WEBVTT - Open AI solves 90 year old math problem 

0:00:00.070 --> 0:00:03.730
<v Speaker 1>So I've been following this really interesting story about what's

0:00:03.750 --> 0:00:07.320
<v Speaker 1>happening in the maths world. And for those of you

0:00:07.330 --> 0:00:09.600
<v Speaker 1>that don't know about it, a quick summary, right? So

0:00:10.260 --> 0:00:13.930
<v Speaker 1>on the morning of Tuesday, September the 8th, mathematicians at

0:00:14.090 --> 0:00:19.509
<v Speaker 1>OpenAI announced that a group of 10,000 autonomous AI agents

0:00:19.610 --> 0:00:23.150
<v Speaker 1>under their direction, running on an advanced model that's not

0:00:23.190 --> 0:00:26.229
<v Speaker 1>available to the public, had found a singularity in the

0:00:26.290 --> 0:00:29.960
<v Speaker 1>Navier-Stokes equations in three dimensions. So in other words, what

0:00:30.000 --> 0:00:34.019
<v Speaker 1>happened was they solved one of the six Millennium Prize

0:00:34.159 --> 0:00:37.470
<v Speaker 1>problems together. These were problems that were posed in 2000

0:00:37.470 --> 0:00:40.169
<v Speaker 1>by the Clay Mathematics Institute, and each one of them

0:00:40.229 --> 0:00:43.479
<v Speaker 1>carries a $ 1 million prize. So in other words, AI

0:00:43.630 --> 0:00:47.360
<v Speaker 1>solved this math problem that humans have been battling to solve.

0:00:47.380 --> 0:00:50.120
<v Speaker 1>So there's been all kinds of fallout as a result

0:00:50.320 --> 0:00:54.340
<v Speaker 1>of that. Dr. Liam Baker is a mathematics lecturer and

0:00:54.360 --> 0:00:58.620
<v Speaker 1>number theory researcher at Stellenbosch University. Dr. Baker, good afternoon

0:00:58.660 --> 0:01:00.810
<v Speaker 1>to you. Thank you so much for your time.

0:01:02.170 --> 0:01:03.470
<v Speaker 2>Hi, thank you for having me here.

0:01:04.260 --> 0:01:08.400
<v Speaker 1>Firstly, can you explain the significance of the fact that

0:01:08.920 --> 0:01:11.670
<v Speaker 1>AI seems to have solved this problem? Put it into

0:01:11.709 --> 0:01:12.490
<v Speaker 1>perspective for us.

0:01:14.069 --> 0:01:16.850
<v Speaker 2>Yeah. So as you mentioned, this is one of the

0:01:16.970 --> 0:01:20.860
<v Speaker 2>Millennium Prize problems. And they were, I mean, they're called

0:01:20.890 --> 0:01:24.020
<v Speaker 2>the Millennium Prize problems because they were announced, you know,

0:01:24.080 --> 0:01:30.690
<v Speaker 2>when like 2000 happened. as a sort of list of

0:01:30.800 --> 0:01:34.880
<v Speaker 2>seven big problems that are like the biggest, not the biggest,

0:01:35.000 --> 0:01:38.820
<v Speaker 2>but big problems that if mathematicians can solve these, it

0:01:38.840 --> 0:01:42.220
<v Speaker 2>would really move the field of maths forward. And a

0:01:42.260 --> 0:01:44.940
<v Speaker 2>lot of these have been unsolved for a very long time,

0:01:45.000 --> 0:01:50.570
<v Speaker 2>like this Navier-Stokes equation problem. So, yeah, the problem is

0:01:50.650 --> 0:01:56.070
<v Speaker 2>basically about whether... these fluids in three dimensions. We can

0:01:56.130 --> 0:02:01.330
<v Speaker 2>describe how fluids work, like water and other fluids, mathematically.

0:02:01.910 --> 0:02:06.290
<v Speaker 2>And the question is whether the equations we work actually work,

0:02:06.430 --> 0:02:09.169
<v Speaker 2>in a sense, and we're looking for mathematical proof of that.

0:02:10.930 --> 0:02:14.690
<v Speaker 2>So the proof was actually announced by OpenAI. There were

0:02:14.730 --> 0:02:17.710
<v Speaker 2>some other mathematicians that had also been working on it

0:02:18.690 --> 0:02:24.020
<v Speaker 2>with the assistance of AI independently. So both announcements sort

0:02:24.060 --> 0:02:27.680
<v Speaker 2>of came out at the same time. The significance is

0:02:27.740 --> 0:02:31.340
<v Speaker 2>that now it seems to be the case that AI,

0:02:31.540 --> 0:02:35.160
<v Speaker 2>if you put enough compute power into it, like you mentioned,

0:02:35.180 --> 0:02:38.250
<v Speaker 2>there was like 10,000 autonomous agents working on it at

0:02:38.260 --> 0:02:42.010
<v Speaker 2>the same time. If you put enough compute power into this,

0:02:42.270 --> 0:02:47.630
<v Speaker 2>AI may be able to do actual mathematical research, like

0:02:47.710 --> 0:02:53.000
<v Speaker 2>cutting-edge stuff, So in the math world, this is having

0:02:53.100 --> 0:02:57.500
<v Speaker 2>a big sort of shake-up, and there's a decent amount

0:02:57.560 --> 0:03:01.270
<v Speaker 2>of uncertainty as to what the job of a mathematician

0:03:01.310 --> 0:03:03.150
<v Speaker 2>is going to look like in future.

