StarTalk Podcast: Cosmic Queries – Rise of the Machines with Matt Ginsberg

[Music] this is star talk sports edition a kind of a cosmic queries on ai sports march madness and what happens if we take sports to other planets i got chuck with me chuck hey neil all right and the person who gives authenticity to the entire concept of sports edition gary o'reilly gary hey nia all right an old footballer from the uk i was browsing the internet and i found a wiki page on you [Laughter] right there you were on the soccer field and you were looking looking buff and with that sexy sexy lids baby sexy legs so uh interested in gary you can dig him up on the internet so uh we we don't have particular expertise in march madness although we've done a lot of thinking about it so we have to go to our our a man about town who thinks about this kind of stuff and that's matt ginsberg matt welcome back to star talk thank you it's great to be back yeah so matt is like was a wonder kind with getting his phd in math at age 24 from oxford excuse me god you're such a slacker expert okay i took a year off at caltech okay and research mathematical physics my wife's specialty is mathematical physicist mathematical physics uh you're a scientist an entrepreneur an author you've got a book out there a fictional book the factor man i like fiction that is deeply informed by math and physics in the universe and that's one such book where you introduced god's algorithm the secret formula that will solve all uh all the problems of the world and who and the fight to take command of it i love that maybe we'll talk about it later i love where that's going also one of your companies what else excuse me excuse me yes i think those are called what are they called humble when you hear someone say well one of my companies not all of the companies right just one of them no it's just i get i get bored so i start a company and okay and it's like oh i gotta go do something else and i start another company and that's why i have this you know trail that's good so so the the the the inside another humble brag yeah yeah matt just shut up there just just stopped it like you know what uh some people when they get bored they turn on television when i get bored i start a company we're the potato chips you provide statistical support for professional sports teams in fact we delved into that topic the last time you were a guest i'll invite people to dig up that episode and what i love about what i learned in that episode is you wrote a computer program that can enter uh crossword puzzle competitions and it's called get this dr phil nice f-i-l-l very very clever so uh anyhow matt let's get let's get straight to march madness they're brackets you know teams win they it's win or go home is that right it's not it's like single elimination last i checked and is there any way to predict the whole thing that you know of that you've invented that you can share with us and if so how how do you how do you use it and what do you do do we all need a phd in math to make that happen can we get warren buffett's money so there you go can we really get warrants get right down to the basic question so my understanding is that in order to get warren buffett's money you have to predict every single game in march madness right and so i actually looked at this i knew we were going to talk about this that i looked at this last night and i think people look at this as you know what are my chances of winning and the answer is they're basically zero and i looked at it the other way around and i said let's say but the chance of winning would be the chance of getting every single game predicted correctly okay so if you're if you're even money on any particular game then you have you know you have one chance one time and two to the 64 will you actually get them all right and to the 64 is some giant number so you have no chance but computers are helping us make better predictions so i thought okay what if i wanted to have a one in a thousand shot which it's still not terribly good but it means that if i get to try 20 times then i have a 1 in 50 chance of actually getting it sometime in the next 20 years which is sort of as far out as i can look and it turns out that in order to have a one in a thousand shot at winning warren buffett's money you have to be able to predict with 90 accuracy who's going to win any particular game and some of the games that's pretty easy right when a one seed plays a 16 seed and you bet on the one seed it's probably about 90 percent so that's good the problem is when the seven seed plays the eighth seed yeah that game is really a toss up and getting to the point that you can predict that game with 90 accuracy is really hard there are computer methods that will help those computer methods one of the nice things about them is they'll actually tell you how how sure they are so you can put in all this information about two teams and say they're playing well did you get to put in how are the players feeling today i mean isn't that a factor that you're not considering here it is and you can put in obviously i i mean you could call them and ask i don't know that that would actually help um it would help them you know hi it's matt just want to know how you're doing today there you go and um if they took the call then you could make some judgments um i think that that a lot of the information that you need is actually buried there so for example one of the things that i've you can put in whatever you want so you can put in for example how many minutes of playing time