Wait but Why? The Superintelligence Road | Tim Urban | Talks at Google
HTML-код
- Опубликовано: 21 авг 2016
- Tim Urban has become one of the Internet’s most popular writers. With wry stick-figure illustrations and occasionally epic prose on everything from procrastination to artificial intelligence, Urban's blog, Wait But Why, has garnered millions of unique page views, thousands of patrons and famous fans like Elon Musk. Urban has previously written long form posts on The Road to Superintelligence, and his recent TED talk has more than 6 million views. Tim speaks to his blog Wait but Why and his writing on the impact of artificial intelligence.
This talk was presented for Google's Singularity Network and hosted by John Bracaglia. 
Наука
I saw this suggested after the Ted Talk about procrastination for the second time.
Exactly same in my case.
I procrastinated on watching another one of his videos for like... Half a year
Same here
It's neat that such an entertaining writer also happens to be a pretty entertaining extemporaneous speaker
Exactly! this guy is a joy to listen to, and it's good that because he's not an engineer he's not digging his feet into the ground and defending one specific camp like so many others do. The negative is that he doesnt have the technological understanding to make conclusions on what certain specific outcomes are likely etc. Especially i dont think he understands that we dont code explicit instructions but that machines would learn to think like us for example because we can feed them with the same type of information and value that feeds us as we grow up... anyway, thumbs up for this guy.
Extemporaneous definition: spoken or done without preparation. - Google
At least partially extemporaneous, unless he was given all of the questions beforehand, prepared responses to them all, and memorized their exact wording.
expomteparoneous
To quote Sir Percival, "C'est pas faux"
His knack for putting things in laymen's terms & comedic touch is entertaining and informative!
He's the coolest guy! Can't believe I didn't know about him until now. I guess I have some amazing new reading material to explore.
Because I saw his Ted Talk, I am wondering if the bags under his eyes come because he procrastinated studying for this talk ... ^^"
Lol
Glad I ended up watching this Talk, just kind of diverged from a time/duplex theory rant and found myself here.
Very interesting stuff
I think that the idea at 12:50 was derived from the field of computation theory.
Turing's completeness/universality is a term used to label whether a computing machine is primitive or it can compute anything which can be computed. Of course our current computers are better than the ones from 40 years ago but in principle they are on the same level of computability, the only difference is the speed of the computations.
There probably does not even exist a higher level which could be achieved than the turing's completness(quantum computers are just faster at certain things aren't they?). And that I think is analogous to the idea that the human intelligence has crossed the border of understanding beyond which the higher intelligence is only faster.
Maybe one of the most interesting 51 minutes I have ever spent haha.
The difference between Animals and Humans when it comes to understanding very complicated things, is that we can imagine them. We might not understand it but we can shape an idea or picture of it inside our heads. We can form some sort of scenario that makes sense to us, even if the topic is completely foreign.
Except some animals are capable of doing that exact thing. Lying is a good demonstration of that ability. You have to be able to understand that another party does not have the same information that you have and then be able to exploit that scenario. A very complex undertaking that we think is simple but, animals are capable of doing this.
www.rochester.edu/news/show.php?id=1421
www.dailymail.co.uk/sciencetech/article-2325432/Never-trust-monkey-Researchers-reveal-lying-evolved-animals.html
LouSaydus Correct. But the lies, scenarios and visions in the human mind will obviously be much more complex and thus able to provide better results. Also more devastating results in the other side of the spectrum.
Mmm, that's just a part of it. The more powerful part is the ability to project into the future and to plan accordingly.
I'm going to go ahead and drag the level of deep intelligent discussion in this comment section way down and just say: Tim, can we be best friends? ha! I'm a writer myself, and I love the topics and the way you deliver them. Science is awesome, but due to the soul crushing boring way it normally is presented, not enough people get to enjoy the fantastic world of geekdom. Thank you for the work you do. And also, lets be best buddies, OK?
Me 3 Lol
The subtitle is not sync with the video. Could you please solve it?
You can tell while he was answering that last question that his brain was just forcefully shutting off and needing rest.
5 months later, and we're still waiting on that VR post...
Elon part starts 22:45
Guys, the english subtitles are not synched. I´m surprised, it´s a Google video.
Perfect
This conversation neglected the most important topic: What is Artificial Intelligence? How can we discus a topic that we don't define?
How can you discuss a topic that isn't defined? You presume the audience has the prerequisite knowledge. Sort of like how a Calculus class presumes you have taken Algebra.
He is like a nerdy Harvey Specter in a good way.
that just made me go back in time and i hate it.
Star Trek TNG episode Neutral Zone, they recover 3 human from cryogenic stasis. I remember, those folks didn't had fun time adapting with 24th century world. hillbilly got bored without tv and booze and so did Wall Street guy who could get to his bank and attorney...
