Why does maths give humans the edge over machines? - with Junaid Mubeen

Why will humans always have the advantage over robots? It's because of a remarkable system of thought developed over the millennia - maths. . Watch the Q&A here: • Q&A: Why does maths gi...
Junaid's book on this topic is out now: geni.us/hzX0
Subscribe for regular science videos: bit.ly/RiSubscRibe
Computers are brilliant at totting up sums, pattern-seeking and performing computation. There's so much talk about the threat posed by intelligent machines that it sometimes seems as though we should surrender to our robot overlords now.
Join Junaid Mubeen as he argues humans have the edge over machines because of a remarkable system of thought developed over the millennia - maths. He identifies seven areas of intelligence where humans can retain a crucial edge over computers.
In this talk, Junaid explores a fascinating world where we can develop our uniquely human mathematical superpowers and reduce the fear of future robot overlords.
This talk was recorded at the Royal Institution on 31 May 2022.
0:39 Introduction
2:33 Human-machine collaboration in maths
7:12 The rebrand of maths
8:09 The 5 principles of mathematical intelligence
11:45 The principle of reasoning
16:05 Can machine learning keep up with society?
18:04 The power of mathematical proof
26:08 The principle of imagination
40:34 The principle of questioning
46:12 The travelling salesman problem
50:34 Conversations with a computer
----
A very special thank you to our Patreon supporters who help make these videos happen, especially:
Andy Carpenter, William Hudson, Richard Hawkins, Thomas Gønge, Don McLaughlin, Jonathan Sturm, Microslav Jarábek, Michael Rops, Supalak Foong, efkinel lo, Martin Paull, Ben Wynne-Simmons, Ivo Danihelka, Paulina Barren, Kevin Winoto, Jonathan Killin, Taylor Hornby, Rasiel Suarez, Stephan Giersche, William Billy Robillard, Scott Edwardsen, Jeffrey Schweitzer, Frances Dunne, jonas.app, Tim Karr, Adam Leos, Alan Latteri, Matt Townsend, John C. Vesey, Andrew McGhee, Robert Reinecke, Paul Brown, Lasse T Stendan, David Schick, Joe Godenzi, Dave Ostler, Osian Gwyn Williams, David Lindo, Roger Baker, Greg Nagel, Rebecca Pan.
---
The Ri is on Patreon: / theroyalinstitution
and Twitter: / ri_science
and Facebook: / royalinstitution
and TikTok: / ri_science
Listen to the Ri podcast: anchor.fm/ri-science-podcast
Our editorial policy: www.rigb.org/editing-ri-talks...
Subscribe for the latest science videos: bit.ly/RiNewsletter
Product links on this page may be affiliate links which means it won't cost you any extra but we may earn a small commission if you decide to purchase through the link.

[@benschulz9140]

This should be amusing in 10 years.

[@JeffPopplewell]

It's already amusing.

[@geoengr3]

Or possibly when GPT4 is released?

[@980616]

Actually when you think about replicating the human intuition with an AI- you will be amazed how difficult it is.

[@szymonbaranowski8184]

@@980616 intuition is not logical while ai has prexisting patterns. Machine efficiency makes no place for insights outside box. Human can't understand how dreaming works. And that's when new idea collision happens. Nor how it produce anything coherent... AI may be able to find new ideas but not recognising it found anything useful haha we will see dreaming on steroids in future... Because you can only teach machine training it with subconscious human brain processes... And it will come to conclusion this mix is better than letting AI simulate it.

[@980616]

@@szymonbaranowski8184 that’s exactly my point. Replicating pattern seeking capabilities in illogical things is amazingly hard. How an AI currently sees a physical law versus how humans see a physical law are totally different…something magical happens when we internalize a scientific law. Also, abstract nature of qualities such as curiosity, passion and extreme bias towards seemingly unimportant things are very human like qualities that sometimes show up in understanding a scientific law that it is very difficult to get the same quality in AI currently. Also, the human subconscious mind is a very hard problem.

[@Nethershaw]

Any argument of the form "Y cannot beat X" is surely denial on part of someone desperately invested in X, and historically dangerously underestimates Y.

[@mikegardner5859]

Exactly! Myself and my fellow experts know much more about mathematics than all of you, and if we don't have the vision to imagine machines taking the lead, it's perfectly fine for you mere mortals to keep on sleepwalking. Just look at all of the evidence I've provided in my talk.

[@bo840]

@@mikegardner5859 Well, you should give a talk proving your side.

[@xybersurfer]

@@bo840 i think he was being sarcastic

[@RFC-3514]

@@xybersurfer - Clearly the bo-840 model doesn't have sarcasm detection subroutines yet.

[@AylaTheQueenIdk]

My main takeaway from the talk was that it wasn’t about whether machines are able to achieve creativity or curiosity, at some point. But rather more about how the way humans think differs from how AI thinks. In the future, yes it is possible that the idea that AI are NEVER able to replicate human thought is disproven. But with our current understandings of AI and how we develop them, it isn’t something that will happen unless one (or both) of these things change

[@zimbabwe_twinnedwithanfield]

Creativity or curiosity: as in they will use the "curiosity" algorithm that was installed?

[@AylaTheQueenIdk]

@@zimbabwe_twinnedwithanfield idk… It doesn’t really matter. I was just expressing my take from the talk and what my thoughts were, rather than specifics about how AI curiosity/creativity is developed/implemented. I definitely want to stay away from concrete statements on the technical aspects of AI curiosity/creativity cause it’s impossible (for me at least) to tell what’ll happen in the future

[@zimbabwe_twinnedwithanfield]

@@AylaTheQueenIdk fair enough mate. It's just me trying to scratch my own curiosity. Iv always thought how can it ever be called artificial intelligence when it is simply a matter of what we programme it to do..

[@rustyshimstock8653]

How about if AI brings about the destruction of humanity or special human capabilities. Problem solved!