did this guy have over the last three weeks that's very important and that's sort of a proxy for how tired is he yeah and you can put in how many minutes do you have the last three weeks of the last three days when was he last injured how has he been playing recently so all of this information is public and you can at least in theory push it all into a machine learning system and out will come i think this team is going to win and the probability is whatever wow now how they typically for a problem like this they typically use something called gradient boosting and that doesn't really help you much um what they do you mean is hearing you say that it's gradient boosting doesn't help as much well does crazy and boosting not help much i know which which is it that doesn't help radio boosting is very slick what it does is it says okay give me tons and tons of data about march madness historically going back as far as you have it just buckets of data and then it says okay i have data on 400 000 games i'm going to take a thousand of those games out and not look at them and i'm going to set them off to the side and then it tries to figure out what tendencies there were in the 399 000 games it has left and then it says okay this is what i think is going on and then you go and you get the thousand games and you bring them back in and you check and you see how well did i predict right now in order to get warren buffett's money because that's the uh that's the objective of this whole program no it's not going to be in a second in order to get warren buffett's money you need to predict these seven versus eight games with ninety percent accuracy in order to get vegas's money you probably need to predict them with sixty percent accuracy so if you're that good at predicting who's going to win these games you should go to vegas you shouldn't you should not try and win warren buffett's money it's a much it's a much further reach and it's not that much i mean it's a million dollars right right yeah for a year for the rest of your life if you can predict you can get that money from vegas and it's easier right if you can predict 60 of the time because you just come back every year and you get the money you get the vegas money every year but matt if we use that model you've just described it talks to me about history now from a player's point of view that's great it means nothing to me because we're not playing those teams those players are no longer playing we're different players this for in my mind that's history it has no relevance to what is about to happen if you look back historically you will see that his that five years ago lebron james scored way more points than average and yes that's history but it certainly bears on today i mean he was just a better player and you can you can also look and you can gather information about how right now today he sucks right yeah how old is he how do people age how do people like lebron age and you can just there is so much data now now what makes it hard the reason people don't just go beat up on vegas with this stuff is because there's so much data in something that happens with machine learning when you have too much data programs start what's called overfitting which means they look and they say wow whether or not the lakers won last year was correlated with the phase of the moon oh dear and that's probably that is history and that's just luck but if you have so much data some of it's going to look relevant when it really isn't and the reason machine learning is hard is because you have to somehow filter that out you have to somehow say which of this actually matters and one of the reasons people have this thousand games off to the side right is because then they say wow the lakers in the phase of the moon and look at the thousand games and it's like that didn't work right and then they go and they try again so you're using the the data uh itself as part of the predictive model kind of like what they do with climate you you put in the history of climate to show whether or not the predictive models that you're using right now are effective yeah that's right that's right that's right but so just i want to emphasize something that you said matt um and you said it brilliantly and beautifully i i don't want it to go by without sort of adding further punctuation if you put in there's there's a point beyond which you have so much data that you can find car relations that have nothing to do with cause and effect and and so you need enough ways to check against that to remove those correlations from the analysis is that a fair way to to summarize some of what you just said it is and and the effect is very real i mean you can you can go online and you can you can look for weird correlations and you know it's things like the number of loaves of bread eaten in each year in denmark is incredibly well correlated with whether the u.s stock market goes up or down something you know they're they're just craziness and and as chuck said it really is a matter of you have there are words for it so you have one set of data it's called the training data and that's the data you use to build your model and then you have this other set of data which is called the validation data which is the data that you use to test your model and see if it actually turned out to work and it's what you can't do what it seems like you want to do is when you train your data and you look at the validation data and it doesn't work because you were looking at the number of loaves of bread eaten in denmark so you make a