Babies with Power Tools

Scrolling through LinkedIn at the moment, it is almost impossible to avoid conversations about Artificial Intelligence. Every other post seems to be announcing the next great leap forward, explaining how AI is going to transform every industry, or warning that organisations which don't adopt it quickly will be left behind. The message is often the same: embrace AI now or risk becoming irrelevant.

Among all of this excitement, one phrase appears repeatedly. Artificial Intelligence is "just another tool".

It is an understandable comparison because, in many ways, it is true. A hammer is a tool, a chainsaw is a tool and a power drill is a tool. We create these things because they extend what humans can do, allowing us to build, cut, repair and create things that would otherwise be much harder. A skilled craftsperson using the right tool can achieve remarkable things that would be impossible through effort alone.

The difference with AI is that we are not simply creating a tool that makes our hands stronger or our work faster. We are creating systems that can analyse information, generate content, write software, support decisions and perform tasks that have traditionally relied on human judgement and expertise.

We are creating systems that can analyse information, generate content, write software, support decisions and perform tasks that have traditionally relied on human judgement and expertise

That makes the comparison with a simple tool slightly uncomfortable.

A chainsaw is powerful, but we understand how it works. We know what it can do, what it can't do and what happens when something goes wrong. Artificial Intelligence is different because the capability has developed so quickly that our understanding of its limits has not always kept pace.

Nobody would hand a three year old a chainsaw and tell them to work it out. Not because chainsaws are bad. They’re not. They’re incredibly useful tools that allow people to clear woodland, cut timber and complete jobs that would have taken far longer using older, more traditional, methods. The problem isn't the chainsaw itself. The problem is giving something powerful to someone who doesn't understand the risks involved.

That is why the image of "babies with power tools" feels strangely appropriate for where we are with AI. The comparison isn't about saying that AI is immature or incapable. These systems are already doing remarkable things. They can analyse information, write code, help researchers and support people in ways that would have seemed unlikely only a few years ago. The comparison is about the relationship between capability and understanding. We've developed tools with extraordinary power, but our ability to fully understand their behaviour, limitations and consequences is still catching up.

We've developed tools with extraordinary power, but our ability to fully understand their behaviour, limitations and consequences is still catching up.

I remember borrowing a friend’s electric drill when I was a student. He showed me how to use it and watched while I made one perfectly good hole in the wall. That was enough to convince me, and him, that I knew what I was doing. A few minutes later after he’d gone, I drilled straight through an electric cable. The drill had behaved exactly as it was supposed to, but I hadn't.

The problem wasn't the drill. It was my confidence. I knew enough to be dangerous, but I didn't know enough to recognise what I didn't know. Now I know that I am useless with DIY and we get the experts in! Looking back, that feels like a useful analogy for where we are with AI. We've seen what these systems can do, but we're still discovering where the boundaries are and, in some cases, whether we even know where those boundaries should be.

One of the uncomfortable truths about modern artificial intelligence is that even the people who design these systems can't always explain exactly why they produce a particular answer. That doesn't mean researchers have built something mysterious and have no idea what is happening. They understand a huge amount about how these models are created, how they're trained and how they behave. The difficulty is that understanding the overall process isn't the same as being able to explain every individual output.

Traditional software generally works because people write instructions. A programmer defines the rules and the computer follows them. Large language models are different. They're trained on vast amounts of information and gradually adjust billions, sometimes trillions, of internal values until patterns emerge that allow them to produce useful responses. Nobody has written a separate instruction for every question the system might be asked. Instead, the model develops complex relationships from the information it's been exposed to.

That is what makes AI so fascinating, and also what makes it difficult to manage. This is because it hasn't been programmed in the way many people imagine. There isn't a giant list somewhere saying, "If this question appears, give this answer." The system has developed ways of recognising patterns and generating responses based on the information it has processed. Of course, the word "learned" needs to be used carefully here because AI doesn't learn like a human being learns. It doesn't have experiences, emotions or an understanding of consequences. Yet it can produce answers that look remarkably like understanding. And that's where things become complicated.

AI doesn't learn like a human being learns. It doesn't have experiences, emotions or an understanding of consequences. Yet it can produce answers that look remarkably like understanding. And that's where things become complicated.

AI isn't alive. It doesn't have a mind hidden inside the machine. It doesn't have ambitions or intentions. But complex systems often produce behaviours that are difficult to predict simply by looking at their individual parts. A single brain cell isn't intelligent, but billions working together create something extraordinary. A single ant can't build a colony, yet the colony displays organised behaviour that no individual ant understands.

AI is not biological, but watching it develop has a familiar quality. We create the environment, provide the information and then discover what emerges. Sometimes the results are exactly what we expected. Sometimes they surprise even the people who built the system. The difficult part is that this is happening incredibly quickly.

Technology has always moved faster than society. The printing press, electricity, calculators, computers and the internet all changed how people lived and worked. Each created excitement and opportunity, but each also created problems that people had not anticipated. AI feels different because the technology itself is changing while we are still trying to understand previous versions.