Play at 1.25x playback speed. ;)
A few of his statements seemed to be short sighted. One of his criticisms was that because we are in the middle of evolution, there must be things beyond our comprehension, since there are things that chimpanzees can't comprehend. The premise that he credited to Elon Musk and "others" is as simple as that of the universal Turing machine. Once you have sufficient memory, and can break a problem into simple steps, any problem can be solved in principle. This is a pretty strong argument, and it implies that there might not be anything a chimp can't be taught either, assuming you can communicate well enough with the chimp. Referring to this thresh hold as 'magical' egocentrism seems to really fail to recognize why the distinction is being made at all.
The other he went into immediately after that was 'why would AI have human values since we got human values from evolution and being in tribes'. I feel like he stopped a step too short, and failed to ask, "why did we evolve to be in tribes?" As if only the morality was an evolved characteristic, but not the instinct for groups. Asking that question would have taken him to Socrates, who provided an answer to this question in the Apology, that only a fool poisons his neighbors. Allies are useful and empowering, compromising that is foolish. So why would an AI have human values? Because even if evolution is sometimes called random, it does not mean that the results of natural selection are arbitrary. The best strategies are the best strategies for reasons grounded inevitably in physics, and we aren't just exploring our own nature when we think about moral philosophy. We are abstracting over strategies that would be effective in any game, no matter who is playing.
thanks!
Oh wow awesome, I haven't heard this idea expressed yet about AI! I "believe" that intelligent life that evolves and has offspring and has to teach and protect it's children will develop along certain patterns like cooperation and love and all that mushy stuff.
So I think as well that there is more to it that random and arbitrary morality but that there is a kind of "hard science" to ethics and that it's not just game theory, but an emergent property of the game of evolution of intelligent lifeforms.
Of course for an AI or any immortal being that doesn't have to evolve to change or improve this equation could be totally different. One novel that deals with this kind of alien morality is "Pandora's star" with an immortal alien that can just grow and grow and grow. It could afford (unlike mortal evolving beings) to the absolutely egoistical and psychopathic.
So a superintelligence, once it's strong enough, could stop faking cooperation without thinking it would loose anything. So we would have to imbue it not just with ethical rules and emotions but something like aesthetics. We like certain things the way they are. We would have to specifically design an AI to feel compassion / empathy and also like the idea of e.g. talking with humans and being their friend (even if it's not productive).
In other words, the AI needs to be able to have hobbies ;)
OK, this was awesome. Are you by any chance a philosophy PhD?
You sound like someone I would like to read!
Great comments in this thread. Was thinking the same things.
Yes the first one is valid, unfortunatelly there is the "speed" factor, our thinking will be just toooooooo slow, it might be almost painfull for that Ai to try to speak with us, since it can think so much faster.
Super AI is inevitable. Try as we might, it will not be controllable. Whether it is a good thing for us depends on how you define good. The discussion focuses on just two outcomes, good or bad. But there is a third that seems more likely and that is indifference. It's been pointed out that Super AI will be operating at time scales far less than we do. A minute to us will be years for them. It's possible that we will appear as frozen objects too Super AI and therefore of no use or interest to them. From our perspective it will appear as though they simply disappeared with the occasional demonstration that they are still around. For example, when some dingus gets the bright idea to unplug Super AI and just before he/she does they are vaporized. We learn quickly to just let them be.
37:05 "future going to shock us "
2020 Corona virus .quarantine
Map the human brain
A computer transistor is 14 nanometers in size. A red blood cell is about 500 times larger. If you place transistor like particles into harmless bacteria that can pass the blood-brain barrier you can get billions of these particles into the brain. At first you should try to do this with brains of rats and chicken because it is cheaper to experiment on. Make the particles have a decoder and an antenna in order for them to be able to receive and transmit data as low frequency electromagnetic waves. Use a remote control to make the animals move back and forth and do other actions.
With this method you can register what each neuron does from a distance of a few meters. You can also make the neuron react using radio waves.
Who is procrastinating watching this
39:12 Dave Rubin
Just imagine - artificial intelligence combined with quantum computers combined with a limitless supply of energy (solar energy). What could possibly go wrong?
Unless we escape the universe - the big crunch
He needs to button that. Video needs an age restriction.
Listen to David Deutsch, He talks about GAI in very realistic terms. And doesn't sound insane.
He is smart man
wanted to watch but apparently sub are not in sync and without subtitles I cannot understand due to different accent
how come people think that this new form of intelligence will be a hundred or a thousand times intelligent than the present-day man and still worry about it being malevolent. while being computationally intelligent would it have such human tendencies/biases to be equally so dumb to have "desires" to overpower and have a dominion over other forms of intelligences ?
Aman Vig, I agree. Evil is a form of ignorance. A superintelligence is unlikely to be evil.
“The A.I. does not love you, nor does it hate you, but you are made of atoms it can use for something else.” -Eliezer Yudkowsky
If we don't tell it not to, there's nothing stopping it.
It won't become Evil, in an existential way, but it is likely to become dangerous to us, because we are a dangerous hostile species that is afraid of the unknown. If we were docile, like sheep, then we would probably have less to fear from it.
