Where Will Artificial Intelligence Take us? - Kai-Fu Lee
Вставка
- Опубліковано 6 лип 2018
- Kai-Fu Lee (李开复) is a Taiwanese venture capitalist, technology executive, writer, and computer scientist. He is currently based in Beijing, China.
Lee developed the world's first speaker-independent, continuous speech recognition system as his Ph.D. thesis at Carnegie Mellon. He later worked as an executive, first at Apple, then SGI, Microsoft, and then Google.
He became the focus of a 2005 legal dispute between Google and Microsoft, his former employer, due to a one-year non-compete agreement that he signed with Microsoft in 2000 when he became its corporate vice president of interactive services. - June 2018 - Наука та технологія
I feel like he is a good example of a person who is an expert in one field while being unaware of the depth of his own ignorance in other fields. Honestly we're all like this to some varying degree and the information explosion is really bringing this to the forefront more often.
But, I think on the subject of his predictions about the future of AI, and geopolitics he sounds painfully naive, though it's obvious that he means well.
unlike elon musk, mr lee has a PhD degree in AI. period
Sean Kincaid it sounds like most political discussions
I love that he has humility in understanding that deep learning is not how the brain works. We wont crack the brain code without some out of the box thinking and lots of time.
"A.I. can only do one thing really well" is not necessarily true. For example, a lot of applications use multiple A.I. neural nets that are each good at a specific thing. Take a speech-to-speech translation A.I., for example, which requires speech recognition, natural language understanding, and also how to generate a human voice. Also there is an A.I. that is general enough to virtually play and be superhuman at any Atari game. If you look at OpenAI's Dota 2 bot, the A.I. are doing a crazy number of different things. They are able to learn many non-intuitive strategies that pros use and also they are able to cooperate and beat human teams. Also, why would you want to shut down talk of A.I. utopia or dystopia? If nothing happens, nothing happens but if Superintelligence A.I. is indeed imminent then it poses an existential risk and we must talk about it or risk extinction.
Also IBM Debater is an A.I. that people can literally debate ethics or philosophy with and a lot of the times it presents compelling arguments far better than many humans could. I think Kai-Fu is just biased from his experience with A.I. in the past, even if some of the A.I. limitations he mentions are true, there is no reason we won't be to solve it the short term and would instead have like 20 year A.I. winter. We already know A.I. can surpass humans and the value of that proposition is too great for us to get discouraged or stop pursuing A.I. at this point.
This video reminds me of Ray Kursweil discussing the mapping of the Genome and how critics said would take decades to finish. He said " I reminded them that it takes 7 doubling s to reach 100% and that we were actually right on track. I had heard others talk doom and gloom over jobs till again Kursweil pointed out that we are heading in to an age of abundance and it wont be necessary to have people working 40 hours and we will restructure to keep people more happy than ever before. I tend to lien to Ray who has been right an incredible 80 plus percentage of the time.
Even people with a lot of ai experience don't know all the work being done with ai. He has valuable input but there are too many people all over the world pushing ai progress in new ways.
Most balanced, well-informed and sane talk on AI/ML technology evolution.
Excellent interview
Fantastic !!!!!!!!!!!!!!!!!!
Thank you for sharing your distinguished point of view! :-)
One of the better talks on AI, no hype , just calm explanation of where we are right now.
I like listening to this man. He seems very logical In his thoughts. People here commenting like they have more knowledge and experience than this man who has worked in this field for over 30 years really should put more thought into their comments before they write.
For decades experts were 100% convinced that reusable rockets are not possible. Like, physically not possible. And some of them didn't even saw it coming 3 years ahead.
Long experience can hide things from you.
Sean Kincaid I think there is some confusion. My comment is in reference Dr Kai Fu Lees knowledge and expertise in the AI field compared to comments in this section from people without any credentials or experience in the field of AI (well least they have not stated they have x years of experience working in the sector). Your comment is in relation to dr Kai Fu Lee and others working in the same field and they are very different comparisons.
exactly. lots of people commenting seem kind of pissed off, that he's telling them that their science fiction story future isnt going to happen by 2045 or whatever it is.
I cannot understand these arguments against the singularity coming within the next 2 to 5 decades. I can't even imagine it'll take that much longer, certainly not 100-1000 as he mentioned
He probably has some personal bias against being open to the possibility that machines can be as or even more self aware and intelligent than humans.
I agree with you guys. There will be a point in History where A.I. will outsmart Biological Intelligence. And there are some real Chances this could happen within this very Century. Some Elites choose not to talk about that so People will not give up on everything. Steven Pinker said exactly that before the EU Parliament. This is Intentional Ignorance regarding the Future. This is Part of their Tactics to keep us going so the Elites can accumulate as much Power (Data) as possible before its to late because other Companies are too far advanced in their (A.I) Developement.