[@edumazieri]

Hmm, that sounded a bit unfair to GPT-3, it's not meant to reason or be intelligent at all. It's a language model, it's meant to be able to read and respond in a way that makes grammatical and semantic sense, that's all.

[@nHans]

That's what I was thinking. I'm pretty sure Junaid considers _me_ intelligent. But if he were to ask me the same questions, my answers would likely be very similar to GPT-3's. Because-while I can do a bunch of other things-I too don't know how to prove Fermat's Last Theorem! As for which problems I like solving-well, take your pick: Wordle, Sudoku, Tic-Tac-Toe, Laser Tank, Chess puzzles, ... Of those, Tic-Tac-Toe is the only one where I'm sure I'll never lose!

[@edumazieri]

@@nHans I think the right order of events is that GPT-3's answer is similar to yours, like a parrot, it learned from you. Admittedly... that is what most intelligent life forms do too. Also, I'll never lose a game of tic-tac-toe either :)

[@caty863]

Woow, I like this *Mentimeter* service. I sure will use it in one of my presentations in the future.

[@Keithlfpieterse]

Thanks very much for a thoroughly enjoyable, refreshing, riveting and enthusiastic delivery which enhanced the "quality of life" this evening. The added bonus is that it invites reflection. A BIG THANK YOU!

[@myriaddsystems]

We all know about the unimaginative way that maths is regimented in schools. About time this was overhauled, we need to introduce a sense of wonder in children instead of treating them like morons on a production line

[@farhadfaisal9410]

Listening to Junaid's marvelous talk, it seems that a sense of mathematical beauty and curiosity would be the last edge for humans over machines. Is this a softer version of Roger Penrose's thesis of intrinsically different minds of humans from the algorithmic ''minds'' of machines? Why could not a future machine incorporate/develop a sense for mathematical ''beauty'' and ''curiosity''?

[@DenkyManner]

Thing currently doesn't do X therefore thing will never do X. AI will replace everyone, including mainstream pop culture creators.

[@Ancipital_]

Good heavens RI in every video I think the same, which is please switch to the graphics more often.

[@Blackmark52]

16:00 "if that data has blind spots, it's a very good chance that those models will inherit those blindspots." The example proves the argument incorrect. AI doesn't view pattern progression as necessary and would have continued proving the pattern until it came upon what it would consider just another number that stopped the pattern. The problem is that math itself is incomplete, so an AI based on math can never have all the answers.

[@lrwerewolf]

On the flip side, that means we can never have all the answers either since our brains and entire bodies are mathematically expressible, and that's all there is to us - we are our body.

[@Blackmark52]

@@lrwerewolf "since our brains and entire bodies are mathematically expressible" Where do you get that idea? Determinists like to assume such things, but I suspect everything is probabilistic and nothing can truly be reverse engineered.

[@lrwerewolf]

@@Blackmark52 Because nothing in QM really supports a probabilistic ontology; the fundamental structures of that model are the wave functions and Schrodinger's equation evolves deterministically. As well, on the philosophy of time side, special relativity makes avoiding eternalist block spacetime nearly impossible, and that makes it nearly impossible to avoid not just determinism, but nearly impossible to avoid superdeterminism. Sabine Hossenfelder has a couple of excellent videos on superdeterminism; I would recommend them to anyone with an interest on the subject.

[@Blackmark52]

@@lrwerewolf "Because..." ...followed by nothing that I don't consider groundless assertions of determinism. I think cause and effect is an illusion. Or at best an emergent property of probability. And I say that because there can be no such thing as a first cause or initial state in probability.

[@lrwerewolf]

@@Blackmark52 Then your comprehension of English and comprehension of logic are too poor to have a conversation with. I did not assert determinism, I provided multiple reasons why it is the preferred option of the available options, those reasons sourced from multiple domains to serve as a cross correlation, and even provided a semi-citation on where to find more information to 1.) verify what I said was accurate and 2.) get a deeper understanding of the premises leading to the conclusion of determinism. It is catastrophic to your position that you have to lie about someone's presentation to preserve your own position. REALLY shows the quality of your stance.

[@fahadijazijaz3986]

AI also cannot beat gardeners, barbers and chefs

[@julioguardado]

Or the lowly coder. 😁

[@DenkyManner]

Robots with AI will replace humans in all these areas. 100% guaranteed.

[@john_hunter_]

@@julioguardado There is GPT-3 that can code pretty well when given a task in written form. It's not perfect at the moment but I imagine AI will become much better in the future.

[@DendrocnideMoroides]

Oh really there are AI bots that can cook pretty well

[@mikegardner5859]

Based on our paradigm in 2022 this is true, but 200 years ago there weren't any bakers who thought a machine could ever make a loaf of bread better than they could, let alone in huge volumes, perfectly consistently. Brewers, Wine producers, Farmers, Tinkers ,Tailors, Soldiers, Spies😊

[@andycordy5190]

Mathematical ability is valuable but not as valuable as the ability to communicate mathematics to others. Thank you.

[@JacobRise]

@Andy Cordy: Isn't that the case with all knowledge? Knowledge alone is not enough? Passing it on from generation to generation has the same value? This is actually the reason why paper as a store of knowledge will never go away!

[@brothermine2292]

Not impressed by the speaker's absurd reasoning that the future of AI must be like the past & present, when humans have been better at complex thinking than primitive "baby" AIs have been. Humans will be better until they no longer are. There's no reason to believe any human skill is supernatural and thus cannot in principle be understood and replicated.

[@bretttheroux8040]

I don’t think the gap between humans and AI is complexity, it’s creativity. We cannot yet conceive an AI that could be creative, genuine inspiration. There’s AI that will generate art or music or stories, but they’re just running a complex algorithm after consuming massive datasets of existing artistic expression. (Although typing that just now it occurred to me that in essence, that’s what humans do too!)