new model that doesn't consider denmark and you try that on your validation data and it doesn't work and then you do it again and again and again until you find something that works on your validation data well you've cheated now your validation data is sort of dirty because you've been using it over and over and over again and your basic models think they're working with just the training data but you're working with the validation data right so you're stuck right so you need you need clean data and clean data is incredibly valuable because what happens you have these four hundred thousand games feels like a lot and then you take ten thousand aside and say uh i'm going to hold them for validation and you notice oh maybe that's not enough i need i need another validation set and you take ten thousand more aside and then you take ten thousand more aside and and once you touch data it's dirty forever so how you split data up between validation and training data turns out to be important and hard and it really matters that you not look at your validation data very often how many relevant data points would you expect someone to want to use for a decent outcome it depends on what you're trying to do some things you need huge data sets so if you're trying to predict with ninety percent accuracy seven versus eight you need at least as much data you need everything if you're trying to predict if a number one seed is going to be the 16 seed you probably buy a much less so different problems require different amounts of data the one by the way somebody somebody or some committee seated the number one team as number one and the number 16 is number 16. so they're using data to do that yep yes so so you're not you're not approaching those two teams from a pure you you are already biased by the setup for that and i remember as a kid and i would i didn't i was kind of very literal and they would say oh uh oklahoma upset texas today by beating them in a basketball game and i say of course they upset them because they lost yeah and so i didn't know just the concept of upset was not didn't i didn't understand that it's because you won beyond the expectations of some group of people who decided that you should have lost so does that add another sort of variable of unpredictability to what you're doing sort of i mean the bottom line is that the the seating committee of the nc 2a bracket is looking at very limited amounts of data they're typically looking at who be whom and perhaps by how much yeah and they look at the you know how well those teams do and the records of your opponents but it's a it's a tiny amount of data relative to everything you actually could use like how long have people rested who's injured where are they playing if i uh oh no or not or not and because if they were accurate no one would ever be upset yeah in a tournament correct and right the only person who would be upset would be warren buffett because he would keep losing but um probably the data you would use if you were trying to do this seriously is a vast superset of the data that the march madness guys you know the nc2a guys actually use there's um there's something else that i think it's important to to realize here and i think uh and this has to do with how people solve problems and how machines solve problems so i often draw a distinction between what i like to call a 99 problem and a 49 problem so in a 49 problem you're trying to distinguish between a 49 probability and a 51 probability so the stock market if you can only invest in stocks that are 51 to go up you will make a killing because you'll just you know the stock market's 50 50 and just that little edge you'll do phenomenally 99 means you have to just get it right so this my sort of standard example of a 99 problem is stop lights if you identify 99 of stoplights correctly and just drive through the rest because you don't realize they're there you're going to die you're going to have an accident you're going to die so machine learning the kinds of things i was talking about turns out to be really good on the 4951 problems and really bad on the 9900 problems basketball is interesting right because it's sort of in between the stock market is on one side traffic lights on the other side people turn out to be pretty good at the 9900 problems and we're not so good at the 4951 problems so the the answer the answer is we're not getting that money single-minded [Laughter] the answer is you're probably not you might go go for it yeah yeah go send your spend your mortgage money i don't want to dash your hopes right exactly yeah go ahead chuck go ahead bet your kids 5 29. guys we got to take a quick break but when we come back this is really a a patreon cosmic queries for march madness and i want to get a few of those questions in uh related to that but also uh maybe we can think about sort of sports on other worlds and what that might be like when you return we're back star talk sports edition march madness that's what we're talking about i got chuck and gary here guys yes yeah we can't do this alone we needed someone who could bring the analytics into the house and we've got our fame favorite analytic guy matt ginsberg matt always good to have you here on star talk with his with his math phd just spilling it out whatever he's got to do it plus i want to tell i want to take sports off world and see where that can where that goes too oh cool with questions that patreon members have but so all this cred you have mathematically do you ever do you ever game the system you know with with with any