0:03:03.470 --> 0:03:08.050
<v Speaker 1>So is there a segment of the community that is obviously,

0:03:08.230 --> 0:03:11.829
<v Speaker 1>there's a backlash because they're saying that this is risky

0:03:11.910 --> 0:03:12.770
<v Speaker 1>or shouldn't be happening?

0:03:14.150 --> 0:03:18.299
<v Speaker 2>Yeah, so there's some group of mathematicians who just don't

0:03:18.419 --> 0:03:22.440
<v Speaker 2>like AI at all. And they're saying that AI doesn't,

0:03:22.960 --> 0:03:25.220
<v Speaker 2>because the way it works is that it's not an

0:03:25.340 --> 0:03:29.600
<v Speaker 2>actual person that's thinking about it. AI doesn't actually understand

0:03:29.720 --> 0:03:34.820
<v Speaker 2>what it's writing. So that's one of the groups. I mean,

0:03:34.840 --> 0:03:38.369
<v Speaker 2>because an AI can solve a problem and spit out

0:03:38.510 --> 0:03:44.170
<v Speaker 2>like a 300-page proof, but no human being actually has

0:03:44.250 --> 0:03:47.270
<v Speaker 2>the time to read through all of that and understand

0:03:47.350 --> 0:03:51.040
<v Speaker 2>it and The proofs that I have seen, sometimes it's

0:03:51.040 --> 0:03:53.440
<v Speaker 2>100 pages. You can cut it down to 20 pages

0:03:53.500 --> 0:03:56.190
<v Speaker 2>and just be a lot more concise and clear about it.

0:03:56.640 --> 0:03:59.380
<v Speaker 2>So some people just don't like AI at all. There's

0:03:59.440 --> 0:04:03.470
<v Speaker 2>another group of mathematicians that are like just fully embracing

0:04:03.530 --> 0:04:07.130
<v Speaker 2>like AI should become part of the way we do

0:04:07.250 --> 0:04:11.610
<v Speaker 2>math every day. So yeah, there are different communities in

0:04:11.670 --> 0:04:14.010
<v Speaker 2>the math world with different opinions about this.

0:04:14.560 --> 0:04:16.780
<v Speaker 1>And what does all of this mean, Doctor? Does it

0:04:16.860 --> 0:04:19.979
<v Speaker 1>mean that, you know, how is it going to practically

0:04:20.060 --> 0:04:22.620
<v Speaker 1>change the world for us if we're able to solve

0:04:22.680 --> 0:04:23.900
<v Speaker 1>some of these millennial problems?

0:04:26.790 --> 0:04:34.750
<v Speaker 2>So many of these millennium prize problems have basically no, like, direct, like,

0:04:34.810 --> 0:04:40.700
<v Speaker 2>day-to-day applications. Some of them do. Like, for example, this

0:04:40.900 --> 0:04:45.570
<v Speaker 2>P versus NP problem, which hasn't been solved yet. If

0:04:45.790 --> 0:04:49.790
<v Speaker 2>it was to be shown, because the mathematical question is

0:04:49.850 --> 0:04:53.370
<v Speaker 2>whether this one thing called P is equal to this

0:04:53.470 --> 0:04:56.550
<v Speaker 2>other thing called NP, and if you can prove that

0:04:56.589 --> 0:05:00.850
<v Speaker 2>those two are actually the same, that basically means that

0:05:00.890 --> 0:05:06.120
<v Speaker 2>a lot of things like your bank encryption, like encryption

0:05:06.150 --> 0:05:08.500
<v Speaker 2>that gets used in a lot of financial things and

0:05:08.540 --> 0:05:14.260
<v Speaker 2>security things, can actually be hacked more easily. Now, the

0:05:14.279 --> 0:05:16.669
<v Speaker 2>question is, you know, is P equal to NP or not?

0:05:16.700 --> 0:05:19.790
<v Speaker 2>So this, that particular problem has a lot more like

0:05:19.850 --> 0:05:23.270
<v Speaker 2>direct application to, you know, our day to day life.

0:05:24.010 --> 0:05:27.890
<v Speaker 2>Most of the others don't really have that much application

0:05:27.950 --> 0:05:30.190
<v Speaker 2>to like an average everyday person.

0:05:31.190 --> 0:05:34.440
<v Speaker 1>Fascinating stuff. Doctor, thank you so much for explaining all

0:05:34.460 --> 0:05:37.580
<v Speaker 1>of that to us. Dr. Liam Baker is a mathematics lecturer.

0:05:37.920 --> 0:05:41.280
<v Speaker 1>at Stellenbosch University, explaining this. I wonder, what do you

0:05:41.300 --> 0:05:43.610
<v Speaker 1>think about this? Do you think AI should be used?

0:05:43.630 --> 0:05:46.029
<v Speaker 1>You've got AI that is solving one of the six

0:05:46.089 --> 0:05:51.330
<v Speaker 1>remaining Millennium Prize problems, and some mathematicians are for it,

0:05:51.529 --> 0:05:53.469
<v Speaker 1>some are against it. What are your thoughts?