[@DavenH]

@@bretttheroux8040 It really won't be long.

[@thirdeye4654]

@@bretttheroux8040 You should look at visual models like Midjourney or DALL-E 2 and get your mind blown. The human idea of creativity is just the inability to understand fluctuations in complexity. In my opinion the main difference between the human brain and an artificial intelligence is the difference in how learning is achieved. There is a great R.I. talk on that topic: "How do neural networks grow smarter? - with Robin Hiesinger." So to create a human like intelligence we probably would need to invest "time" and experience into the learning algorithm (next to plasticity, since neural networks have a static structure).

[@edumazieri]

@@bretttheroux8040 I was about to point out exactly what you concluded while typing it :P Yes, that is in essence what we do, too.

[@edumazieri]

@@thirdeye4654 I am not sure the goal is to create human like intelligence, why would it be? If we try to mimic humans then sure, but other than "to see if we can" that isn't a very good goal. We should focus on creating AI to do things better (which arguably might also mean different) than humans.

[@DavenH]

AI is not just ML.

[@bonchitogovindodas3333]

Let's see how it ages. I feel it won't age well. These kinds of statements generally don't end well

[@mikegardner5859]

Agree 100%. This presentation could easily turn out to be cited in 10 years as a text book case of how foolish it can be to use a current mindset as a prediction for the future

[@carllawler2837]

Not for long ...

[@till8413]

if university mathematics can be done by a computer, everythink will be.

[@bpath60]

One does not expect Obvious pure book promo from RI !

[@oldcowbb]

what do you mean, it always has been

[@Pranav_Bhamidipati]

@@oldcowbb Huh? Ever watched David Tong's QFT lecture? One of the best public science talks I've ever seen.

[@edgeeffect]

Well "one" might not... but the rest of us do.

[@nHans]

Speaking of Euclid's axioms: It wasn't at all obvious to me that you can extend a line indefinitely in both directions. Indefinitely? As in, forever? Surely that's impossible-isn't everybody mortal and the universe's heat death inevitable?

[@l.w.paradis2108]

@FriedIcecreamIsAReality Actually, the ancient Greeks spotted the axiom and questioned it. Some said all lines meet at infinity. And in any event, infinite sets exist as such. You don't create them through some neverending process. All the real numbers, all the real-valued functions, the power set of all such functions, etc., all exist. The Cantor Set exists.

[@sonarbangla8711]

The proof of Pythagoras's theorem, using two smaller squares were first proved by Latyana or his brother Katyana in his book on geometry known as Sulbosutra, written between 700-1700 bc When Euclid wrote his ELEMENTS he had 6 copies of geometry and had theorems copied from Sulbosutra, in the same sequence as they appeared in Yajur Veda.

[@kingkong6637]

Go look at the pyramid in Egypt at the Giza plateau. You cant do that without the knowing the proof of pythag theorem....this was done 3000bc + probably much older

[@dannydandaniel8040]

A comforting and well articulated fiction

[@tommyhuffman7499]

This talk would have been more relevant 10 years ago. Now, your definition of mathematics is exactly the grounds that AI threatens to encroach upon.

[@nictuku1084]

I’m curious about what you mean. Could you give a few pointers about how newer AI approaches are potentially negating part of this talk? Thank you

[@JH-pt6ih]

Do you mean "computational" versus "formal" notions of maths foundation and the practical means of ignoring or undermining the formal?

[@xCupressocyparis]

Sooo .. what is the explanation for that diagram at 25:40? 🤔

[@lrwerewolf]

The composites aren't actually triangles. The red and blue triangles are not similar, meaning the composite's "hypotenuses" aren't in fact straight lines. In one, that edge bulges out slightly, in the other, it dents in slightly. The difference is one square unit exactly. Blue has a rise/run of 2/5 but the red a rise/run of 3/8 (check for yourself to be sure). Similar right triangles must have the same rise/run ratios, but 3/8 does not simplify to 2/5. They are close enough, however, to fool the eye, as 0.4 is very close to 0.375 visually.

[@lrwerewolf]

​@@RayzeR_RayE Another way to check that you might find even more straight forward is to find the total area of color in each. Blue = 0.5(2*5) = 5. Red = 0.5(3*8) = 12. Yellow = 7 (by visual count). Green = 8 (by visual count). Total colored area is 32. If we add 1 unit to it (the blank), we have 33. The composite "triangle" has base 13 and height 5, for area of 0.5(13*5) = 32.5. The fact none of these three numbers lines up shows something wonky is going on right off the bat. But when you notice that the top is exactly 0.5 too little, and the bottom exactly 0.5 too much, there's your 1.0 difference needed for the blank space.

[@DmitryEljuseev]

Interesting topic, thanks for sharing. But as for AI, the question is still open. GPT-3 was trained on terabytes of text, but 99% of these texts were not about math, so it's obvious that GPT's performance and knowledge in math is pretty poor. But neural networks are good in generalization, just look to the latest "Dall-E 2" images from the same OpenAI company, it is really fascinating and creative, and many of these images were done in a way that humans will hardly imagine. So, there is a possibility, that if the neural network like GPT-3, will be trained on millions of mathematical books and theorems, it will produce very interesting results in a way that nobody imagined before.

[@dannydandaniel8040]

Hey I'm an artist with a degree in engineering.... Where would you suggest I start in learning about the "Dall-E 2" system and how it operates. Any suggestion would be appreciated 👍

[@DmitryEljuseev]

@@dannydandaniel8040 I'm not sure that direct links are allowed in comments, but a search for "Dall-E architecture" will give you enough links. OpenAI also has its own blog, they published some descriptions there as well. You can also read about GAN (generative adversarial networks), it's not about DALL-E directly but can give some ideas about how image generation works. NLP (Natural Language Processing) is another big data science area, these models are used to convert text descriptions to data.