of your programming any of your machine learning any of your mathematical prowess do you ever put that to work and just say you know what i'm gonna i'm gonna go ahead and make a little little wait wait chuck i have the answer so don't wait i can answer for you matt here so either he's a miser or the answer to that question is no because look at the room he's in there's no butler there's no staircase going up to three-level mansion the dude is just in a regular room so um it's not it's not fun right i had a friend who went and played blackjack in vegas and counted cards it was very mathematical at the end of it all he said he figured out that he had made 15 cents an hour and it's such a grind and i i like i'll tell you the the best it's a grind making money it's such a grind oh the best thing that ever happened to me in terms of mathematical abilities is my daughter uses me as a calculator so i'll be walking around the house and she'll dad what's 17 times 36 and occasionally we have house guests who look at her and they say scott did you just use your father as a calculator and she said yeah that's what he's for so that's great well guess what how old is your daughter well now she's 22. okay i was going to say because at this point she's using you as a cash machine dad can i have 17 times 36 worth of dollars that's right that's cool all right well that's a good answer yeah all right so so give me that chuck who's got the first uh patreon question um i don't care gary you want to go all right yeah i'll jump in so these are patreon patrons our exclusive patreon patrons nancy diaz she says here's a shot at march madness question how can ai take into account things like motivation styles of play and performance under pressure or other factors challenging to quantify in determining the outcome the example she cites is virginia's national championship when in 2020 given their embarrassing loss in 2019. so matt what do we think um is this sort of enrolled in your in your data points that we told you yeah can you quantify motivation probably i mean in theory you can sort of quantify anything motivation probably shows up let me let me look at this exact example and you know in the data is going to be well how well do teams who got to the finals do if they lost in the finals the previous year you've got a reasonable sampling there and that's what that can you know and then these machine learning algorithms that just you just drop all the data in and they just turn it churn it up and try and look for patterns and if that's a pattern it'll find it i've looked for patterns like that and i've never found them college athletes seem to sort of always be about as motivated as they can be because they love the game they're trying to get into the pros they're really working but if there is a pattern it should be somewhere buried in the data and you should be able to pull it out interesting wow okay all right that's pretty cool actually and so what hap what happens when you have a team of four starting freshmen because there won't be any data on them so there's there's data on a variety of things there's there's data on for starting freshmen yeah wait wait wait you have the whole season behind them what did you have about to say you have the whole season behind you by the time you get to march madness right um you have data on how does a freshman's performance in march madness compare to where that freshman's performance over the course of the season do freshman choke right that's okay exactly that's exactly a point that you need to recognize whether or not the this the occasion crushes the player and one of the things that's that's cool about these machine learning algorithms like gradient boosting is you don't have to figure out what you're looking for in the data you just pour all the data in and the algorithms figure out oh this is a pattern oh that's a pattern and they do the mental heavy lifting for you now they're stupid so they might find some patterns that aren't really there and that's an issue but that's what this validation data is for so you in theory they are these very general purpose algorithms that are capable of finding the signal in the noise got it so so it's not as though you are quantifying the thing itself is you're looking at the statistics in the larger data set of the manifestation of that of that motivation right so uh so in other words you can't go up well maybe you can go up to the one person see they're really jacked they gave a really good pep speech pep talk the coach gave a crept talk and now they're just gonna win you can't put that in after the fact right you don't know that you're not adjusting these these correct the statistics in real time you have to go in with the bet already placed i have to decide what data i want to put in and train the thing and then what comes out is sort of how i'm going to make my bets right and i'm not because motivational coaches can have their influence as well but that will be there right because you'll see oh this coach players playing for this coach do a little better when the same players went and played for a different coach so this coach must be good i don't know if it's motivation right so i could imagine a world where for some strange reason on your 18th birthday you can't play sports you just you become uncoordinated on your 18th birthday exactly and then it gets better in the data that might look like freshmen occasionally have a bad day because they turn 18 on their somewhere and i wouldn't know if freshmen tend to choke because the pressure