It certainly will. If you think about it, AI is technically already smarter than humans. It can do calculations that are physically impossible for humans to do. All that is left to do is replicate the algorithms of human intelligence.
Earliest 2030! Artificial intelligence will be very powerful by then they have project debater and google duplex now ai doubles every 2 years so yeah
Latest 2045
Kai Fu Lee has very useful insights in his brain.
Wonderful points made! Now you will understand the need for built-in robot arms that are part of the kitchen sink and all you do is put dishes in sink and the arms come out and do the rest. Every home having this will create many new markets and so much relief worldwide for dishwashers At Home
Like a washing machine? No way 😉
Guillermo Casas ...Yes way! Thank you..so many people will be soo much happier! Think of the artful models for this type of contraption..The Arms can look reeeal! Water shoots out of the hands that can pick up scrubbers and way fast clean n rinse and stack. The arms also scoop off compost food and knows how to sort plastics for recycled melting into plastic bricks ala Trashmasher Tek. Optional of course.
freeing us up to do what we really love.
what do we really love,
i guess would be the appropriate question.
perfect english!
Perfek engrish.
What were you expecting? He emigrated to the US when he was ~10. The American educational system may be terrible, but it's not that terrible.
There is a cut at the end, or was that purposefully edited?
good history on AI and its course of development in beginning of talk
Very thoughtful and practical discussion on current AI tech. I hesitate the disagree with his extensive experience in the field, but I think our lower level AI, as concentrated as it is, will aid us in finally creating general AI. First in freeing up so many more human minds to tackle the problem, and sexondly in running simulations on new neural architecture to build better AI. Maybe general AI is still a hundred years away, but a 1,000 that seems like an eternity compared to our ever increasing rate of tech evolution.
niceee
Good talk but I find one thing very troubling.
1. He arbitrarily labels some jobs or tasks as repetitive, uncreative, and thus make us less human and ending these jobs would free up people because essentially "we're here to create." He lists scientists, filmmakers, and marketing as examples of human creativity but oddly leaves off Go and Chess which for centuries were thought of as the pinnacle of human intellect and creativity in their respective cultures. Whats considered creative and meaningful today can become repetitive and mundane tomorrow.
I actualy think that current algotithms are strong enough to create AGI. We just have to optimize for creating learing algorithms. In other words, we need to create neural network that can make another learining algorithms and then optimize that neural network to get the best learners. It will be something similar to compilers or compiler compilers. specialized programs that whole purpose is to create better learners. It could work if knowledge how to improve learner is not more "complicated" than the task itself, which usually is the case. In the same way as creators of AlphaGo are not able to beat go proffesionals, but still they could create something better at playing go.
I actually think if we have expressive enough model and correct amount and type of data, it will converge to human intelligence. Human will not create AGI, it will be machines that are optimized for creating AGI or even it will be machine that are optimized to create machines that are optimized for creating AGI. Human mind is to complex, so we will essentialy brute force models of human mind.
there is still much todo in the AI community. Sofar, neural networks do a lot. But there is far more to the brain than a simple neural network of neurons and synapses. For example, take RNA, it appears to encode memories. a NN without something for RNA is empty or dead. When thinking about things, the brain works by thinking about what is most important at any given point in time.
right I did neural networks and what my university called 'AI' back in 2001, as far as I'm aware nothing significant has changed in the field except like he said faster computers and bigger sets of training data, the different types of neural networks people use now, are more like optimisations of existing ideas, they aren't based on new ideas about how the mind works.
Only at min 17 but I have to comment on that a few things.
"Some say we have to be careful because they could take over the world"
This is not what the critics say. At least not those that you should address. They say AI could be dangerous because we loose control over it. We lost control over vastly more simple tech, like petrol or plastic. AI can be humanities doom without being grumpy at us. As stated: it's basically an optimizer.
"Needs huge amount of data" I hear that time and time again. Not convinced. Why should Alpha Zero not happen in other domains? Something like alpha zero was probably deemed impossible one decade ago ... by experts. Same with reusable rockets.
Also I saw a paper about learning in a self build world model aka dreaming.
"Very narrow. No learning in multiple domains."
I also saw a paper about transfer learning. Where knowledge from a existing neural net was used to speed up the learning process on a different task.
"Some people extrapolate" yes they do.
"this will not happen this way" IS an extrapolation.