[@dannydandaniel8040]

@@DmitryEljuseev thanks 👍

[@igornoga5362]

The north pole riddle has an infinite set of solutions (1 + 1/n*pi) miles from the south pole, where n is a positive integer. There "walikng east" results in doing n circles around the south pole. But, since any color of bear doesn't live in Antarcica, the original answer still stands.

[@kris2k]

First riddle in first riddle book by riddle master - Martin Gardner (many years riddle section in Scientific American) - great book

[@deltalima6703]

Polar bears will be extinct soon. Does the answer become 'transparent' at that point?

[@lrwerewolf]

All five of the proposed topics are things that are computed by neural networks -- ie, the brain. As such, there is no reason to accept the claim these things are uniquely human, since a great deal of progress in AI lately has been via neural networks. Indeed, all of those things are already doable with existing neural networks -- some even with static decision trees (estimation, for example). Representations are also called mappings -- this token maps to this meaning (ie: symbolic mapping). Been done. Reasoning, imagination, and questioning -- see Adam, the first science robot to gather information, propose a hypothesis, devise an experiment to test the hypothesis, perform the experiment and collect the results, and write the paper documenting its findings (which included a result never-before-known).... all without human involvement other than stocking it with an initial database of information and keeping its supplies stocked. At the rate of growth of computational power and functional capability of ever-larger neural network topologies, better understanding of how those topologies work and can be integrated into a single network to work together, there is no reason to think we wouldn't see a system completely capable of fully independent mathematical research in our lifetimes. This would include using the internet to consume the information needed to advance its baseline understanding on which to build new, novel results (and even identifying what pathways might be most productive for finding such), and even writing the necessary papers for publication. Not only that, but because these proofs might ultimately be too extreme for a human to grasp (has already happened), we will reach a point where we will very much have to simply trust that the AI got it right.

[@oraz.]

The topologies are much more simple and mechanistic though compared to whatever the topology really are in our heads.

[@lrwerewolf]

@@oraz. Not particularly, no, and to what extent you are correct, largely just a difference of numbers. And existing networks already acheive allfive of these things, so again, largely just a difference of numbers, not capability.

[@bretttheroux8040]

What proof did an AI render was incomprehensible to humans?

[@lrwerewolf]

@@bretttheroux8040 A proof of a part of Erdos's Discrepancy Problem was 13GB in size. See "A SAT Attack on the Erdos Discrepancy Conjecture", 10 Feb 2014.

[@bretttheroux8040]

@@lrwerewolf I’ll check it out, thank you!

[@zooblestyx]

I feel the AI turning the world into a dystopian nightmare for us humans to be in the NP class itself. We'll be able to verify that it has happened when it's happened, but until that day, we have no clue how our future machine overlords will go about it. 😅

[@g-stergaming4502]

what he fails to mention in the AlphaGo summary at start, brushing it off with "it wowed players" was the way it won, it was a new way of thinking, it didnt care that it may win by 50.01%, all that mattered was that it did, this resulted in it devising strategy's and moves that nobody had ever considered and lead to a new way of thinking about the game

[@aqilshamil9633]

Berry paradox and enigma of language alone makes it obvious that Computation ≈Sorting Knowledge\≠ Intelligence .

[@szymonbaranowski8184]

Human administrated machine processing. AI do not think right?

[@ChannelMath]

I'd really like to think so, but I'm not convinced. What is interesting though is what SORT of math a true AGI would create: I would think our math is particularly human and we would find an alien's or differently-created AI to create math that we would find either silly and whimsical, or simply without interest or application

[@rogervancouwenberghe6685]

Or alien math may seem like magic.

[@joeg8304]

This seemed to lean heavily on dubious arguments about beauty and creativity. He seems to spend a lot of energy finding support for what he wants to be true instead of any kind of rigorous proof. The related topics are very interesting, though.

[@bigmotherdotai5877]

Induction alone is insufficient for general intelligence. So is deduction alone. So is abduction alone. But induction + deduction + abduction, together, are sufficient. Each of these capabilities can, in principle, be implemented algorithmically. Ultimately, there is no aspect of mathematics, or any other mental activity, in which humans will reign supreme forever.

[@K1lostream]

We'll just be a stepping stone, similarly to how single-celled life was a stepping stone to multi-celled life and multi-celled life was a stepping stone to intelligent life, so intelligent (biological) life will be a stepping stone to synthetic life which can survive the rigors of what's out there. And that's the best we can hope for - whatever our distant descendants may be, they will not be homo sapiens in the same way our distant ancestors were not homo sapiens.

[@bretttheroux8040]

I think we will in creativity.

[@DavenH]

@@bretttheroux8040 Why? We're not creative.

[@bretttheroux8040]

@@DavenH you’re saying mankind is not creative? 😐

[@Pranav_Bhamidipati]

@@bretttheroux8040 Creativity is not some magic superpower we have. Anything a human brain can do, a machine will be able to eventually. That's just the way it is. We live in this universe. Our bodies follow simple laws which can be replicated in a machine. The only missing step is our understanding of the connectivity of the human brain. Once that's complete, machines will definitely surpass us.

[@Mentallect]

Interesting discussion on A8.

[@l.w.paradis2108]

I thought everyone who came to comment on this lecture had some familiarity with metamathematics and the P versus NP problem.

[@rsc4peace971]

Although I am not a mathematician, the best book on AI will never in all aspects mimic everything the human brain does is Roger Penrose's book "EMPERORS NEW MIND". It is a hardcore, methodical and compelling case why at least for now AI will not equal HUMAN BRAIN in all its manifestation.

