Skip to main content

tv   Bloomberg Best  Bloomberg  May 25, 2018 10:00pm-11:00pm EDT

10:00 pm
>> i am reminded of all the beauty, innocence and gun free fun available from our neighbors in the north. ♪ as followed breaks out in canada canada.ll breaks out in there is the majesty of toronto. and gallons of maple syrup that you can chug openly and feel free. for this open syrup is pure and nourishing -- this maple syrup
10:01 pm
is pure and nourishing. changes on the season are a perfect time to encounter one of canada's most creatures. the artificial intelligence nerd. ♪ not too long ago, these beings were rare and hidden away in university dachas. today, they are among the best paid professionals in the world. ♪ creatures dide something truly remarkable without anyone paying much notice. that gave birth to an ai revolution. canada,ned canada, yes, into one of the greatest ai superpowers. this is the story of how all of this came to be. questory of one nation's
10:02 pm
to teach computers to think like humans. the story of what this says experiment will mean for all of our lives, and for the future of the human species. ♪ human, or someone trying to imitate one, he will want to pay attention. ♪ ashlee: ever since people first came up with the idea of computers, they have dreamed of imbuing them with ai. , as im a smart fellow have a very fine brain. is able a computer that
10:03 pm
to mimic or simulate human thought or a beer. within that, there is a subset called machine learning, that is now the underpinning of what is most exciting about ai. ashlee: by allowing computers to learn how to solve problems on their own, machine learning has made a series of breakthroughs that was seemed nearly impossible. it is a way that computers. can it now understand your voice, spot of friends face in the photo, or steer a carlos car. the reason why humans are talking about the coming of ai. whether there will be a wonderful thing or a coming end of days thing. many people who made this moment possible. but one figure towers above the rest. i have come to the university of toronto, to see the man they call the godfather of modern artificial intelligence, jeff
10:04 pm
hinton. because of a back condition, he has not been able to sit down for more than 12 years. would muchtanding, i rather sit down, but i have a difficulty when i sit down. ashlee: at least standing desks are fashionable now. guest: i was standing when they weren't fashionable. [laughter] ashlee if you can't sit in the car or a bus, he walks everywhere. . a lot about al him and his results. he has been trying to get computers to learn as people do. a class that many thought was crazy or hopeless, right up until the moment it revolutionized the field.
10:05 pm
amazon thinks it is the future of the company, my own department thinks it is probably nonsense, and we shouldn't be doing any more of it. [laughter] so i told everybody, except my own department. [laughter] >> pretty early on, he became obsessed with the idea of figuring out how the mind works he started off getting into physiology, the anatomy of how the brain works, then he got into psychology. finally, he settled on a computer science approach to modeling the brain and got into artificial intelligence. >> my feeling is that if you want to understand a really come pick it a device like a brain, you should build one. you could look at cars and think that you understand them, but each time you build one, you suddenly discover that there is
10:06 pm
some stuff that needs to go under the hood, otherwise it doesn't work. ashlee: as he was taking about his ideas, he got inspired by ai researchers across the pond. specifically, this guy. frank rosenblatt. >> in the late 1950's, he developed what he called a percentron. a computing system that would mimic the brain, a numeral network. >> the basic idea is a collection of small units called euros. they are little computing units modeled on the way that the human brain thinks. they take the data and actually learned so that the neneuron can learn. ashlee: his hope was that you feed a numeral network a bunch of data, like the faces of men and women, and he would
10:07 pm
learn how to tell them apart. like humans do. but there was just one problem. it did not work very well. >> his numeral network was the urons. layer of nen and it was limited in what i could do, extremely limited -- neurons. his college wrote a book in the 1950's which showed the limitations and that the whole area of research, into a deep freeze. for a good 10 years, nobody wanted to work in this area. they were sure it would never work. ashlee: well, almost no one. geoff: it was just obvious to me that it was the right way to go. the brain is a big numeral network. so if stuff like this, it has to work. alssion is a big neur network. ashlee: what kept you wanting to pursue this when everyone else was giving up? geoff: i need that everybody
