Mathematician explains Turing's halting problem | Edward Frenkel and Lex Fridman

Turing was um Alan Turing who was considered as the father of modern Computing right so he actually did something very similar so he had a sculpting problem that he proved that halting problem cannot be solved algorithmically that you cannot out of all computer programs roughly speaking you could not you could not you cannot have an algorithm of choosing out of all can be possible computer programs which ones are meaningful which ones will not which ones will halt very depressing results all across the table or on the contrary life affirming depends on your point of view because everything is full of paradoxes so that means everything so you're right it's depressing if we are sold on a certain idea from the outset and then suddenly this doesn't pan out but okay so which my I report what if what if he proved that actually you know everything can be proved so then what what is left to do if you're a mathematician so that would be depressing to me and here there is an opportunity to do something new to do to discover something new which may be a computer will not be able to again with the caveat according to our current understanding maybe some new technology some new ideas will be brought into the subject and the meaning of the word computation like now we think of condition in a particular framework during machines or church thesis and stuff like that but what if in the future another Genius Like Alan Turing will come and propose something else the theory will evolve the way you know we went from Newton's gravity to Einstein's gravity maybe in the framework of that concept some other things will become possible you know so um it's not to me it's kind of like not so much about deciding once and for all how it is or how it should be but kind of like accepting it as an open-ended process I think that's much more valuable in some sense than deciding things one way or another you know I wonder I don't know if you think or know much about cellular automata and the idea of emergence I I often return to Game of Life yeah and just look at the thing amazing right and wonder the kind of things they can do with such a small um you know tools that from Simple Rules a distributed system can create complex Behavior and it makes you wonder that maybe the thing we'll call computation is simple at the base layer but when you start looking at greater and greater layers of abstraction you zoom out with blurry vision maybe after a few drinks you start to see some uh something that's much much much more complicated and interesting and beautiful than the original rules that are scientific intuition says cannot possibly produce complexity and Beauty I don't know I don't know if anyone has a good answer a good model of why stuff emerges why complexity emerges from a lot of simple things it's a why question I suppose another but every why question will eventually have a a rigorous answer not necessarily I could have an approximate answer which still eludes something like quantum mechanics 99 maybe we will be able to describe it was 99 certainty or 99 accuracy yeah and then maybe you know in in 100 years or you know next year somebody will come up with a different point of view which suddenly will change our perspective you know to this point I want to say also you know one thing that I find fascinating speaking of paradoxes and so on do you remember how everybody was freaking out about this blue dress and the blue was it blue or was it black yeah it was the yellow I think yellow or and white or black and blue it almost broke Twitter you know I remember that yeah that night so there are many examples like that where you can perceive things differently and there is no way of saying which is correct which is not for instance uh you got this uh the bars the Rubens bars you know where you have from one perspective it's a verse from another perspective it's two faces then there is this dark rabbit picture where you can Google it if somebody doesn't know they can Google it and find it it's very easy actually Ludwig wittenstein devoted several Pages Duck Rabbit in his book and so on there are many others there are like squares where you can square you can see the from different perspective this way that's why it's on so when we talk about neural networks we're talking about training data and stuff and so that you have some pictures for example that you feed to your program and you try to find the most optimal neural network which would be able to decide which one is it that the dog or cat or whatever but sometimes it doesn't have a definite answer so what do you do then so do actually it's a question I actually don't know has modern AI even come to appreciate this question that actually sometimes you can have a picture on which you cannot say what it is in it from one perspective it's a rabbit from an episode is a duck how are you supposed to train if you have a neural network which supposed to discriminate between distinguish between ducks and rabbits how is it going to process this you see well so the the trivial trick it does is to say there's a this x probability that it's a duck and this probability that it's a rabbit but that's a good approach but also I would say there's no like given percentages for instance actually at some point I was really curious about it and I looked and for some for each picture of this nature and there are a bunch of them you can easily find online my mind immediately interprets it in a particular way but because I know that other people have could see it differently I would then strain my mind and strain my eyes and stare at it and try to see it in a different way and sometimes I could see it right away and then I could go back and forth between the two and sometimes it could it took me a while for some pictures so in that sense even if this probabilities exist they are subjective some people immediately see it this way some people may see that way and I think that nobody knows not psychologists not neuroscientists not philosophers what to make of it the best answer the of course as a scientific mind I will I'm even though I say no don't look for interpretation leave some place for mysticism or mystery right I say that but of course I want a theory I want an explanation so the best explanation I find is from Niels Bohr's complementarity principle so it is like particle and wave that there are different ways to look at it and when you look at it in a particular way another side will be obscured think about it like the other side of the moon you know so like we are observing the Moon from one side and then we don't see the other side there is a complementary perspective where we see the other side but not the side we normally see but the Moon is the same you still there it's our limitations of being able to grasp the whole that's complementarity and we know that from quantum mechanics that our physical reality is like that rather than being certain rather than being one way or another and we should just as a smaller side in terms of neural networks mentioned that at the end of the day there's humans is built on top of humans or with Chad gbt that it's using reinforcement learning by human feedback we're actually using a set of humans to teach than our works yeah and that's the thing that people don't often talk about because or I I sometimes think about that those humans all have a life story each human that annotated data that fed data to the network or did the rlhf yeah uh that they have a life story they grew up they have biases so biases there's some things that they like there's some things they don't like which can kind of appear under the Raider screen they may not be aware that they are exercising those biases that's the point what you brought up is a very important issue here not so it might issue but it's not a bug it's a it's a feature in my opinion that implicit in the discussion of the question is thinking computational and so on is the idea that our conscious awareness covers everything within our psyche and we we just know that that's not the case we have all of us have observed uh other people who have had sort of destructive Tendencies so obviously they did things distracting for themselves and many of us have observed ourselves to doing that as part of human nature right so and there is great research in analytic psychology and you know in the past 100 years uh strongly suggesting if not proving the existence of what Carl Jung called the unconscious personal unconscious and also Collective unconscious this is a kind of a circle of ideas which are under the radar screen which lead us to some strong emotions and Inspire us to act in certain ways even if we cannot really understand so if we accept that then it the preposition that somehow everything can still be covered by our actions which are totally kind of neutral and totally like righteous and totally um conscious that it becomes really tenuous

Channel: Lex Clips

Views: 15,575

Keywords: ai, ai clips, ai podcast, ai podcast clips, artificial intelligence, artificial intelligence podcast, computer science, consciousness, deep learning, edward frenkel, einstein, elon musk, engineering, friedman, joe rogan, lex ai, lex clips, lex fridman, lex fridman podcast, lex friedman, lex mit, lex podcast, machine learning, math, math podcast, mathematics, mit ai, philosophy, physics, physics podcast, science, tech, tech podcast, technology, turing

Id: _iFeZptwFMc

Length: 10min 25sec (625 seconds)

Published: Tue Apr 11 2023