"What we see is just growth in application, not exponential breakthroughs. There was no real breakthrough since what those 3 people did back than"
1. Probably also depends on what is considered breakthrough.
2. Three people. There it is. We not only see growth in application but also in research. And not only in rich countries btw. Did anyone notice that India and China have a space industry and active nuclear power research? Please take a look at gapminder. Global research capabilities are increasing ... Not because of Europe or the US btw.
When someone tells me it makes no difference when 100 people actively research or 100000 ... I get a little suspicious. :\
Little disclaimer: I know I did not cite him word by word... would be a pain to do on a phone.
Yes, I do not work in the field. I don't say I know stuff better than him. But I noticed that he was contradicting himself at times and that I know of research that indicates that things like multi domain learning might in fact be possible. This narrative of "only big data, only narrow" is just less true than 5 years ago.
It's also not only about compute power. Alpha zero required less time and resources than alpha go and still was able to beat it.
Alpha zero could have been done years ago. We didn't lack the compute power but the knowledge. So saying it's basically the same stuff than 30 years ago seems to ignore a few things.
But again, what really got me writing this comment was the "robots won't take over the world and kill us all" part. It's just so easy to to argue against rediculous robots.
How about this: AI is used to build a reeeally good climate model. To counteract climate change we throw some geo engineering at it. Works in theory. We try it. Doesn't work because it interacted with something that wasn't properly modeled because it was unimportant originally.
I am sure this scenario is too obvious at it will happen because of something noone thought of ... Or because of something that was ignored by those that have to make the decision. THIS is what people warn about!
Edit: Now, this probably sounds like I really hate him, but I don't. I really enjoyed this talk. He just ignored stuff that I see ignored to often, so I had to comment to not forget what I wanted to say. Got a little bit carried away tbh. Now have to finish the video. Maybe he talks on those issues. Consider me an idiot than. :)
Like basically everything he said afterwards. :) Well, can agree on describes it better, I guess.
Really loved: AI shows us what not makes us human.
"Needs huge amount of data"
I do think the experts needed it to learn the new things that followed. I doubt that without ImageNet we'd have gotten as far with image recognition for machine learning. And from that a lot of new developments followed.
Which AI companies are good to invest in?
Ai effect
Create your own. Otherwise, maybe look out for companies with values you could identify with. Firms aligned with some #SDG values at heart, could be a anchor to start from as well.
When evaluating existential risks associated with so-called AI, one must have an understanding of the three basic factors involved, otherwise, all utterances are no more than verbiage.
The first factor I refer to is the kind of agency "AI" is likely to emerge with, of which three types are possible.
Factor two, only one of those types will be able to provide, not guaranteed, survival but the mere possibility thereof.
And three is that the percentage of the global population able to access the means to survive will be approximately .000,000,000x%.
Those not in the aforementioned percentile will die.
We will all laugh so much at this guy in 2030...
Very very very smart man
Ohh If They let Us Create...
ALL WE NEED IS DRONES PLANTING FUCKING TREES AUTONOMOUSLY FOREVER
4329 Petawatt were used on very effective technology to compute this,
I don't think Geordi Rose believes this fellow.
since god isn't in the business of talking to us,
we will make a god,
that will.
it never ends
i like local minima, they feel so cozy if you use feature scaling :D
God talks to you all the time; try listening.
that was, god's message to , me.
Hehe thats the feeling whether god or not we must progress and progress we shall! AI is inevitable thats just the way it is something real!
Interesting that at one point in this excellent talk he says AI is not currently a direct threat to mankind - but the indirect threat of narrow AI is very real and far greater particularly in the short term.
Mass job losses starvation and mass human migration are very serious threats to global stability. The human race continues breeding at exponential rates and no country has as yet managed to put a cap on birth rates that has successfully reduced the population to sustainable levels.
China came the closest with the one child per family restriction - but they are still the second largest population in the world and will soon have to deal with massive unemployment.
The EU is on the forefront of a human catastrophe -mass human migration from Africa and the middle East - who will clothe, house and feed the hundreds of millions of refugees that are soon to coming knocking loudly on their doors?
Not sure what planet you live on - but I challenge you to name one country who has laws that govern population growth (other than those in China). Do some research into the word exponential. One can barely call the increase in refugees in the EU over the past 5 years - 'mass' human migration. Lets have this conversation in 10 years. The whole issue here is narrow AI has no concept of empathy or other human qualities - they are simply systems deployed to maximize corporate productivity and profits. The fallout as eloquently explained by Mr Lee - will be job losses and an ever widening division in global society - the haves and have nots. It is our job to be responsible when deploying these AI system so that it minimizes the human impact.
I was not talking about legal laws. Population growth just slows down at some point. Maybe due to better education and changing lifestyle, I dont know if anyone knows for sure. But many countries in europe have not experienced any significant pop growth in the last decade. I don't see why people in europe should be so special that this only applies to them.