10:08 pm
else was wrong. ashlee ok! [laughter] hinton decides that he has an idea of how these neural nets might work, and he will pursue it, no matter what. he's bouncing from research institutions in the u.s. around, and he gets fed up that most of them are funded by the defense department. he starts looking for somewhere else he can go. he suddenly hears that canada might be interested in funding artificial intelligence. geoff: that was very attractive. i could go in just get on with it. so i came to the university of toronto. in the mid-80's, we discovered how to make more complicated numeral nets so that we could solve those problems much more simply. >> he and his collaborators created a multilayered neural
10:09 pm
network, and it started to work. then they hit a ceiling. early: in the 1990's and 2000, he was one of a handful of people and the planet still technology.s he would show up at academic conferences and be banished to the back rooms. treated really, like a pariah. was consumed and would not stop. he pursuing the idea that humans can learn, until about 2006, when the world catches up to 's ideas. ck ou geoff: now, it is behaving like i thought it would behave in the mid-1980's. solving everything. of superfastrrival chips and the massive amounts of data produced on the internet, gave his algorithm a magical boost. suddenly, computers can identify what was in an image.
10:10 pm
then, they could recognize speech and translate from one language to another. nets12, words like neural and machine learning were popping up on the first page of the new york times. >> next up, we have professor jeffrey hinton of the university hinton, of--geoffrey the university of toronto. [applause] >> thank you. ♪ ashlee: this is obviously a really redemptive moment. now, he is basically a technology celebrity. sor canada, it is a country' moment as well. they have more ai researchers that just about anywhere else in the planet. the question now is to see what they can do, starting companies and pushing that technology forward. canada to seess
10:11 pm
the best in canadian ai technology and get a feel for how far the technology has come and how far it still has to go. ♪ >>
10:12 pm
10:13 pm
10:14 pm
10:15 pm
♪ ashlee: here is a city that get right at the central tension of modern life and the unfolding ai revolution. it is montreal. a place filled with beauty and old world charms. it asks you to move slowly through its streets and chill for a while, reflect and think deep thoughts. ♪
10:16 pm
at the same time, it is one of the world top ai research centers. people flock here from all over the globe to get deep with intone learning and take his ideas, and figure out how to turn them into products we all use. to see just how successful they have been, look no further than your pocket. >> all this stuff started out as hard-core computer science, but over the last five years, ai has invaded our everyday lives. your smart phone is packed full of ai-powered apps including something like google translate. it lets you plug your phone in a magazine that is written in french and read it like a local. they have been trying to get computers to translate text like this for decades. but it was neural nets that made this possible.
10:17 pm
geoff., jef and it is not just a cell phone. they are headed out on the roads. meet my friend, stephan. the head of montreal's tesla fun club. >> i have been driving one for about a year and a half. ashlee: do have people asking you for rides all the time? >> yes. ashlee: maybe it is because of his fancy pants autopilot. tesla's semi-autonomous driving system that kicks in when driving conditions or right. that it, autopilot is on? >> yes, it is driving by itself. we need to pay attention, we don't have to drive. crazyt'sts craz [laughter]
10:18 pm
self-driving car's are full of technology that lets the cars build a picture of the world. the technology has a long way to go, but this tesla can monitor all the cars around it, switch lanes and park, all by itself. thanks, geoff. so you are living in the future? >> yes. when you try it once, it is very difficult to do without it. i can relax, and just hang out. that is why we still need to pay attention. [laughter] ♪ ashlee: a big part of hinton's legacy lies beyond these examples of ai in the world. he is also inspired a region of disciples -- a legion of disciples spreading the good
10:19 pm
