It flies, but not like a bird

AI is changing surveying. Get insights, risks, updates and practical ways to use it responsibly.

In early human history, if you wanted to fly, you copied the bird.

You strapped wings to your arms and you flapped. Some people died doing it. The bird was the only thing anyone had ever seen fly, so the bird was the model.

It never worked. Flight arrived when people stopped copying the bird. Fixed wings instead of flapping ones. An engine instead of muscle. A different mechanism that produced the same result. The aeroplane flies, and so does the bird, but they do it in entirely different ways. And the machine that stopped imitating the bird went on to fly further, higher, and faster than any bird ever could.

We keep reaching for the bird

When we try to understand AI, we reach for the thing we already know. The human mind. We say the machine thinks, learns, understands, reasons. We picture something like us, only faster.

It is the wrong picture. AI does what a mind does the way an aeroplane does what a bird does. The output looks familiar. The mechanism is nothing like ours. It has no understanding in the way we mean the word, no experience, no judgment. It reaches the result by a completely different route.

The computer scientist Edsger Dijkstra made the point decades before any of us were using these tools. He said the question of whether machines can think is about as relevant as the question of whether submarines can swim. A submarine moves through water. It does not swim. Asking whether it swims like a fish tells you nothing useful about what it can do. The submarine goes deeper, and stays down longer, than any fish.

That is the trap with AI. We keep asking whether it thinks like us, as though that were the point. It is not the point.

The machine produces the work by a route we would not recognise, and it is getting better at producing it very quickly.

Why this is so hard to see

Here is the consequence of the aeroplane and the submarine. Because AI does not work like a mind, we have almost no reliable intuition for where it goes next.

Nobody watching a sparrow could have pictured a passenger jet. The bird gave you no way to imagine the thing that was coming. When the pattern of progress does not match anything you already understand, your instinct for what happens next stops being reliable. That is where we are with AI. We are trying to forecast the jet by studying the bird.

There is hard data that gives the pace some shape. GPQA Diamond is a test of graduate-level science questions in physics, chemistry and biology, written to be difficult even with a search engine to hand. Subject-matter PhDs score around 65 percent on it. When the test was published in 2023, the best AI models performed close to guesswork. By 2026, the leading models score in the low nineties. A test built to stretch experts was, in effect, beaten in about two years.

And that is not a one-off. It is the pattern. Benchmark after benchmark has followed the same curve. A hard new test is released, models trail badly, and within a few years they saturate it and the field has to build a harder one. It has happened often enough that researchers now treat it as routine. We keep having to invent tougher exams because the machines keep passing the old ones.

A word of caution, because it matters. A high score on an exam is not the same as competence in the real world, and I would not claim otherwise. But the direction is not in doubt, and the direction is steep. Capability that would once have taken a career to arrive now arrives in a couple of years.

You do not have to accept any particular forecast to take that seriously. The tools your firm uses are getting more capable on a timescale measured in months, and none of us can see clearly where it lands.

What it looks like inside a firm

This stops being abstract the moment it touches real work. Picture a scenario that is not widespread today, but is well within reach.

A diligent surveyor carries out a building survey. Thorough site inspection, careful notes, a full set of photographs. Back at the desk, they use an AI tool to help draft the report from their notes and images. The tool produces clean, confident prose. It describes a defect, names a likely cause, and does it well. It reads like the work of an experienced professional, which is precisely the difficulty.

Because the tool has gone a step further than the notes actually supported. It has firmed up a judgment about cause and consequence that the surveyor had held more loosely on site. Nothing was rushed and nobody was careless. The output simply reads so well that the small gap between what was observed and what was written is easy to miss.

That is the risk, and it rises with the sophistication. A tool that tidied grammar was one kind of exposure. A tool that drafts a professional judgment about the condition of a building is another kind entirely. The more capable it becomes, the more of the actual work it touches, and the more it touches, the more sits on the line where a mistake carries real consequences.

This is exactly what the standard is reaching for. The RICS Professional Standard on the Responsible Use of AI, mandatory since 9 March 2026, applies heavier duties once an AI output becomes material, meaning it can influence the delivery of the service. And materiality is not a fixed line. It moves in one direction, because the tools keep getting more capable and more of what they do crosses into it.

The survey example is not material because someone decided it was. It became material because the tool got good enough to shape a professional opinion. The surveyor who used the same tool a year earlier for a lighter task may not have crossed that line at all. Same tool. Different exposure. The line moved underneath them.

The response is not to keep up

We are not going to fully keep pace with this.

The technology is moving faster than the profession, faster than the guidance, faster than most firms can absorb. But not keeping pace with the technology is not the same as being exposed by it, and this is where surveyors have an advantage they sometimes forget they have.

Surveyors assess risk for a living. You already know how to work responsibly with something you cannot fully predict or control. You do not need to understand every mechanism inside an aircraft to fly safely, and you do not need to master the internals of an AI tool to use it responsibly. What you need is to know where it sits in your work, what it is influencing, and to be able to show the judgment a qualified human applied on top of it.

So the response is not to master the machine. It is to stay clear-eyed about three things.

Where AI sits in your work. Not a vague sense that people are using it, but a real view of which tools touch which tasks, and where an output could shape a professional judgment. You cannot govern what you have not seen.

That the line keeps moving. The tool that was harmless for a light task last year may be reaching into serious judgment this year. What was safe once is not safe forever, because the tools do not stand still.

And the human decision behind the work. When a claim arises, the question will not be whether AI was involved. It will be whether a qualified human understood the tool, made the judgment, and can stand behind it. The record of that judgment is the thing that protects you.

That is the principle the GUARD framework was built on. You cannot see inside these tools, so you govern the human judgment around them and keep a clear record of it.

Look up

The aeroplane did not just do what the bird did. It changed what was possible.

We are at a version of that moment. AI is not a faster surveyor and it is not a mind in a box. It is a machine that produces something like our work by a route we do not fully understand, improving on a timescale we cannot clearly see. The profession that treats it as a novelty, or waits for it to settle down, is studying the bird. The profession that maps where it sits, watches the line move, and documents the judgment behind it, is the one that stays in control of its own work.

We cannot predict exactly where this goes. We can decide how we meet it.

Until next week

Nina