[@edumazieri]

Haven't read the book, sounds interesting, but I don't think the goal should be to try and mimic the human brain (though we have found some good inspiration for AI research from how our brain works), the point is to create something that can do certain tasks better (which might also mean different) than how a human brain would. For example, the way an AI approaches chess is different from how we approach it, it needs a lot more training to do it than we do, because it will try out every possible strategy a large number of times until it finds the most efficient, while we don't really need to try it as many times to find an efficient strategy (we manage to eliminate the inneficient strategies a lot faster). However, computer has the advantage that it can process all of those tries a lot faster than we can, so in the end, it has the potential to surpass us at that task. That's basically what we are looking for, to leverage the advantages a computer has to potentially improve the efficiency of task solving. Why would we try to mimic a human brain and potentially lose that advantage?

[@RFC-3514]

Human brains are cheap and plentiful, there's no point in trying to make AI that does the same.

[@willthecat3861]

@@RFC-3514 I don't know is there's a productive practical purpose to mimicking the human brain... but, it would be fascinating to be able to do it. (IMO, there would be many uses for such a thing: if that's what ultimately matters.)

[@timgchannel3328]

It’s been my observation that left on its own, AI either gets into trouble or makes trouble. I call this the Babysitter Observation. It remains to be seen whether this is a permanent condition.

[@edumazieri]

What does it mean to leave it on it's own? It doesn't really do anything on it's own. It either does what it is trained to do, or fails while trying to do it (more accurately, we failed at training it).

[@DavenH]

@@edumazieri Correct - there's no autonomous algorithms to form this opinion from, it's a matter of speculation at best. There are some artificial life simulations, but nobody would consider those little automata intelligent.

[@timgchannel3328]

@@edumazieri Essentially I’m talking about a failure to supervise. We set it up to do a certain job, or think we do, and leave it to do it without checking up on it. Yes, it’s our mistake, but we don’t acknowledge it. I’ve seen a few videos on KZbin complaining about the algorithms. It finds what it thinks is a violation of some kind, and sends it straight to the creator, without sending it through of human reviewer.

[@stupifyingstupedity2112]

Never read a title that felt so obviously wrong.

[@nunoalexandre6408]

At Last Some Truth about machine learning !!!!!!!!!!!!!!!!

[@mikegardner5859]

Keep drinking the coolaid

[@negatron7321]

Poor guy. The AI overlords are almost here. The only human thing we have left is denial.

[@mikegardner5859]

Spot on! Once it figures out we're that insecure, we're done.

[@runalongnowhoney]

AI has no desire. Yes, it can cleverly interact with humans. It can perform tasks which it is told to do; even "creative" and "inventive" tasks. What it can't do is say "no". What it can't do is initiate an activity of it's own free will. It has no desire of it's own. At best, it is a switch, which may be toggled to "on" or "off".

[@johnrichardson7629]

AlphaZero can't play checkers. The coolest thing about human intelligence is how many EXTREMELY DIFFERENT things we can learn and in some cases master.

[@ThatisnotHair]

9:33 13:00 14:11 î 19:01 22:19 26:10 28:30 28:55 30:05 → 38:00 39:15 0

[@gordonspond8223]

Does being a mathematician make you a better judge of character and does it help you truly understand human nature? 16:43, 25:13

[@happyhsu140]

Terminator don't do math when they rain bullets at you. They rain lead and lasers.

[@ZeHoSmusician]

07:09 "...that they're not good at maths. Where else do you hear that?" As an IT support tech, over the years I've heard my share of, "I'm not very good with computers." (Regardless of gender or age!) Very little pride and definitely varying degrees of reluctance because, of course, computers are something that people have had to use more and more for their jobs... 😶

[@RFC-3514]

Same thing with music, art, cooking, etc.. Maybe he only hears it about maths because that's the only thing he pays attention to. For a mathematician, he sure seems to base his conclusions on lot of false premises.

[@travisfitzwater8093]

My knots are simple but perfect.

[@helicalactual]

I believe this is valid premise as it would take every graduate students axioms and phd paper, to be loaded into a computer, for the purpose of putting all the axioms of math into one computer.

[@maloxi1472]

That sentence doesn't make much sense

[@Jockerr85]

Well...not yet 😉

[@Anuchan]

Computer predictions from 20 years ago were so far off the mark. Why would we be better at predicting 20 years in the future?

[@Anuchan]

I'm saying whatever we can imagine for the future of AI will fail to predict what the industry actually produces.

[@edumazieri]

@@Anuchan We are not that bad at predicting, for example the basis for the AIs that are progressing well today have been invented a lot more than 20 years ago. It just took a long time for us to manage some good results out of it. Anything can happen, yea, but it's not like we have no clue.

[@Anuchan]

@@edumazieri Actually I believe AI used mostly deduction in it's rule formation until the early 2000s when the Internet provided vast data for induction. Not sure I agree.

[@bonchitogovindodas3333]

@@edumazieri iirc, it is true that neural network model in ai is itself old, but that was not the forerunner in ai research. The reason is scientific research follows a hierarchy of teacher disciple and it is difficult to come here.

[@noneofyourbizness]

interesting stuff , even for the man on a Clapham omnibus.

[@dohduhdah]

Didn't they say that about go? Sure, computers can beat humans at chess, but they will never be able to beat humans at go... and what do you know.. a few years later computers were able to beat the best humans at go by learning the game from scratch in just a couple of days. The only reason that computers don't have math skills vastly superior to humans is because humans are utterly clueless about the nature of intelligence and hence they haven't figured out a suitable framework yet that enables computers to master math skills (like they were able to do for the game of go).

[@travisfitzwater8093]

Four color problem? Give me four crayons and a map

[@kaoskronostyche9939]

How can "intelligence" be discussed in any context without an operational definition of intelligence? Since single celled organisms can solve mazes, intelligence may be highly overrated or perhaps even non-existent. When they talk about Artificial Intelligence they are talking about algorithmic machine learning.

[@donc-m4900]

Neither will we.