An organisation might introduce an AI system because it wants to improve efficiency. Staff receive training, policies are written and risk assessments are completed. Then a newer model appears with capabilities nobody had considered and suddenly some of those assumptions need revisiting.

It is a little like watching a child grow up faster than you expected. One minute they are learning to tie their shoes, and the next they have found the kitchen cupboards and are attempting to bake a cake. You're pleased they are becoming independent, but you're also wondering whether its worth checking if they know how to safely use the oven.

The problem with AI isn't that it makes mistakes. Every technology makes mistakes. The problem is that AI can make mistakes confidently. A broken tool is obvious. A chainsaw that won't start is frustrating, but you know something is wrong. An AI system producing an answer that sounds professional, thoughtful and convincing can be much harder to challenge, especially if the person using it doesn't have enough knowledge to recognise the mistake.

An AI system producing an answer that sounds professional, thoughtful and convincing can be much harder to challenge, especially if the person using it doesn't have enough knowledge to recognise the mistake.

Most people have already experienced harmless versions of this. A Satnav confidently directing you towards a road that has been closed for months. Autocorrect changing a perfectly normal message into something embarrassing. A search engine misunderstanding what you meant and taking you somewhere completely different. Usually, we laugh, shrug the shoulders and carry on (unless the Satnav has left our car in the ocean).

The consequences are very different when the same thing happens in healthcare, recruitment or public services. An AI system supporting doctors may identify patterns that humans miss and help detect disease earlier. That potential is genuinely exciting. But what happens when it encounters a patient who doesn't resemble the examples it was trained on? What happens if the information it learned from contains hidden bias? What happens if a clinician starts trusting the recommendation simply because it came from a computer?

The technology doesn't need bad intentions to cause harm. It only needs to be wrong and for people to stop questioning it. And that’s the real danger. Not that machines will suddenly become evil, but that humans may become too willing to accept answers from systems that appear more certain than they really are.

The same applies to employment. AI will change the workplace. Some tasks will disappear, many will change and new roles will emerge. That pattern is familiar from previous technological changes, although that doesn't make the transition easy for those affected. Someone developing a new career using AI may see enormous opportunity. Someone whose job is changed or removed may see a very different future. It’s easy to talk about technological progress when you’re not the person trying to work out what comes next.

It’s easy to talk about technological progress when you’re not the person trying to work out what comes next.

The answer can't simply be to tell people to adapt. Organisations need to think carefully about why they’re introducing AI and what they expect it to achieve. Is it helping people work better? Is it improving services? Or is it being introduced because reducing human involvement appears cheaper? Those are very different questions. And this is why guardrails matter.

Regulation is often described as something that slows innovation, but history suggests a more complicated picture. Many of the technologies we rely on every day operate safely because rules, standards and oversight were introduced alongside their development. Civil aviation is regulated through organisations such as the Civil Aviation Authority, with requirements covering aircraft safety, pilot training, maintenance and operational procedures. Medicines and medical devices are subject to strict approval processes through bodies such as the Medicines and Healthcare products Regulatory Agency before they can be widely used by patients. Buildings are constructed under systems such as the UK Building Regulations, which set standards for areas including structural safety, fire protection and accessibility.

These rules don't exist because the technologies themselves are bad. They exist because powerful technologies can create serious consequences when they are poorly designed, incorrectly used or introduced without sufficient checks. Regulation is not about stopping progress. It is about recognising that innovation and responsibility need to develop together.

AI deserves the same level of thought. Not every use of AI carries the same level of risk. A system helping someone draft an email is very different from one influencing a medical decision, assessing a job applicant or supporting decisions about access to essential services. The more an AI system affects people's lives, the more important it becomes that we understand how it works, test it properly and know who remains accountable when something goes wrong. We don't need to put the toolbox away. We need to make sure the people using it understand what they are holding.

The more an AI system affects people's lives, the more important it becomes that we understand how it works, test it properly and know who remains accountable when something goes wrong.

Recently I watched someone pick up a nail gun for the first time. Before they touched it, the owner stopped them and explained what could go wrong. It only took a few minutes. At the time, it probably felt unnecessary to the person holding the tool. But it wasn't. Those few minutes were necessary because somebody else had already learned the lesson the hard way.

AI needs that same attitude. Not fear. Not blind enthusiasm, but a bit of humility from us. We need people who understand enough about these systems to ask difficult questions. Where did the training data come from? How has the system been tested? What happens when it gets something wrong? Who is responsible when the answer causes harm? Those questions aren't barriers to progress. They are part of using powerful technology responsibly.

Artificial Intelligence may become one of the most useful technologies we've ever created. It could transform healthcare, research, education and many other areas of life. But capability has arrived faster than understanding, and that creates a responsibility for all of us. We don't need to fear the toolbox. We don't need to pretend the tools aren't powerful. We simply need to recognise that handing over capability without enough understanding is where problems begin. Because the biggest risk isn't the technology itself. It's what happens when we give enormous power to something before we have learned how to use it properly. And in that case we’re just babies with power tools.