world of neural nets. this one is a professor at the university of montreal one other researchers and glam on to hinton's ideas when it seemed to make little sense to do so. over the years, he has formed a mind meld with hinton and have come together -- they have together come up with concepts of ai. >> it seems like this totally went from computer science and research, to we see it everywhere in our lives. ashlee: are you surprised in the last few years? thees, the rate at which progress and the industrial products have been coming out, is totally something we did not expect. ♪ even now, it is hard to predict where we are going, is it going
10:20 pm
to slow down? or are we going to continue with this exponential increase? yoshua it is thanks to that montreal is full of top notch ai graduate talent. this has brought tech giants like google and facebook to town, along with their ample checkbook's. ks. ashlee: it seems like if you are good at ai, you can mix 200 or $300,000 per year. it is crazy to see how much these guys get paid now. >> a million dollars is something quite common. ashlee: have you ever had a country offer you an incredible amount of money to come and set up a lab there? >> not a country, but companies, yeah. ♪ has rejectedosua the lucrative offers of
10:21 pm
nets.ive neural he remains committed to academic, which is a better fit for his philosophical approach to ai. ashlee: you have guys like elon musk and stephen harper game, people who painted these technologies as things that can run a mock and start doing things on their own. what do you feel when you hear people say things like that? guest: i am not concerned about technology running amok. that terminator scenario is not very credible. i feel that if we are able to machines that are ours smart -- that are as smart as us, they can also understand our values and moral systems, and act in a way that is good for us. i think there are real concerns, essentially the misuse of ai to influence people's mind. it is a really happening with political advertising. ashlee: right, like the stuff from facebook. guest: we should be careful about this and try to regulate its use, in cases where it is morally wrong. we should ban it and make it
10:22 pm
easy go. ♪ illegal.t eas ashlee: it is comforting that he has these concerns. but in university, in reality, what's left of it becomes messier. this tiny room is the home to a startup called liarbird. founded by yoshua's former students, it has build an app that can clone your voice. >> this is huge, it can make is say anything. ashlee: one of its founders is this guy. mexican ex-pat, jose. he taught me the art of the close. >> you record yourself for a few minutes of audio. ashlee: thousands of letters danced across the amateur author's screen. when you start to eat like this,
10:23 pm
something is a matter. you guys better with politics. i wonder ravel and came from. [laughter] -- i wonder where that one came from [laughter] ok, creating my digital voice now. it takes at least one minute. one minute? my god! >> yes, before, you would need to record yourself for at least eight hours. ashlee: test your voice. so now i get to type something? >> yes. ashlee: ok. magiche ai has worked its anyr i am done typing, words i put into the app can be played back in my digital voice. here's a crazy thing, even words and never actually said in the first place. artificial intelligence technology seems to be advancing very quickly. should we be afraid? ashlee eakin definitely hear my
10:24 pm
i canin their -- ashlee: definitely hear my voice. those were words that i did not a. it is flawless in being able to take any word and just repeat it. >> hello world is the best show i've ever seen. [laughter] ashlee: this technology seems sweet, but lends itself to all manner of trickery. i popped back to my hotel to test out the technology a bit. you can see some really obvious ways that this could be abused. this is fake donald trump talking. >> the united states is considering in addition to other options, stopping all within the country doing business with north korea. can picture, you somebody taking over your voice, and creating some mayhem in your
10:25 pm
personal life. to really put my computer voice. to the test, i will call my dear sweet mother, and see if she recognizes me. [phone ringing] ♪ >> hey, mom. what are you up to today? did not have we any electricity early this morning, we're just hanging around the house. >> i'm just finishing up working, waiting for the boys to get home? >> ok. >> i think i'm coming down with a virus? >> oh, you feel bad? ashlee [laughter] i was laughing because you are talking to a computer. >> i thought i was talking to you! it's amazing. ashlee: is that scary? >> it could be scary if it was something really important. [laughter] you. sounds like ashlee: it does? >> yes.