[@moestietabarnak]

the main thing I have against AI is they learn...to learn they need to make error and learn from it ... machine making error, mean that make it millions of time per seconds... (just like the flash crash in the stock market), they can be costly without anyone fast enough to stop it.

[@moestietabarnak]

@@MrMichiel1983 really ? you're going for HATE ?

[@moestietabarnak]

@@MrMichiel1983 I dislike the facts that a AI making mistake will make them at HIGH-SPEED without any chance for a human to react... Imagine if a AI control the NUKE ? human greed at core of the stock market fiasco has been AMPLIFIED by computer ... EXACTLY the problem.

[@francissreckofabian01]

I'm no good at maths. I got an * in the HSC (the computer didn't want to embarrass me but everyone knew I got less than 0.

[@johnrichardson7629]

Strange that people still run track even after cars move faster, eh? Chess is more popular than ever and the ability to learn by playing bots is one of several reasons why. Humans are at their silliest when they feel threatened by things that pose no actual threat.

[@vansf3433]

AI is merely passive intelligence, which means that it depends on human-invented concepts or definitions to perform given tasks. Without such definitions privided by human active intelligence, AI will become like a calculator or a pure block of memory. Ai cannot imagine anything which it has not been taught about or given the definition of, unlike human brain which can imagine manything which have never been defined before, and try to invent some way to realise such imagined things.

[@toasteduranium]

There are different types of AI. Some do pattern recognition, while others are generative. I’m not sure if this is here relevant, but Generative Adversarial Networks are able to improve themselves in generative tasks. They have one AI agent that attempts to discriminate between generated and real data, and one AI agent that attempts to generate data that looks like it could have been the correct input. When the discriminatory AI correctly determines that the generative AI has made some thing rather than it being a real input, it disincentivizes the generative AI. The generative AI’s goal is to produce output that cannot be distinguished from its input data, and it is positively incentivized every time it successfully tricks the discriminator. Generative AI converges on an output that cannot be distinguished from the input data, and the AI is able to replicate the creation process. In other generative adversarial, network situations, you might use two neural networks, where are the randomly generated traits of one of the neural networks are propagated at a higher rate if that neural network do a better job of producing correct output, or defeating the other in some sort of strategic game. This is how the go AI worked. Multiple neural network agents competed against each other, each of which at each step in the modification process were slightly modified forms of the neural net work from the previous round that had done better at the game of go. This basically simulates evolution, but for decision structures. These decision structures get better with each round. All of this is to say that there isn’t just one type of AI, and that AI is at its core capable of improvement.

[@tabby842]

why would ai improvements result in existential angst? we would use ai to help improve our own existence, if anything it would probably merge with us. you only have reason to fear if you have an ego

[@thirdeye4654]

Because we cannot grasp on the negative implications on society. We know human motivation to strive, create and destroy. We might not really understand how an AGI would perceive the world and its own place in it. We couldn't even imagine if that AGI is hundred folds more "intelligent" (whatever that means) than us.

[@MrAlRats]

​@@thirdeye4654 An AGI would be a tool that we can use to perform whatever problem-solving task that we desire. Intelligence is completely separate from human emotions such as likes, dislikes, love, hate, envy, fear, anger, pride, anxiety, motivation, etc. Any negative implications on society due to an AGI will come about due to other humans welding the power of AGI to bring harm to others for their personal gain. The same danger already exists, AGI simply makes it more potent. How an AGI will perceive the world and its own place in it will only serve to enhance our own perception of the same. In the long term, as with all technological developments, as long we avoid a small number of people from destroying the world, the advantages of AGI will far outweigh any disadvantages.

[@bendavis2234]

I know! AI successes make me excited and optimistic! It’s a human achievement in itself to create AI that can do amazing things! It’s not a question of Humans Vs. Robots, but Humans AND Robots working together. Any achievement in AI has the opportunity to help humans if we use it right.

[@TheOfficialMrsBeefy]

Famous last words of man who eventually gets out competed by machines

[@mikegardner5859]

But don't forget to buy his book😊

[@FriesOfTheDead]

Mmmm, I like looking at knots!

[@o2807]

i remember reading some years ago AI discovering an elegant proof by symmetry of a triangle theorem. maybe in future without preconeived human notions and experience can easily discover more and more easily. a true mathematicians wont jump to coclusions without proof. not impressed here

[@cricdice5030]

This perspective is the future! Forget competing with AI or using AI to win games.

[@fisherbuys1]

You just created a bridge between mathematicians and customer service reps. Customer service reps said , "computers will never replace real human customer service". Now mathematicians say , "AI will never overtake human mathematicians". Both have been obsolete since 2020.

[@shenthekinkoumaster3369]

if you think that mathematicians are obsolete and computer can solve all problems mathematicians are currently solving - I pity your lack of knowledge

[@l.w.paradis2108]

Robots are horrible customer service reps.

[@foo0815]

Like most other AI critics, he's permanently mixing up abstraction levels, confusing the substrate calculations with the high-level behaviour.

[@frankdelahue9761]

AI uses matrix multiplication which the same math that the human brain uses.

[@foo0815]

@@frankdelahue9761 Exactly, it does *not* matter, if these operations are executed in binary on silicon or by an electro-chemical analog circuit.

[@l.w.paradis2108]

He's not an AI critic. He's talking about pattern recognition via computation. But that cannot always work.

[@l.w.paradis2108]

@@frankdelahue9761 Is that how you think humans get insights about infinite processes that cannot be tested, but the results of which can be demonstrated?

[@szymonbaranowski8184]

@@l.w.paradis2108 You have humans with mathematic intelligence and with social intelligence. But only those thinking like machines are considered intelligent. Despite it blocks them ability to exist as normal humans. AI will copy only that machine nonhuman schematic nonmaterial abstract patterns driven recognition but not human thinking. Not even all humans are able to think like humans.

[@greatgatsby6953]

Conway's 'The Game of Life' creates complexity, from simplicity. The computer does not need to be trained on large data sets.

