No One Really Knows the Future of AI

Lately, I feel as though there are fewer YouTube videos predicting the future of AI.

Not long ago, I was constantly seeing videos claiming that “jobs will disappear within five years,” “programmers will cease to exist,” or “you need to learn this technology right now.” Recently, however, I have not come across nearly as many.

I suppose they have run out of material.

Then again, I have also been writing fewer blog posts about AI lately. I use AI every day, but I cannot find many topics worth turning into an article. So I am hardly in a position to criticize anyone else.

Recently, I have reached one conclusion about the future of AI.

In the end, no one knows what it will look like.

The More I Understood How It Worked, the Less I Understood Its Future

The other day, I went back and studied how large language models work.

I gained a much clearer understanding of areas that had previously been vague to me, including the transformer architecture and the attention mechanism. I come from a science and engineering background, so I do not mind working through equations.

Yet the more I understood the mechanics, the more mysterious the whole thing became.

How can all those calculations produce such accurate computer programs?

At its core, an LLM predicts the next token. Yet it can understand complex requirements, work out a design, and even find bugs. How do abilities like these emerge from a system that merely predicts what comes next in a piece of text?

Of course, researchers are making progress in explaining these mechanisms. But we still do not completely understand what is happening inside a model.

So what will happen as these systems continue to advance? Will they match human intelligence? Could they even develop something resembling consciousness?

I have no idea.

Researchers around the world are debating these questions. There is no reason to think I could find the answers simply by revisiting the subject for a while.

Even Experts Are Not Very Good at Predicting the Future

Looking back, predictions about the future of AI have been wrong again and again.

Some technologies once thought to be decades away suddenly become practical, while others proclaimed to be just around the corner remain unfinished years later.

Even a few years ago, I did not expect generative AI to become this capable of writing usable code.

Progress has also become so rapid that an AI explainer video from only six months ago can already feel like ancient history.

Under those conditions, what evidence could anyone possibly have for confidently describing society five or ten years from now?

What I find especially amusing is that experts in forecasting the future do not lose their jobs when their predictions are wrong. They can simply publish another prediction and have a new topic for their next video.

It is quite an enviable profession.

Keep Up with Change Instead of Trying to Predict the Future

I have worked in IT for nearly 30 years, but I did not choose my career by accurately predicting the future.

I simply learned whatever technologies were needed at the time.

When AWS became widespread, I learned AWS. When AI became practical, I started using AI. If I need a new programming language, I learn it. I now use Claude and ChatGPT every day, and the way I develop software is completely different from how I worked only a few years ago.

This was not the result of exceptional foresight. Whenever a new technology appeared in front of me, I simply tried it. I have survived by repeating that process.

Come to think of it, the people who last a long time in IT are not necessarily the ones who are best at predicting the future.

They are the ones who quickly change their minds when their predictions turn out to be wrong.

The difficult ones are those who cling to their past successes and continue rejecting new technology. The technology changes, but they never update their own thinking.

Computers get a new version every year. It is a problem when the humans remain on the old one.

My Future Self Will Probably Laugh at This Blog Too

In the first place, what I write on this blog is closer to a diary than a set of predictions.

I write about what I saw or heard that day and what occurred to me while working. When I reread these posts a few years from now, I will probably find some that make me wonder why I sounded so self-important.

There will probably be quite a few of them.

But that is fine.

If I think differently now from how I did a few years ago, it means I have had new experiences. I would be more worried if I kept saying exactly the same thing year after year.

The same applies to the evolution of AI.

Will programming disappear as a profession? Will AI surpass humanity? Will it even develop consciousness? I do not have the answers today.

Rather than spending my time trying to predict the future, I find it far more interesting to experiment with technology I can use today.

In the end, my approach amounts to little more than hitting whatever pitch comes my way.

Then again, after nearly 30 years in this industry, I am starting to worry about my reflexes.

For now, losing the ability to read the screen because of age-related farsightedness feels like a far more realistic threat than failing to keep up with the evolution of AI.