10:26 pm
10:27 pm
10:28 pm
10:29 pm
>> the digital divide is splitting this country. we're trying to get kids off on too
10:30 pm
>> the real artificial intelligence weirdos in canada live here in edmonton. ♪ very, veryarge but cold and very, very flat city that is more or less in the middle of nowhere. it's the kind of place that has a giant butter vault to help people survive the lieeantean
10:31 pm
winter months. canadians like to put the best possible spin on how it brings out th the best in people, like this guy. >> edmonton is one of those cities that isn't automatically listed in the top cities in canada in terms of size or scale, but it has always had a really neat quality to it of that western, independent spirit you see very much in alberta in general combined with conscience and a thoughtfulness. >> over at the university of alberta, some of the most far out ai research in the world is taking place. the man i'm here to see is the university's very own ai godfather, rich sutton. he's considered one of the great revolutionary thinkers in ai. >> if you are not canadian -- i am, but not by birth.
10:32 pm
i was born in the u.s., now i'm just canadian. >> what brought you to canada? >> the politics. i wanted to get away from difficult times in the united states. it was invading other countries in 2003 when i came here, and i didn't care for all that. ♪ >> sutton entered the field in the mid-1980's, and he was a big believer in neural networks. but he has a different idea about how to further the technology. unlike the method of feeding no networks reams of data and telling them what to do, sutton wants them to learn more naturally, from experience, an approach called reinforcement learning. >> reinforcement learning is like what animals do and people do. they try several things, the ones that work best you keep doing. >> how do you teach a computer?
10:33 pm
-- the computer must have a sense of what is good and bad. they receive a special signal call the reward. reward.d a >> to see reinforcement learning in action, i followed one industrious young brazilian who has created an ai to play his video games for him. it plays the game thousands of times and gradually learns from experience how to do better. >> so the goal of this game is that you are a yellow block and you have to get as many potions as you can. >> and this is the ai going at this for the first time. >> the ai bumps into things. if it dies it is unhappy. ai starts to figure out that what i wanted to collect the potions. ai thatan look at the
10:34 pm
has ran for 5000 games. this is what it looks like. ♪ >> you can tell it is smarter about its strategy. >> yes. >> what happens if you run it 500,000 times? >> we get to superhuman level. >> so notching a high score is the note was pursued. learning has turned out to have all kinds of other applications. it's behind the algorithms. it recommends movies and tv shows on netflix and amazon. it beat the world champion go player, previously thought impossible for a computer. soon, it could read your brain waves and determine whether you have a mental disorder. but for sutton, all that is just the beginning. >> we're trying to make a real intelligence. we are trying to re-create human intelligence.
10:35 pm
easy is reinforcement learning as the path to what the futurists call the singularity, the moment when our ai creations rge past and su human level intelligence. >> do you have a date for the singularity? >> it's quite a broad probability. the median is 2040. equal chance of being before or after 2040. the rationale goes like this. by 2030, we have a hardware. it gives guys like me another 10 years to figure out the algorithms, the software to go with the hardware. it's going to be exciting, where we are going. 2040 seems like a long time to wait to meet a smart robot, but don't fret. over in the experimental wing of the university, there are coeds hard at work learning the line between human and machine. >> are you human?
10:36 pm
>> afraid not. but that shouldn't keep us from chatting. >> case in point, homegrown edmontononian genius corey mathewson. >> tell me about this guy. >> sure. so this is blueberry. on blueberry, i have deployed the improv system. >> yes, that's right. cory does improv comedy with a robot. >> i have been doing it longer than computing science. i've been doing it for 12 years, and i thought there's no more natural conversion than taking the state-of-the-art systems and putting them on stage. >> one day we will make it to the moon if this planet is not to be our last. the moon and the universe, the sun! the sun! >> it was like a ventriloquist,
10:37 pm
a new age -- a new, strange toy. the piece that is different is that i don't know what it will say. you, i downloaded a voice into your brain so you could perform in front of these people. i don't know what you're going to say. >> to give blueberry the power of surreal canadian improv, corrie made use of text that should be familiar by now. a neural network. he seems the network the dialogue from a bunch of movies. 102,000 to be exact. >> all the movies. every movie for 100 years. he canthat's just so learn language. >> that's right. he is kind of a language model.
