Talking Garbage

Every now and then, an article comes along that makes you stop and reflect, not because it's saying something new, but because it puts into words something that many people working in healthcare data have known for many years (and would have told you if they'd been asked).

A Health Service Journal article, published today (16 July 2026), has one of those headlines that immediately caught my attention: "Without Better Data the NHS Will Squander the Opportunities Offered by AI." And, to be honest, it's difficult to argue with that.

I've written a lot about artificial intelligence on LinkedIn and other platforms (even though many of my articles haven't had as wide a reach as I would like). The pace of change in AI during this period has been remarkable. Tools that once seemed like futuristic ideas are now available to almost everyone. AI can write, analyse, create images and videos, help with research and perform tasks that would have sounded impossible only a few years ago.

It's definitely exciting, but it's also slightly unsettling. Not because AI is something we should automatically fear, but because the speed of development has sometimes moved faster than our ability to think through the wider consequences. We’re becoming very good at creating systems that can process information, recognise patterns and produce answers. The harder question is whether we're equally good at making sure the information behind those answers can be trusted. Because while we spend a lot of time asking what AI can do, we perhaps spend less time asking a more basic question.

What are we actually giving it to work with?

We’re becoming very good at creating systems that can process information, recognise patterns and produce answers. The harder question is whether we're equally good at making sure the information behind those answers can be trusted

This isn't a new concern. I've spent most of my career working with data, statistics and information standards, and one thing has remained constant. The cleverest analysis in the world is only as good as the information underneath it.

Back in the 2000s, I spent some years as the Public Health and Statistics Standards Lead at the Information Standards Board for Health and Social Care, and for one year I also stood in as Head of Standards Services. Much of that work focused on questions that are becoming increasingly important in the age of AI. Do we all mean the same thing when we record information? Can different systems exchange data and actually understand each other? Can standards be implemented safely in real clinical environments?

Those questions might sound less exciting than talking about artificial intelligence, but they’re the difference between technology that genuinely supports healthcare and technology that simply creates more uncertainty. They are also questions that existed long before anyone was talking about large language models or generative AI.

There’s an old phrase from the early days of computing that still gets used today: "garbage in, garbage out". It’s not exactly a phrase you’d expect to see on a strategy slide, but it remains one of the simplest explanations of how information systems work. If the information going into a system is incomplete, inconsistent or poorly understood, the output will reflect those weaknesses. The challenge with AI is that the output can look so convincing. A traditional data error might stand out because a number looks wrong or a calculation doesn’t add up. An AI generated response can be beautifully written, confidently delivered and completely plausible, even when something underneath it is wrong.

The challenge with AI is that the output can look so convincing. A traditional data error might stand out because a number looks wrong or a calculation doesn’t add up. An AI generated response can be beautifully written, confidently delivered and completely plausible, even when something underneath it is wrong.

AI doesn’t know that a piece of information has been recorded incorrectly. It doesn’t know that one department uses a definition slightly differently from another. It doesn’t recognise that a change in recording practice has altered the meaning of a trend unless those changes are identified and accounted for. It simply works with the information it receives.

A human expert might look at a dataset and think, "That doesn’t look right." They might remember that a coding system changed, that a new service opened or that a particular group of patients has historically been underrepresented in the data. AI can identify patterns at a scale humans would struggle to match, but it doesn’t bring the same experience or context. That’s why the information environment around AI matters so much.

I sometimes think of this like a Satnav. Most of us use satellite navigation without thinking about what sits behind it. We simply enter a destination and follow the instructions. But imagine using a Satnav with an outdated map. This is unfortunately something I am experiencing while waiting for mine to be updated. Roads are missing. Junctions have changed. One way systems have changed. Bus lanes have appeared. A new bypass has opened but the system hasn’t caught up. The satnav still confidently tells me where to go, but it’s now wrong some of the time. That can be a frustration and sometimes, worse, lead to a fine. The problem isn’t the technology but the information it relies on.

Healthcare data faces a similar challenge. The NHS is one of the most complex organisations in the world. Its information systems have developed over decades, often at different times and for different purposes. Some systems work extremely well within their own environment but struggle to exchange information with others. That isn’t necessarily because people have made poor decisions. Healthcare has had to evolve while continuing to provide care every day. Replacing information systems across the NHS isn’t like updating a phone or installing a new piece of software. The scale and complexity are enormous. And that complexity creates consequences.

Healthcare has had to evolve while continuing to provide care every day. Replacing information systems across the NHS isn’t like updating a phone or installing a new piece of software. The scale and complexity are enormous. And that complexity creates consequences.

Two organisations can record similar information in slightly different ways. A term that seems obvious to one team may mean something subtly different somewhere else. A field that is completed routinely in one setting may be missing in another. Individually, these might seem like small issues. But when information is combined across thousands of services and used to support decisions, those differences can become much more significant.

This is why information standards have always mattered. Standards are sometimes seen as bureaucracy. Another set of rules. Something that systems developers have to adhere to. Another thing for busy professionals to understand and comply with. But good standards aren’t really about paperwork. They’re about creating shared understanding.

If a clinician records information, a researcher analyses it and an AI system uses it to support decision making, everyone needs confidence that they’re talking about the same thing. Technical interoperability is only part of that challenge. Two systems can successfully exchange information while still misunderstanding its meaning. Semantic interoperability has been lost. The data has moved, but the understanding hasn’t.

That was one of the important issues we considered at the Information Standards Board. It wasn’t simply about whether information could travel between systems. It was about whether it remained meaningful, whether it could be implemented safely and whether it supported better care. Those principles are just as relevant today.

There’s also the question of trust. Healthcare depends on trust at every level. Patients trust professionals with their personal information. Clinicians trust the systems that support their decisions. The public trusts that NHS data will be handled responsibly.

Artificial intelligence doesn’t remove those responsibilities. If anything, it makes them more visible. People will quite reasonably ask who has access to their data, how it’s being used and whether the benefits genuinely improve patient care. Those questions shouldn’t be seen as barriers to progress. They’re part of making sure progress happens in the right way.

People will quite reasonably ask who has access to their data, how it’s being used and whether the benefits genuinely improve patient care. Those questions shouldn’t be seen as barriers to progress. They’re part of making sure progress happens in the right way.

The interesting thing is that improving data quality helps healthcare whether AI succeeds or not. Better information helps clinicians make decisions, supports research, improves planning and allows organisations to understand what’s happening within their services. AI simply makes the consequences of poor information harder to ignore.

The NHS doesn’t have to choose between artificial intelligence and better data. It needs both. The most advanced AI system in the world will struggle if the foundations underneath it are weak. A strong information environment, however, creates opportunities that go far beyond AI.

After decades working with statistics, information standards and healthcare data, I’ve learned that the most important questions are often the least exciting ones. How was this information collected? What has happened in its journey to me? What does it actually mean? Can we trust it? Can somebody else understand it in the same way?

These sorts of questions are unlikely to generate many headlines. They won’t attract the same attention as the latest AI breakthrough. But they may determine whether those breakthroughs genuinely improve healthcare or simply create more complicated problems.

The future of AI in healthcare won’t only be decided by how clever our machines become. It’ll also depend on whether we’ve been wise enough to make the information feeding them trustworthy. Otherwise we’re just talking garbage.