AI has reached top-level mathematics. What happens now?

CB News
4 followers
#ai #artificialintelligence #maths
This month, OpenAI claimed to have found a solution to the Navier–Stokes equations, one of mathematics’ Millennium Prize Problems. It is considered one of the hardest open questions in the field of mathematics and has left the math community in shock.
Ten thousand AI agents were able to solve the task in just 88 hours.
AI Generated·May contain errors
Video Summary
AI GeneratedThis video explores the profound impact of artificial intelligence on the field of mathematics. Top mathematicians express a mix of exhilaration and despair as AI rapidly evolves from mediocre to superhuman capabilities, solving long-standing problems. While AI accelerates scientific progress, it threatens the professional identity, training, and joy of human mathematicians.
Exhilarated.
Concerned.
I think everybody, if they're honest, how they first feel is despair.
So things seem really positive but also really very dangerous at the same time.
We're still on the beach, we can see the tsunami but it hasn't quite reached us yet.
This is a film about a profession on the front line of automation.
These are some of the very best mathematicians in the world,
from Oxford, Imperial, Cambridge.
The finest human minds and in a matter of months
they have been overtaken by artificial intelligence.
Everybody is surprised at how quickly it's happened.
I'm not sure anybody expected it this quickly.
During the course of this year they've gone from mediocre PhD students
to very good PhD students to experienced mathematician level to top mathematician level.
It's shockingly good.
It gets much much better and it's only accelerating as time goes on.
I believed it later than a lot of my peers
and I think a little bit I just didn't want it to be true.
But at the moment it's impossible to deny.
I've sort of slightly hoped that there would be some reason that things would slow down,
that it wouldn't scale up to the really hardest problems.
These are wonderful and terrifying times.
The rate of progress is astonishing.
A few years ago AI could barely do GCSE maths.
Now it's solving problems that humans have wrestled with for decades.
In September OpenAI announced the proposed proof of the Navier-Stokes equation,
one of the Millennium Prize problems.
A group of seven renowned problems set in 2000 to inspire human mathematicians to greatness.
AI has reached the point where even the best mathematicians struggle to assess its capabilities.
I'm not going to say superhuman.
I mean it has done something that humans couldn't do.
I think it's still too early to say whether there are
kinds of problems that it's better at than other ones.
I don't believe that generative AI is going to solve all the problems there are.
It simply isn't.
What it's doing is we've put every method we ever discovered onto the internet.
It has access to all of them and it's putting them together,
wiping out problems that humans were struggling with because we didn't have that synthesis.
Of course this poses huge questions for mathematicians about their work.
My own view is that people are going to be forced to use AI
because if not they'll just be left behind by the people who do use it.
I don't use it very much myself and that's a personal preference
and it's because to me it kind of destroys the joy of mathematics.
Mathematics is about exploration, it's about understanding,
it's about figuring out which questions to ask in the first place
and this is something that's still as far as I can tell exclusively a human endeavour.
My excitement is that science and mathematics is moving forward faster than ever in human history
driven by human beings and my worry is that we're losing sight of the humanity.
We've got to also remember that in six months time these systems will be much,
much, much better. I think they will be able to teach us things.
What if AI does get to the point where it can just, is better than humans basically.
If there's a problem that AI can't solve then humans won't be able to solve it either.
If we get to that stage then the role of mathematics will no longer be,
or mathematicians I mean, will no longer be to solve problems.
A big worry is the careers of future mathematicians.
The situation for young people feels basically dire. I think many people are upset and traumatised
by what's happening because it's unclear what the way forward for the profession is.
They don't think there's a future because they don't have access to the machines
or the machines hide their contributions and hide their work.
We just have to change the way we train but I'm convinced that we'll solve these problems.
And it's the field itself, the very nature of what it means to do mathematics
that has suddenly been thrown into question.
I see what's happening now as an opportunity to push mathematics forward
at a much faster rate than it's ever gone before.
One thing that's not coming with it is an understanding of how the problems were solved,
of the ideas that a human would have come up with along the way.
And if we continue like this then we run a real risk of not knowing what to pursue next,
not seeing the big ideas.
Some mountaintops we've been chasing for centuries and suddenly we have this
brand new means of suddenly jumping up.
It's like we can knock down the side of the mountain instantaneously
and in some sense we partially destroy the mountain.
What worries me the most is that we have an incredibly powerful tool
that can produce solutions, very deep solutions to mathematical problems
that go beyond what experts would be capable of discovering in a comparable amount of time.
But the motivation for people to spend the years it takes to reach a level of expertise
where you can actually understand what the AI is doing
won't be there to anything like the same extent.
And so we may find ourselves with an incredible tool that nobody can use.
I don't want to throw any stones but I don't think these tech companies are
solving these problems because they care about advancing knowledge.
That's not our incentives.
Our incentives are to discover truth and understand it.
Mathematicians of course are unusual in terms of the work they do
but what they're experiencing right now might not be that unusual.
You know automation can make us faster, more capable
but it can also take away something that we thought really belonged to us.
Identity, purpose, expression, joy.
Up until now the advancement of mathematics and the advancement of humanity have been the same thing.
It feels like AI might somehow prise those two things apart.
Yes of course that brings gains, maybe big gains to humanity as a whole
but on a personal level it can bring losses.
It might be that that hard bargain is one we all face as AI continues to progress.