gets them and they behave a little bit less evenly or freshman or there's this miraculous 18th birthday thing now if their birthdays were in there well then i would see oh look it always happens to people on the day they turn 18 and then it's because they got [ __ ] faced that night of their birthday well that might be the reason but it's the thing everything is lurking in the data it's very difficult to imagine a phenomenon that that both matters and is not somehow present in the data for you to tease out if only you knew how to look so that's the lesson here that's it really yeah all right let's that is the lesson all right let's let's fly this thing out of the earth's atmosphere shall we uh abby chris heyo experts he says i've been watching the expanse which made me think how we could we conduct basketball tournament with people who are from various planets and asteroids that have been settled by humans like for example champions of mars versus champions of jupiter's moon right now this for me is interesting because you've got a whole new set of metrics to factor into your machine learning here what are some of the things that we would need to think about to keep the field equal and fair for all teams love the show and everything keep informing fellas so you're welcome wow so let me just as a way of lead into that um so so matt when people started training for the olympics at high altitude no it's not another planet it's this planet but it's a different environmental conditions under which their body is getting trained and now they all go into the same stadium and some people outperform the others so that's just an interesting realization there that maybe the environment in which you train the gravity the air quality the air density the uh that can can definitely show up in your performance i think that's right and i don't think we don't historically i mean i have no idea what's going to happen when we have people from mars but historically we don't try to compensate for that so you know kenyans win marathons it's just how it is and and there's but there's not anything in the rules saying that any kenyan entering a marathon has to have ankle weights as a as a handicap as a handicap there's nothing in the nba to try and make it easier for short people to play give him a step shoe just put a little trampoline for their dunk [Music] so i think that if um if martians you know people from mars have some physical difference that makes them better at a game hopefully we will just celebrate with them probably it sucks to live on mars so we'll celebrate with them that they're better fascinating i love watching marathons because the people who are so good are so good and i don't know of anybody i mean i never bemoaned the fact that i'm i'm not going to ever be a professional wrestler it's just not going to happen there's no time there's still time you're being vicious there chuck and i mean i mean if we bring it back to this planet just temporarily if i have to cross from the west coast to the east coast and if i am altitude and we come and play at sea level there's recovery then there's oxygenation and ability to approach these things are these not factors that are relevant of course they're relevant just like home field advantage is relevant right not anymore because there's no fans that's right no but playing it fenway is different than playing at wrigley field the green monster matters and and boston players boston selects their players in part because they're looking for people who have the natural skills that will exploit the peculiarities of their park as did the yankees for so many years everywhere the right field line was one of the shortest of all ball fields it was something like 296 feet or something it was very short and so the yankees had a lot of lefty sluggers right that racked in the home runs simply because of that short porch out on right field and in fact if you you're right matt we don't go back to the record books and say you know half your home runs were 310 feet and they would not have been a home run in any other stadium but you happen to play for the yankees so we're going to subtract those we don't do that we just allow the circumstances to be the expression of that of that ability so that's an interesting take on this do you know uh the tug of war used to be in the olympics as an as an event and the rule was everyone in the tug of war had to be the same profession they had to be a group that made sense that they competed together they all had to be like medical doctors so they all had to be you know soccer players they all had to be policemen or or and so it turned out that the mounted police always won because they you also had to wear your native uniform and they all had these steel reinforced boots and then just realize this is a stupid event and let's just get rid of it they're also all used to pulling on the reins of an obstinate horse so like let's go so i mean the closest i've seen to this my wife um was a hydroplane racer wow and wait wait she was a what a hydroplane hydroplane racer she was the national champion that's a thing crap yes these little boats that fly across the water and it's actually it's one of the coolest sports ever it is so her class had a weight limit and she was like the only woman she weighed way less than everybody else and they actually made her put a plate of lead in the bottom of her boat right so that she could meet the weight limit and she hated it because everybody else you know you're there and you're driving and you lean and you move your