Muhammadjon OzodjonovMuhammadjon Ozodjonovblog

Scene 8 · NOV 2026 · 5 min

AI looks backwards

Everyone Looks Backwards

Why “AI only looks at the past” is the wrong criticism, and what the real difference might be

In competitive programming, you get good by looking backwards.

You solve a problem, then another, then a hundred more. After a while you stop seeing new problems. You see old ones in disguise. This is just shortest paths. This is binary search on the answer. This is the same DP as last month’s contest, but with the array turned upside down. By the time I was winning regional olympiads, most of my “ideas” in a contest were really memories: patterns I had seen before, recognised fast.

Then I reached the national stage, and that stopped being enough. The hardest problems there didn’t match any pattern I knew. Nothing was wrong with my memory. The problems were built so that no single old trick would work. You had to take two techniques you already knew and bend them into something neither one was on its own.

I’ve been thinking about that feeling since I read a line on someone else’s blog: “AI is data, and data can only look backwards.”

My first reaction: so do we

The sentence is meant as a criticism. AI is trained on what already exists, so it can only rearrange the past and never produce anything truly new.

My first thought was: isn’t that true of people too?

When I think I’ve had an original idea, I can almost always trace where the pieces came from. Something I read, a problem I solved years ago, a conversation, a mistake. I didn’t make any of the ingredients. I only combined them.

And this isn’t just me. Newton wrote in 1675, “If I have seen further it is by standing on the shoulders of Giants.” Darwin’s theory of evolution looks like a lightning bolt, but he built it from parts that were already lying around: Malthus’s essay on population, which he read in 1838, what farmers had known for centuries about breeding animals, and years of his own notes from the voyage of the Beagle. Every piece was old. The combination was new.

So at first I wanted to write a post arguing that every new idea is a copy of an old one, and that AI and humans are doing the same thing.

But that’s not quite right.

Copying is not the same as combining

A copy reproduces something. What Darwin did, and what you do on a hard olympiad problem, is different. You take pieces that already exist and put them together in a way nobody has before. “Nothing comes from nothing” is true. “Everything is a copy” isn’t.

That matters, because it means looking backwards was never the problem. Everyone looks backwards. The past is the only raw material anyone has, human or machine.

So if the original blog is wrong about where the difference lies, is there a difference at all?

I think there is. But it’s not about the past. It’s about the question.

Who chooses the problem?

Think about how an olympiad works. Someone gives you a problem. Your job is to search everything you’ve learned and find a combination that solves it. You don’t choose the problem. You don’t decide whether it’s worth solving. It arrives, and you answer.

That is almost exactly how an AI works. A prompt arrives, and it searches what it learned for the best answer. And honestly, it’s getting very good at that. In most ways, better than me.

Now think about something nobody handed you.

In 9th grade I started Code Caravan, a hackathon initiative for high school students. Nobody assigned that problem to me. Nobody graded it, and there was no answer key. I started it because I decided it should exist. The pieces I used were all old: hackathons I had heard of, things I learned from competitive programming, how other student projects were organised. But the decision that this was a problem worth working on came from me.

I think that’s the real difference:

We both look backwards. Only one of us has a reason to look.

There are millions of ways to combine old ideas, and almost all of them are useless. What makes a person creative isn’t access to more of the past. It’s caring about something enough to know which combinations are worth trying. You could call it taste, and I think taste comes from caring: from having a problem that bothers you, a want, something at stake.

Darwin didn’t read Malthus by accident and then stumble into evolution. He had been obsessed for years with a question nobody asked him to answer. When he read the essay, he already knew what he was looking for.

Where this argument could break

I want to be honest about the weak spot in my own argument.

AI systems are no longer only answering prompts. Some can now run experiments, use tools, and pursue goals over many steps without a person guiding each one. They are starting to take actions in the world and learn from what happens, something I would have said only humans do.

So the line I’m drawing may not hold forever. If an AI ever chooses its own problem, not because a person told it to but because it wants the answer, then this difference disappears too. I don’t know if that will happen, or how we would even recognise it if it did.

But I’m fairly sure the original sentence was aimed at the wrong target. “Data only looks backwards” doesn’t separate AI from humans, because looking backwards is what all of us do. The better question is: who decided what to look for?

For now, I think the answer is still us. Which also means that the most important skill I’m building isn’t recognising patterns faster. A machine can already do that. It’s learning to notice which problems are worth caring about in the first place.

FADE OUT.

What do you think?
CUT TO:Scene 7. Purpose