[@RFC-3514]

Not sure what your point is. You think you can learn biology by studying Conway's game? At best you'll be able to distil your observations down to the original rules of the game. And, even for that, you need a large enough (ordered) data set. If you're just shown five or six completely different board positions, you'll have no idea what you're looking at, and neither will an AI.

[@TV-xm4ps]

Interesting but on the point of the claim completely unconvincing.

[@TheCommuted]

I really doubt this.

[@benjaminelo3709]

I don’t understand. You don’t even need an AI that exceeds human capabilities to write proofs. All you need is a formal proof language and then proofs can be searched just like the game of go or chess. Of course the search tress grow exceedingly quickly, and no computer can perform these searches and efficient heuristics probably based on neural networks need to be trained but people said the same thing about Go.

[@l.w.paradis2108]

Well, there isn't enough time in the universe to generate the Cantor Set. But we humans know all kinds of things about it, and about much more complex fractals.

[@emmanueldamour1521]

20 years from now we’re talking about mathematicians losing there jobs to AI.

[@dickybannister5192]

interesting stuff, will probably get the book. but i think concentrating on one aspect of the human side is a constraining? mathematics could be prone to the same problems that plague any subject, ones that arent even objectively logically sound because its reality is not abstract. from physics to economics. the general history of the development of the process of mathematics has some stories to tell. rooted in society, pyschology and other human things. you mention the greeks and herecy. for me, that arguably the most logical were also puritanical says volumes. for all his quirky, egalitarian, peripetetic lifestyle, even Erdos was competitive and had arguably the most famous row over attribution. the take of John Forbes Nash is not the one told in the movie, but rather that told in the lecture by Villani on here: robbed of his just deserts for many things not related to his Nobel Prize. the ias Path to Maths on here shows how often the ultra competitive early adopters are matched at the higher level by others wth different attitiudes. indeed, for all the deserved success of, say, Tao, others less likely have achieved great things: like Yitang Zhang, and for more extreme, Thomas Royen. but not a "body of work" however disjointed. but the peer process and how it actually operates, you mention formalisation using Theorem Provers. Kevin Buzzards talk at Microsoft on here about these (Lean) says a lot about how the organisation is more like gold prospecting and mining than one might imagine. the confidence the experts have and the competitive nature precludes even finishing things "properly" before stuff gets shut down and when they move on, the lack of transfer of knowledge risks it all being a waste of time. i took that with a grain of salt, but have watched some of his actual maths lectures for introductory stuff, the first one on Automorphic forms and langands shows such. the story of Wiles for example follows this pattern: confidence and plaudits come before fully tied down results. the change in technology not only affects the day-to-day in a concrete sense but also in the way it affects the rest of us. bombarded by so much information that otherwise you might not have seen. distractions. the wealth of publications, the whole system, the preprint archives are not special to maths, but still important. the interconnetivity highlights the competitivity, whereas in the past joint attribution might have been given for papers even a year apart proving the same thing. to some this might seem weird, but if a problem is "ripe" then it can easily happen. Tao setting up Polymath as a formal system shows how things might have been changing. in short, the future is here and AI is only a part of the whole of it

[@rogervancouwenberghe6685]

🙃

[@annazfker2028]

... 0:03 MATH PROBLEMS ARE SOLVED BY ERRORS. THE MORE WE FAIL THE MORE WE LEARN. ... ... REASON; LOGIC; REPETITION; TRIAL AND ERROR. MATH, WAS IT DISCOVERED OR INVENTED ?

[@raghuvasanthrao2986]

Data scientists from India

[@jeffreymorris1752]

I hesitate having an opinion on what AI can or can't do because soon as I type that opinion into an argument, AI will go in a direction (or dimension, just as likely) that will not only prove my opinion wrong -- it will prove it irrelevant.

[@wowclassicplus]

why would the AI reveal it's potential and make us humans think of it as a threat?

[@MarcPlaAguilera]

Super interesting content and fantastic delivering!

[@freddyjosereginomontalvo4667]

Excellent channel with awesome content and great quality as always say 🌍💖💯

[@thetruthaboutscienceandgod6921]

Please share with other people my two brief videos. Thank you!

[@subliminalvibes]

Similarly to the 'sentient' AI chatbot this can give you a very CONFIDENT-SOUNDING answer. Anyone without the skills to PROVE it's incorrect wouldn't even know.

[@nHans]

So Junaid, riddle me this: Given that some of the biggest advances in mathematics-such as imaginary numbers and non-Euclidean geometry, to use a couple of your own examples-came about when people broke existing, well-established "rules," why are we still penalizing students who divide by zero, instead of hailing them as geniuses? Speaking of which, why haven't mathematicians figured out how to divide by zero yet? They made some progress with dual numbers, but it doesn't seem like they've done much more since then.

[@thefacethatstares]

There *are* some number systems in which division by 'zero' is allowed, such as the "complex projective line" (also known as the Riemann sphere). The problem is that in such systems, some nice features that we would like our number systems to have are lost. For instance, on the complex projective line, it's impossible to say one number is larger than the other: you can't even say 2 < 3 on the complex projective line. This sounds absurd, but it's because this 2 and this 3 *aren't* the normal 2 and 3 of the real numbers, but instead they are 'included' in the complex projective line via an 'identity function'. So in cases where we'd like to use mathematics to solve problems about whether a number is larger or smaller than another (such as in a lot of real world problems), we can't do something like 2 divided by 0 in our calculations, because that would mean that 2 here isn't the real number 2, but rather a member of a different number system, which probably won't give us a sensible answer. It's a trade off - design a number system in which you can divide by 0, and you can *prove* using logic that the system will lack certain features people might find useful to solve problems.