whole weight and she could not move the lead plate it was just stuck on the bottom of the boat and she could lean but she was leaning with much less mass than why didn't they why didn't they competing against load pockets in a vest on her so that she could lean with the weight and match other people's capacities she decided the lead weight was better i mean she was ridiculously and she she's beat them all anyway so she was fine but um she's wow she's just not competitive that's awesome i could have i could have destroyed you instead of just beat you also speaking of home field advantage in an interplanetary contest so it would matter if you played your sport in on your planet right i mean that would matter presumably yeah that would be the the ultimate home field advantage because you know your gravity and your air quality and your your all the peculiarities of your environment plus neil right if i have to travel from planet a to planet b and it's x amount of light years i have to get there and acclimatize so therefore it's recovery i'd have to turn up however many years years ahead of time my journey yeah no you know it'll be it'll be like the the the the uh what do you call it in tennis where there's the four events but they're not all in the same kind of court so the grand slam so the so wimbledon is on grass yeah and who is it on clay uh is it right right right running backwards yeah the french opens on clay and then you have the concrete you know at forest hills so that's interesting so if you if you do the whole circuit then you need a combination of abilities so which is what makes winning the full grand slam so that's so much more impressive than anyone who's only has the talent for just one so no that's cool all right let's get let's get another one of our patreon patrons okay go ahead go ahead he is the delightful name of craig woolhouse and he is from new zealand where he's proudly uh flagging up the fact that they stopped covid congratulations yes um if a game of basketball it helps when you're a tiny island and you don't let anybody in okay go on don't tell them that they won't listen they're very they're a very proud nation right he says if a game of basketball is held on mars indoors with earth's atmospheric pressure would we finally be able to dunk it from the three-point line and would it count as a three or do we need another planet oh i'd like that and we will get we will get to that answer when we come back on star talk sports edition we're back star talk sports edition cosmic queries we started out with march madness and now we're thinking about sports on other planet planets what role ai could play in predicting winners and we've got matt ginsberg with us becoming a friend of star talk so it's not your first rodeo with us thanks for coming back matt so we're picking up on a question i love this question if you could restate that of course it's from craig woolhouse he's one of our patreon patrons an exclusive member and he says if a game of basketball was held on mars indoors with earth's atmospheric pressure would we finally be able to dunk it from the three-point line and would it count as a three or d we need another planet and then he science the one that mj that's michael jordan came from so uh hmm your answers please so matt if you could if you have anything to add to what i say i'd be delighted for you to sort of jump in but on mars there's about 40 percent of earth's gravity there so if you weigh you know 100 pounds on earth you weigh 40 pounds on mars and all of your musculature is accommodating the hundred pounds that you weigh so now you only have to sort of move 40 pounds sort of up against gravity so you can jump higher and but you cannot so so you can jump higher and you fall more slowly oh okay that's the important part so the hang time yeah is there okay so if you can jump higher and you have good hang time and if you get a good running start i'm thinking i didn't run all the equations on this but i'm thinking you could dunk from the three-point line and count it as a three-pointer because you would not have touched the ground and the ball wouldn't have hit the ground in between the three-point shot line and and the main basket matt what do you think of that i think that's right i mean certainly from a rules perspective that's the easy part of the question um you know you take off from behind the three-point line it's a three-pointer i think that whether or not you can do it um i haven't done the calculation either i suspect that um a second is still a second and a meter is actually what's reduced by 40 percent so you probably can jump like two and a half times as far and given that michael jordan can dunk from behind the three-point line anyway no no no no no no no no no no no [Laughter] what he's actually doing which is not obvious unless you analyze it so generally if you're trying to dunk the point where you're dunking is the highest part of your arc because the rim is 10 feet up whereas michael jordan from the free throw throw line he is not still ascending at the point he's dunking he has already peaked in his parabolic arc and he's on his way down so he had to jump that high in order to make all of that happen so if you watch his arc he's on his way down so it doesn't have to be sort of the limit of where you're jumping provided you got up high enough you could just descend into the dunk okay i i want to sound like the smart kids in this conversation so i didn't do the calculations either however i know if we're doing this with this sort of ability i'm making