these ideas in more depth.

Call It What It Is

The article steps away from usual topics like data and statistics to reflect on the culture around us, especially in light of the Covid pandemic. It questions whether causing harm to others can ever be justified for personal or organizational gain. Drawing parallels with phrases like “collateral damage” and “for the good of medical science,” it argues that many still accept hurting people if it benefits a larger goal, such as profits or progress, even when that harm is avoidable or unnecessary.

The post highlights workplace bullying as a serious issue often dismissed or tolerated under the guise of company culture or business needs. It explains how toxic environments wear down individuals, leading to emotional and mental harm, and sometimes even bonding victims to their abusers. The author stresses that bullying should be treated like abuse, not a normal part of work life, and that employers and HR must do more to protect employees rather than just defending the company.

Finally, the piece calls for us to stop sugarcoating workplace mistreatment with soft language and to recognize it for what it truly is: abuse and violence. It argues that if domestic abuse is criminal, workplace abuse should be too. The author urges everyone to remember basic human decency, treat others with respect and kindness, and not settle for anything less.

If this strikes a chord, reading the full post will give you a clearer view on why calling out workplace abuse honestly matters.

Epistles, Enquiry and Ethics

Data availability and use in health and social care have evolved significantly since the early 1990s, moving from limited and anecdotal information to vast, complex datasets drawn from many sources. This expansion goes beyond healthcare to social media, finance, and more. Alongside this growth comes a need to shift the mindset about how data is managed and analysed.

A key issue is the importance of having skilled people handle data, much like companies employ experts for financial matters. Despite this, data analysis is often treated as something anyone can do, which carries risks. Ethical concerns arise because following the law is not enough; questions about consent, purpose, skills, risks, and impact must be carefully considered before analysing data.

The old justification of pursuing data use or research “for the good of medical science” is compared to outdated ideas about acceptable collateral damage. Ethics should be central in data use, with proper training and restricted access to sensitive information. Ethical considerations need to become as routine as technical skills for those working with data.

If you want to explore these ideas further, reading the full article is worth your time.

Ask a Silly Question

The phrase "Ask a silly question, get a silly answer" often discourages people from seeking clarity, especially in adult life. This mindset can make individuals hesitant to question experts or complex information, leading to misunderstandings or blind acceptance of what they are told. The author shares a personal experience of speaking up in a meeting only to discover that nobody really understood the jargon-filled discussion, showing how important it is to ask questions—even if they seem silly.

In data analysis and reporting, accepting figures at face value without asking basic questions can be dangerous. Using unemployment rates as an example, the text explains how a falling unemployment figure isn’t always a straightforward sign of economic improvement. Factors like people leaving the workforce, changes in benefits, or even mortality among the unemployed can influence the numbers, meaning a deeper look is necessary before drawing conclusions.

The article urges readers to challenge assumptions and ask those “silly” questions without fear, especially around data. Doing so reveals complexities that numbers alone can’t explain and prevents misinterpretation that can have real consequences. For anyone interested in understanding how data shapes our views, it’s worth reading the full article to see why curiosity matters more than ever.

Looking Beyond the Label

Shopping can be frustrating when clothes don’t fit the way they should, especially for men who don’t match the ‘average’ sizes stores cater to. Labels like ‘big man’ or ‘one size fits all’ feel limiting and ignore real differences in body shapes. This reflects a bigger problem where marketing and media create perceptions that don’t reflect reality, leaving many people feeling excluded and overlooked.

The problem of labeling shows up just as much at work. Job titles such as ‘data scientist’ get thrown around without a clear definition, causing confusion and sometimes unnecessary competition between roles. Instead of relying on vague labels, it’s better to focus on what people actually know and can do. The skills and experiences everyone brings are unique, and one label can never fully describe a person’s value.

Moving away from rigid labels helps include more people and appreciate what they bring to the table. Whether in shopping or in careers, labels can box people in and deny their individuality. Understanding this encourages us to see people as more than just a name or category. If you want to explore this idea in more depth and consider how labeling affects us all, I encourage you to read the full article.

Statistics, Certainty and Crystal Balls

The Proof of the Pudding

The idea of proving whether one pudding tastes better than another runs into some tricky problems. It’s not possible to prove that two things taste exactly the same because someone, somewhere, might disagree. You would need to test every single person to be sure, which is impossible. On the other hand, proving that one pudding is better can also be uncertain because samples might not represent the whole population or could be affected by chance.

This touches on how science works. It never offers absolute proof, only levels of confidence based on evidence. Even a large group of testers cannot guarantee a definite answer, only a likelihood. That uncertainty is part of why scientific thinking encourages skepticism and an openness to new evidence, rather than demanding final proof one way or another.

When it comes down to it, the question of which pudding is better is less about proof and more about weighing the evidence thoughtfully. It’s a reminder that many things we want to be certain about can only be approached with caution and curiosity. If you want to dive deeper into the challenges of proving “proof” itself, this article offers some interesting insights.

Everyone's an Expert

Ever had that weird neck twinge, rash, or headache that just won’t go away? You pop some ibuprofen, maybe some leftover meds, but nothing helps. So, naturally, you turn to Google. And suddenly, your harmless symptom looks like a rare, deadly disease. Panic sets in—you imagine the worst, maybe even say goodbye to loved ones, and rush to the doctor expecting the worst.

But then the doctor calmly tells you it’s just a mild case of sunburn. No crisis, no drama—just simple advice to drink water and take paracetamol. Relief floods in, along with the realization that you can’t replace years of medical training with a Google search.

Sounds familiar? Now imagine doing the same with data. Everyone thinks they can analyze data because they did math at school and have Google at their fingertips. But data isn’t just numbers; interpreting it correctly requires expertise, training, and experience—just like diagnosing health issues.

If you jump to conclusions without the right skills, it can lead to wasted time or worse—bad decisions affecting finances, patient care, or staffing. That’s why we need trained analysts, statisticians, and data scientists, not just casual dabblers.

If you’re involved with data in any way, make sure your training is solid. Check out my courses to stay sharp and be the trusted expert who gives the right “diagnosis” when it really matters.

Want the full story and more insights? Dive into the full article!

Magna est Veritas

Ever wondered where your family name comes from or what it might mean beyond just a label? I’ve been fascinated by the story behind my surname, Varlow, and the challenges it brings, like getting it mixed up over the phone or even by Google. That curiosity led me to discover my family coat of arms, which includes the Latin motto Magna est Veritas, meaning Great is Truth. This phrase really resonates with me, especially in my work delivering training around data and research.

In the full article, I explain why truth isn’t just a nice-to-have but absolutely essential when working with data, whether you’re analyzing, interpreting, or making decisions. With so much data at our fingertips and powerful tools available, it’s tempting to focus on flashy visuals or quick results, but without honesty and rigor, that data can mislead or even cause harm. I share why Magna est Veritas is more than just a motto, it’s a reminder of our responsibility to defend the truth in everything we do with data.

If you’re someone who works with data, analysis, or research, or if you simply want to better understand how to handle information responsibly, I encourage you to read the full article. It’s packed with insights on why integrity matters, how to ask the right questions, and how to use data in a way that really empowers people. This is more than theory, it’s about real-world challenges and why truth must always come first. Give it a read and see why protecting the truth is at the heart of my training and my approach to data.

Page updated

Google Sites

Report abuse