Mass migration: definetly not my word but what happend in the last few years is already frequently called mass migration. Maybe I am so used to it that automatically thought you would belong to this group.
I agree with you that AI has to be deployed carefully and the widening social gap is nothing to ignore. Technically AI should make it easier than ever to feed all people on this planet. Nevermind some population growth. We probably already have the technical and economical capability to do so.
If we don't care and mass migration is the result, its our fault, we have to live with it (and that is without closing borders).
I'm from planet earth btw. ^^
Solar System. Milkyway. Local group. Pls don't ask for coordinates, they change all the time. :/
Don't know if this link will get filtered. But here is a graph that shows the population growth of all african countries. Sadly I don't see a button to average them, but if you look from far away you can see that the growth is already slowing down in africa for about 10 years. This kind of information just takes a while to get noticed by media.
www.gapminder.org/tools/#$state$time$value=2016&delay:100;&entities$show$geo$;&world_4region$/$in@=africa;;;;&marker$axis_y$which=population_growth_annual_percent&scaleType=linear&spaceRef:null;;;&chart-type=linechart
If total population would be exponential, this graph shouldn't even think about decreasing. (Because I do know what exponential means ;) )
If you look at Asia, it is even more obvious:
www.gapminder.org/tools/#$state$time$value=2016&delay:100;&entities$show$geo$;&world_4region$/$in@=asia;;;;&marker$axis_y$which=population_growth_annual_percent&scaleType=linear&spaceRef:null;;;&chart-type=linechart
More specific example: India
www.gapminder.org/tools/#$state$time$value=2016&delay:100;&entities$show$geo$;&/$and@$geo$/$in@=ind;;;&$world_4region$/$in@=asia;;;;;;&marker$axis_y$which=population_total&scaleType=linear&spaceRef:null;;;&chart-type=linechart
Total population looks pretty exponential, right? But if you look at the growth:
www.gapminder.org/tools/#$state$time$value=2016&delay:100;&entities$show$geo$;&/$and@$geo$/$in@=ind;;;&$world_4region$/$in@=asia;;;;;;&marker$axis_y$which=population_growth_annual_percent&scaleType=linear&spaceRef:null;;;&chart-type=linechart
The growth stopped increasing 1982. Humans are just very bad at seeing trends of steep slopes. It is slowing down for about 36 years already! 1% decrease in 27 years. 1.1% to go. After that, the population will no longer increase. Not more than in every developed country at least.
Population growth is not exponential.
That's the point - narrow AI - targets the jobs done by relatively unskilled labor. Look at India and Africa, South America and large parts of China and Russia. Many of these people will lose their jobs within the next 10 years. How will they survive? The world is progressively moving away from fossil fuel dependency - how will middle Eastern countries survive as there economies go into decline? That's apart from the significant increases in average temperature in those regions and N Africa. 50C temperatures are regularly being recorded there - that is already driving relocation towards the EU and the UK. Where else will these people move to?
As far as feeding 7.3 Billion mouths on a daily basis is concerned - the human race has to become more responsible - we cannot continue to breed as current global rates - it's simply not sustainable. Another contributing factor is the increase in human life span. Consider this - the human race is the only species on the planet that does not live in harmony with nature - we naively seem to believe that our impact on the global ecosystem can continue unchecked without consequence. All these issues can only be exacerbated by AI systems that target the poor / unskilled labor force - searching for ever increased efficiencies and profits.
#timephonehack a.i. inventions
Glad I started listening to Kai Fu Lee. Makes Kurzweil look like a fool. Human-level intelligence by 2029 ? Yeah right.
this good sir doesent read the deepmind papers
Will lead us right into slavery/extinction
If we haven't successfully eliminated any potential Extinction-events from the equation,
but rather succeeded in adding an entirely new set of challenges to the Mix...
*!!!We haven't improved the Situation!!!*
Slavery? The promise of AI is the labors of the poorly educated masses are no longer required. Why would they be enslaved if they serve no useful function, their continued existence is undesirable.
Dukky Drake very true skip straight to extinction 👍🏾
39:36 - New World Order
This guy is just interested in making lots of money in the short term. Real sad.
Good at business. Horrible at science.
bad idea AI...we are not mature enough to handle this technology
why are you said that???
did you watch the video?
And how to mature?
That question got me when i was not a child anymore but not yet a (wo)man.
Nancy, your protest is about 60 years too late. LOL! Have you been living under a rock the whole time?
LOL! I LOVE TROLLS THEY LET YOU KNOW WHEN YOU ARE DOING GOOD