the court bigger and the rim taller the rim higher yeah i'm raising the bar i'm i'm stretching the court and yeah because if i if the hang time is seconds plural then it's a different game if we play it on this people will be flying out the arena that's a better game yeah so wait wait but if you do that then you're neutralizing all these interesting features of the martian basketball court right well but then again it's it's more like basketball because we've kind of made it equivalent in the size of the core as opposed to an earth court i know i think it's it's way more complicated than that right are you going to make the hoop bigger if you're shooting from so much further away you have to make the hoop bigger but then the inside game becomes tremendously different so i think it's act i don't think you can you can't rebalance it it's going to be a different game all right so here's a very good point right just to be clear just again to add emphasis to matt's point you shoot the ball at the basket and there's a certain margin of error in angle outside of which you're not going to make the basket right and that is true in any gravity right so so that you're not helped in a lower gravity by this sort of margin of error angle so if you're going to shoot from twice as far away then you're going to make half as many baskets because that if the angle will no longer accommodate the distance over which the ball is veering off of course unless you're steph curry but that's right but i will i want to i want to just litigate steph curry on mars that's a new movie we got to do that i just want to litigate one quick point for you guys to figure out okay if shooting a ball from behind the three-point line is what entitles you to the extra point why would dunking the ball from behind the three-point line still gets you that same extra point as it's your feet well no the shot actually happens at the rim okay no so here it is a guy standing at the three-point line leaning forward right so his hands are inside the three-point arc and shooting still gets to three points not only that you can jump from behind the three-point line land inside the three-point line and it counts as a three point okay so i'm all about your feet but it's all about the feet it has nothing to do with the actual shot it's just the exit correct all right gotcha yeah so technically it works it reminds me of the the movie which was it a stupid movie but it was entertaining where they invented the ali-oop in the movie and in the alley-oop it's like is that legal you know the first alley-oop you got to look at it and say did anything happen illegal there i can't think it should be illegal but apparently it's not and we kept it all right there you go all right cool all right another question this is another one of our patreon patrons james senior he says uh a question about ai when do you think we will actually have ai in a sense of an actual artificial consciousness also how would this be achieved would it come from an algorithm or from actually uploading the human consciousness into a computer or by other means so we've dropped the ball timeout uh half time whatever you want to call it and we're now thinking about ai matt over to you so um there's some there's some assumptions underlying that question that i don't know if they're right um so i i think getting uploaded into a computer if if that happens to me before i i uh before my i guess my warranty is up um that would be fine with me but i don't think that counts as an ai that's just matt inside a computer somewhere so the question of of what it is to be intelligent actually becomes important and interesting historically the definition of intelligence was able to pass the touring test which is something invented by british mathematician alan turing and it basically says if i'm on one end you're typing into a computer and an entity is typing back and if you can't tell if that entity is a machine or a person the machine's intelligent okay if you get back to what i said earlier about the 49 versus the 99 problems the machines that look sort of quote intelligent unquote are going to be the ones that solve the 49 problems they're going to do things we can't do they're going to solve problems we can't solve but they're not going to look like us they're not going to pass the touring test they're going to i don't know how they're going to deal with with traffic lights but it's scary they're and but they're going to be great at trading stocks they're going to be great at predicting who's going to win sporting events they're going to be great at predicting the weather they're going to do all sorts of things and help us and they're going to have something that we i think will come to think of as intelligence that there's not going to be a moment where all of a sudden they go from not intelligent to intelligent we're already seeing that as machines do predict more and more i think you're also implying that we should not hold consciousness as the metric of whether the thing is intelligent or useful or can get the job done that's right because i don't know what consciousness is i mean are we waiting for a machine to say hey leni alone i haven't had my coffee that's not that's never going to happen why would we build machines like that why would we build machines that are grumpy and machines that need coffee well the fantasy always is that the machine will come to this state of being on its own through some evolutionary process it will achieve consciousness all right it will so it it wasn't designed that way