10:38 pm
>> step two, uses reinforcement learning to train the network, rewarding it when it makes sense, punishing it when it spits out gibberish. time to put this want to be kindle to the test. >> start improvising. >> ok. campers, we are going to get ready for a real baseball game, grab your gloves and bats, let's get out there. especially you franklin,. >> i will not be much longer. >> ok, why aren't you ready? >> i did not have any more -- >> ok, come on, franklin. you know how i feel about you, but you have to keep your head in the game. put down the bat, franklin!
10:39 pm
i've got nothing to hide. this is all i am. obviously some of the responses are a little weird, but it's funny. as you are going along, it hit a couple things perfectly. and that's extra hilarious. >> blueberry may not be ready for its second city audition just yet, but cory has a higher purpose -- making ai relatable. there is a fear in society of ai. so we're kind of humanizing this ai, we are taking it down a peg, saying don't be afraid, look at how cute and naive that is. isn't there a flipside?
10:40 pm
you make it cute and people start to accept it? then we wake up and -- [laughter] >> i don't think that will happen in my time. >> to singularity may be near, or it may be not so near. but is the inhabitants of this oddly beautiful place keep pushing their technology, they just might create something alarmingly humanlike one-day. for rich sutton, it's not a question of whether we will get there, but whether we will be able to accept our mechanized brother. >> our society will be challenged, just like every time. are black people people? are women people? we will do the same thing with robots eventually. are they allowed to own property? are they allowed to earn income, or do they have to be owned? >> but a robot is obviously not a person.
10:41 pm
right? >> no. ♪
10:42 pm
10:43 pm
10:44 pm
10:45 pm
♪ >> our last on, i returned to toronto, home to 2.8 million people, one very tall tower, and of course the godfather himself. there are lots of little processes. but jeff be of import, hinton has done something truly exceptional for toronto. he has turned the city into an ai mecca, where conferences like this one seemed to take place daily. >> we are enormously thankful to canadians for inventing all this stuff.
10:46 pm
we use it throughout our entire business. owe canada.y the tech industry is full of people who adore ai. and also some famous types like elon musk and stephen hawking, who said that ai might be the end of us. to consider such dystopia in the toper light, i've come inonto's geekiest bar to case myself in the steel container with a writer for gizmodo and ai philosopher. the con case around ai? what's the nastiest scenario? >> unfortunately, there's no
10:47 pm
shortage of nasty scenarios. this is what makes artificial intelligence such a scary thing, all the different ways it can go wrong. it can be everything from an accident, where we didn't think it through -- we gave a very powerful computer instructions to do something. we thought we explained it articulately, and it completely took a different path than we thought it would in such a way that it caused great damage. i'm sure you've all heard the paperclip example. you are a paperclip manufacture, they say we need lots of paperclips, and because the ai has so much reach and power, it converts all the matter in molecules in the planet paperclips. scenario, but it is illustrative. we can be dismissive of the perils. that is exceptionally dangerous. early tohink it is too start raising the alarm bells about it. clippyg turned into
10:48 pm
sounds awful. but fear not. we'll have years to ease into that sort of suffering. steadily plucks off one job after another. course, willgo, of likely be the factory workers, which brings us to suzanne gilbert, a budding ai overlord and founder of robotics dart up kindred ai. >> to only about you guys. >> these are research prototypes, some of the first robots we built at cambridge. we tend to work with small robots. it's like if you imagine a a child growing up, it breaks a lot of things.
