I Was Late to AI

Two years ago, when I went back to school, I was probably one of the few people around me who had barely used ChatGPT.

I am not even sure I had an account when I started.

That changed quickly.

Almost everyone around me seemed to be experimenting with AI in some form. Some were using it for writing, some for research, some for analysis, and some simply to see what it could do.

I remember feeling slightly left out.

That was unusual for me.

I have always been curious about technology, but I have also carried a certain skepticism about big technology companies and the way technology, capital, scale and influence can reinforce one another. Living in the United States probably made that skepticism stronger.

So when the excitement around AI started building, my instinct was not immediately to join the excitement.

But I have another habit that usually gets the better of that skepticism.

Once something becomes a real question in my mind, I tend to go deep into it.

I want to understand how it works. Why people are excited about it. What is real. What is exaggerated. What people may be missing.

AI became one of those questions.

My first mental model was fairly crude.

I imagined AI almost like an enormously powerful computer with access to vast amounts of information, memory and computing capacity.

Compared with a human being, that seemed like an unfair contest.

We forget. We get tired. We cannot read everything. We cannot calculate endlessly. We cannot hold enormous amounts of information in our heads at once.

A machine does not necessarily have those limitations.

So of course, I thought, it should be able to do certain things better than us.

But as I started reading and using it more, the technology became more interesting, not less.

At the heart of the current AI boom were large language models.

And underneath many of the impressive things they seemed able to do was an idea that sounded almost too simple.

Generative Pre-trained Transformer.

GPT.

Very crudely put, a system trained on enormous amounts of data, learning patterns and predicting what token is most likely to come next.

That description does not explain everything modern AI systems can do. They can be connected to search, tools, software, memory, databases and other systems. They can perform tasks far beyond producing the next word in a sentence.

But I could never quite get away from that basic idea.

Something built fundamentally around prediction was writing essays, producing code, explaining mathematics, analysing documents, creating images and increasingly being described as something approaching intelligence.

That fascinated me.

It also made me skeptical in a different way.

Was this really intelligence?

Or were we becoming so impressed by the outputs that we were giving the underlying system qualities that it did not necessarily possess?

I still do not know the answer.

And perhaps that uncertainty is one reason I kept going.

The deeper I went into AI, the harder it became to look at it as one thing.

It touches thinking, work, education, creativity, companies, data centres, energy, water and infrastructure.

And that is also why I have become uncomfortable with attempts to explain it through one grand analogy.

The printing press may help us understand one part.

The industrial revolution another.

The internet, electricity, computers and automation may each illuminate something different.

But none seems capable of carrying the whole story.

So over the next few pieces, I want to take AI one question at a time.

Not to prove that it will save us.

Not to prove that it will destroy us.

And not to pretend that I know where all of this ends.

I want to look at what we can already see, what history might help us understand, where the comparisons break down, and where I currently stand.

My views have changed as I have gone deeper into this.

I expect some of them will change again.

That is part of the point.