it becomes that way through so many experiences that it is able to decipher for itself that it indeed is sentient and conscious so the the number of times that things happen in the movies is not a tremendous indicator of how frequently they will happen in real life and i don't i don't chuck he just did that was just a disc chuck in case you know i never said it came from a movie i said it was allow me to just say fantasy that was a diss okay so polite one at that i don't think it's gonna be like terminator i think we're going to find that these machines are our partners we can do things we can solve problems they can't they can solve problems we can't and we will collectively do more than either of us could do individually and i think and that's now that's today and tomorrow as far as will they eventually so completely surpass us that we become unnecessary in some way i don't know maybe in some far-off land but i imagine that far off time i imagine that the team the man machine team is going to be so much better and it will grow we will grow together we will work together to always do things that we can't do individually i'm incredibly optimistic about this i think although you did say you did say just to let the record show that the future of machines probably won't be a terminator that's i don't know how encouraging that is because any terminator at all would be bad yes yes and i i think that you know could we we could get there if we worked at it but that would require an enormous amount of stupidity by a relatively large number of people and that's why it's just probably never underestimated exactly because never underestimate yeah humanity's never proved that they can do that ever i'm i'm just wondering matt if if machines are constantly learning that that part where they can't solve the [Music] 100 problem yeah very well there'll be a point where they can and then are we not redundant i don't know that there will be a point where they can they really do they're different so they're architecturally different right so we have a trillion neurons operating on millisecond time scales yeah machines have even these massively parallel machines have thousands of processors operating on nanosecond time squares scales we're different architectures we should be good at solving different kinds of problems and i you know could you eventually simulate a human brain and computer and make it all sort of the same but then why would you is the point yeah i love it i love this angle on it there's so much there you know it's like it's like somebody your your parents for your 18th birthday buy you a porsche and you say i only want to use it to drive up the driveway and get the mail nobody would do that you're going to use the porsche as a porsche we're going to use these these computers that have abilities we lack in the areas where we need help just learned that i've been using my porsche wrong all these years but you got your mail fast there's the early internet let me get to my mailbox as fast as i can we got time for one last question if it has a quick answer all right i think i think we've got to go back to the factor man and god's algorithm matt ginsberg's book didn't you want to know something yeah i just i just tell me what happens what that's a novel right yeah what is god's equation um it's about a guy who finds what's called god's algorithm it lets him solve basically any problem and my view is that he's anybody who finds this is it's a race whether the government kills him or he takes over the world first and this guy realizes he realizes he's in this race he doesn't actually want to take over the world he he mostly just wants to go to disneyland with his kids and it's about his desire to make the his his attempted journey to make the world a better place before this technology is used to to mess everything up so this is a trailer it's supposed to be fun this is a this is a um do you have confidence that such a an equation exists such an algorithm exists so there's this is the biggest open question in computer science is whether such an algorithm exists and i believe it does um confidence is probably a little bit too strong because i'm in a pretty small minority um people occasionally measure it and i think i think something like 10 of the serious computer scientists believe it something like that and that would give you access you'd be able to tap future knowledge of systems that would be without precedent in the history of civilization that would make you all powerful um then you can move out of your parents basement where you are right now you say this is you kneel yes i can i can go get the battery back out of quarantine you can get the button up get the boat i love that get the butler out of the corner all right we got to call it quits there matt ginsberg great to have you back obviously star talk sports edition thank you and there's more to plumb in your uh in your expertise on these topics and we'll surely come back to you on this thank you okay great excellent and gary always good to have you man my friend all right chuck chucky baby love you there all right i'm neil degrasse tyson your personal astrophysicist bidding you farewell from star talk sports edition as always [Music]

Channel: StarTalk

Views: 122,671

Keywords: startalk, star talk, startalk radio, neil degrasse tyson, neil tyson, science, space, astrophysics, astronomy, podcast, space podcast, science podcast, astronomy podcast, niel degrasse tyson, physics

Id: GM2mO04vcts

Length: 51min 5sec (3065 seconds)

Published: Sat Mar 13 2021