10:49 pm
now imagine the child is six feet tall and has the brain of a six-month-old, it would be dangerous. >> how many of these robots have slapped you? >> i have been hit in the face by a robot a couple times. [laughter] >> suzanne seems nice enough. exotic digital art, and she loves cats to the point where she built a robotic suite of them in the office. a quadruped robot loosely based on the cat, although it's not a very highly faithful replication yet. up, youyou were growing would build these as well? >> that's correct, yeah. so i was really enthralled by electronics at an early age. i guess mostly girls would be -- i was looking at
10:50 pm
resisters and components. >> but don't be fooled by the hobby electronics in the cute cat bots. suzanne is a keen businesswoman. and kindred has recently embarked on its first commercial venture. robots that bank of are learning. they are continuously running, picking up objects. all day, all night. ♪ network,d by a neural these arms can do something that is very easy for human, but very hard for a bot. pick up objects of different shapes and put them down. most factories still use people to do that sort of thing. lots and lots of people. shopping, piles
10:51 pm
and piles of different objects, shapes and textures. how do you pick it up? right now, it is humans. millions of humans picking up things and putting them in another location. so we are teaching robots how to do that. >> what is the hard part? what's the belt, what's the shirt? just how to grab it? >> exactly. things can be in any shape and you have to figure out how to pick it up without dropping it. it takes a lot of training. >> part of the training involves, of all things, humans. robot pilots who manually control the arm, while the ai watches and learns the final points of grabbing. grim about something the human. >> yeah, it's not good to take people's jobs away.
10:52 pm
but this kind of technology coming into the workforce should make a start inking about how we are going to pay people in the future, because ai is not just going to operate manual labor jobs. it will operate things like doctors and lawyers and accountants soon. there are going to be a lot of issues and disruption. >> suzanne is a realist, but she is also an optimist. in her vision of the future, robots won't be mindless competitors to humanity, they will be full-fledged citizens like the rest of us. robot, crazy, you have a working at a factory, maybe it gets paid a wage. it goes to buy a lithium-ion battery. body, having a physical they will have a lot of physical needs like we have. you might have to go to the repair shop to get a motor looked at, and they will have to
10:53 pm
pay to do it. they will be contravening to our economy in the same way we do. us,if they have brains like they will want to explore new things they've never seen before, they will want to learn rest so theirs mind has time to consolidate the information. >> trying to picture it in my head, robot workers that go home and sit on the couch and watch tv after work. >> i don't see why not. they could watch cap videos. >> it is hard to tell sometimes if she is laughing with us or at us. alone in hert cautious optimism for the future. >> i think there's always a sense that technology can be either used for good or used for bad. i'm reassured that canada is part of it, in terms of trying to set us on the right. whole, being responsible
10:54 pm
and thoughtful about the power we are gaining by research and learning is the right trendline, and i don't think ai's automatically doomed to some dystopian outcome. we are told that politicians will come up with policies that address massive job loss and prevent horrific inequality between the classes. we are told that these guys will take so long to become humanlike that we need not be afraid for a while. though, is that we are turning ourselves over to the unknown. so, you know, fingers crossed. >> eventually, i think we will become the ai. we will become the intelligent machines.
10:55 pm
we will understand how things can be smart, and we can deliver really create them. you might think of it as making a new generation, of new kinds of people. humanity is continuing to evolve, and why wouldn't enhanced people, or even designed people, be the next step in humanity? it's really hard to predict the future. i think there's going to be all sorts of things happening we didn't expect, but there's one thing we can predict. this technology is going to change everything. >> goodbye. >> goodbye. >> goodbye. >> goodbye. >> once i power you down, that's it. >> i'll never see you again. >> yeah. that's right.
10:56 pm
deep.t was [laughter] ♪
10:57 pm
10:58 pm
10:59 pm
11:00 pm
♪ emily: i'm emily chang in san francisco. coming up in the next hour, gdp are are now being enforced in europe. tech companies had months to prepare, but are they ready? plus the trump administration has reached agreement that will allow cte to remain in business. and, spacex completes its test mission this year. meantime, china launches a satellite that ushers in a new era and the global a spirit we


info Stream Only

Uploaded by TV Archive on