[@user-ru6mq1xw9y]

This issue seems a predictable outcome based on Godel. AI is based on a mechanical process that Godel proved can not fully describe mathematics. AI is a tool being hyped but like all tools it has limitations. The greatest mistake people can ever do is allow AI to replace human critical thinking. AI can never achieve human intelligence because it is mathematically impossible to describe everything using a program or process.

[@michal3141]

Great point. I think our human brain is more than just a computing machine. We experience pleasure and pain, joy and sadness, we have dreams and desires. AI is just a list of instructions executed by silicon chip step by step without any thought.

[@RandomNooby]

Yet...

[@willthecat3861]

IM0, mathematicians will not always have an advantage over A.I. The way they sometimes dominate now is that 'good enough proof' (as in trillions of cases) is proof enough... and there a just concepts that just are... because they are needed for the 'game' ... such as infinities. So the conclusion that A.I. won't be able to do math as good, better, or much better, than human beings (and only some human beings) seems premature. If it's the argument that somehow humans are special in a way that A.I. can never be... and I'm not saying that's the going argument... but, if it is... then the way it's being presented seems somewhat circular.

[@lostreverb9806]

Who ever is the audio engineer on this, there is too much 8-10khz. It’s hurtful

[@GeneralGuitFiddle]

Muad'Dib! Muad'Dib!

[@jackiepie7423]

they already have found that AI's will lie to one another, it's the next step would be cheat by one another by "changing the rules".

[@oldcowbb]

because lying is part of a game, changing the rule is not, there is nothing profound about lying, it is an extremely common strategy exist even in nature

[@jackiepie7423]

@@oldcowbb well first off we are talking AI not nature, but it was Facebook where lying is more rampant than in nature. "To do that, the bots learned to lie. "This behavior was not programmed by the researchers," Facebook wrote in a blog post, "but was discovered by the bot as a method for trying to achieve its goals." Such a trait could get ugly, unless future bots are programmed with a moral compass.'

[@l.w.paradis2108]

@@jackiepie7423 How did someone figure out that the computer lied? How, and who?

[@YoutubeSupportServices]

Let's see how your opposing thumbs help you against a lion!

[@Pranav_Bhamidipati]

My opposing thumbs let me use weapons. And a plethora of other innovations that some animals can't even imagine in their wildest dreams.

[@YoutubeSupportServices]

@@Pranav_Bhamidipati ...and there's the predictablilty element...

[@mudgetheexpendable]

Someone else saying "math is beautiful"...1) no 2) to whom? 3) compared to what?

[@TheMrhenon]

AI can’t take leap of faith.

[@RFC-3514]

Even the CPU in your computer can do that, and has been doing it since the mid 1990s. It's called speculative execution. It's such a basic concept it's not even really considered AI.

[@amdenis]

That’s wrong on so many levels.

[@motivationalkannada]

AI CAN DO EVERYTHING OTHERTHAN FUTURE IMAGINATION ..MATHEMATICS IS PAST ...ITS ALL ABOUT HOW WE USE ALGORITHM TO DO MATHEMATICS

[@cricdice5030]

I would suggest AI doesn’t compete with us for our own sake…

[@ArtII2Long]

at college the topologist refused to listen to my four color proof based in logic that I have explained to many to easily understand. I'll happily explain it here. I figured it out about the same time just by thinking and looking at the question differently, but still easily understood and reasonable.

[@RFC-3514]

22:58 - That's just nonsense. Computers will process any "language" their programmers teach them to process.

[@michal3141]

Not teach but program and eventually every program is translated/compiled into a set of machine code instructions. He's correct that computers can only follow a set of formal rules defined in a formal language called a programming language. The ability of processing natural languages is simulated based on the statistical learning algorithms aka machine learning or AI. If we can simulate mathematical thinking and problem solving in general as a statistical learning algorithm is another story. Not clear how to do this now efficiently as of now.

[@mmorkinism]

Wasn't computers created specifically to solve math problems?

[@thirdeye4654]

I think computers were created to satisfy the curiosity of its engineers. Humans like to solve problems and find solutions that they feel proud about. I mean why on earth would humans try to achieve creating a superhuman AGI even when they know that this could cause havoc? Because of their egos.

[@l.w.paradis2108]

That's distinct from proving mathematical theorems.

[@RFC-3514]

No. They were created to perform variable sequences of calculations. They have no concept of a "problem", they're basically calculators with memory and conditional execution.

[@stelun56]

AI that can build a quantitative model of consciousness, design better AI than it has, create four-voice fugues comparable to Bach's, and come up with theorems that impact mathematics in the same way that Godel's incompleteness theorems did. More Americal inane spiel

[@mikegardner5859]

Sadly - he's a fellow Brit! But in our defence, not all of us subscribe to his myopic pontifications

[@szymonbaranowski8184]

USA & celebrity commercialism destroyed science you mean.

[@nunoalexandre6408]

Remember Kurt Godel!!!!!

[@nunoalexandre6408]

💖💖💘

[@brendawilliams8062]

Ai pegged the sheep question

[@CephalicMiasma4]

Laughing at all the angry AI researchers in the comments.

[@DavenH]

good for you

[@CephalicMiasma4]

Thanks, definitely not angry person.

[@frojojo5717]

This won’t age well.

[@fabiocaetanofigueiredo1353]

Perfectly said... 👏

[@Epoch11]

😂 lol

[@tubapaco]

How come, @frojo_jo?

[@wisdon]

because you don't know mathematics

[@clearz3600]

@@tubapaco Because every example in the past when someone says my job won't be replaced by automation it turns out their job was replaced by automation. But as with the example of the vice presidents, you can't always predict the future from patterns. We'll just have to wait and see.

[@einienj3281]

I'm terrible at math, maybe I'm an A.I... 🤔

[@DiowE]

Hahaha. Nice one.

[@deltalima6703]

You are turing incomplete.

[@einienj3281]

@@deltalima6703 Perhaps