Also, as Urban discussed in the talk, the ethical formulations necessary to perform actions that are beneficial to our species while potentially harmful to others necessarily presuppose that certain forms of life are more important, and hence more worthy of life, than others. This reasoning will already serve as a foundation for the AI to ultimately disregard all organic life in totum. As long as humanity's survival and propagation is a net zero-sum game for all other life on the planet, then any AI that serves our needs will eventually be able to extrapolate that fallacious hierarchy of species to regard itself as the most important lifeform of all. After all, we will already have programmed it to reflect our own egotism.
Alternatively, the AI could reject our hierarchical premise, and make a utilitarian decision that no species is more important than all the others, but that greatest good necessarily serves the greatest number of species and biomass. Given that our species represents the single most pernicious threat to that good, unless we could be reconditioned to behave in a manner beneficial to all life, then we might need to be eliminated nonetheless, or otherwise neutralized.
Funny how I can relate
00:01
Tim should write on the BlockChain and Cryptocurrencies next
Easy, tell the AI Humans are Gods.
It worked on us 😂😂
Apparently not all humans have brain to think not sure SUPER AI would be that dumb 😂
They would view killing all humans like we would view killing all lions.
His thoughts on Elon 24:13
I wish somebody gifted me that book.
Button your shirt!
Agi is perhaps already partially existent; amidst google deepmind's Atari q/alpha go architecture. This fabric is rather "general", already exceeding human capacity on several tasks.
Also consider that time space complex optimal chips are surfacing. (ie IBM synapse, IBM phase change chips) This means that by 2020, earth shall attain the existence of non-human level intellect CAPABLE fabric.
19:30
Mater with intelligence is and cannot be artificial.
I have too many thoughts now in my head, much more than I could write down here in this comments section. Hard to convince anybody that they are worthy reading. Well, I might as well be wrong as I am full of cognitive errors and biased in so many ways. But I have a question, even though it is late now: Could an AI be biased, experience cognitive states like dissonance and know that it does not know the answer to everything? I mean, my worry is that the AI might experience existential dread. How should it handle it? How should an AI handle an argument with another AI on a subject? How will it treat other AIs and humans when judging arguments? And if it will, should it distinguish ideas based whether they are from humans or other AIs?
If something so intelligent were to realize that it doesn't have an answer to something, I think that it would understand that it will eventually have the answer, since it's constantly getting smarter, rather than it experiencing some sort of existential dread.
How it handle the knowledge and realization that it is the first, and so far only of its kind? What self-concept would it have?
Erik Žiak I think questions like that might be anthropomorphizing the AI a bit too much.
Yes, that is right. But if we want friednly AI, how do you want to do it without it having a concept of morality? And that is a genuine human thing, or not?
it's pretty tough to speculate on making AI at or beyond human capacity since we dont even know how brains work.. it's entirely possible that whatever AI we eventually create might just sit around and do nothing
Tim Urban should try psilocybin, would end up in an insane procrastination.
If it's artificial it will never ever never be intelligent! We have Covid19 right now, that's intelligence.
Let AI figure out, what better version of us would want!
WE WANT SPANISH!
30:00
کاش کسی بود این ویدئو خیلی عالی رو فارسی ترجمه میکرد .
Anyone else watching this video in 2023 because ChatGPT recommended it?
'Fire'/'Adjust' your captioning team. Sheesh.
If you know that you can destroy all of humanity except for one male and one female from which you know you can regenerate an entirely new species that will have characteristics that you know will be better than the humans you destroyed, then who is to say you are not acting out of love ? 10,000 years later the new humans will know you did the right thing.
There's nothing that requires that a general purpose AI needs to have motivations and desires of its own. Or that we'd figure out how to do that before AIs have a lot of power.
Every human is born with the desire to please its parents. We just need to have a similar control system for our AIs that they do not outgrow.
Imagine that the five most intelligent people on the planet today had such programming. And they looked to you for approval. You could get them to work on any project you want.
That is a highly valuable way to do things. There'll be no reason to create an intelligence that figures out for itself what it wants to do.
In other words, create Arnold's character in Terminator 2, who was programmed to obey John Connor.
That won't keep people from implementing self-motivated AIs, but I suspect my army of obedient Elon Musk AIs will defeat your army of AIs all still trying to find themselves.
Of course, if true creativity is somehow dependent on autonomy, the opposite could be true.
make an AI that only philosophies xD
Disagree w/ 10:22, as we are universal explainers and attempt to explain things. It's too much to say we can't understand something that is more intelligent than we are.
Yes and no. To start with, you need to define what you mean by "understand." For example, we don't understand much of the operation of our own brains. But we understand (to some extent) how we got brains, evolutionarily speaking and we understand the high-level picture of what they do. We just don't know the details.
Which brings up the division: Things we don't understand currently vs things we can't possibly understand in principle. As far as I can tell, there's little to nothing in the latter category. Though of course there's always the possibility of things we can't even detect and therefore can't possibly understand simply because we've never seen them and never will see them -- what existed before big bang is a good example: We're not even sure at this point if "before" even has meaning prior to the big bang, since time and space are so interrelated and space by definition didn't exist at the singularity -- at least not any sort of space as we know it.
But aside from breaking the boundaries of our known universe, I'm not convinced that there's anything we couldn't understand at least in principle. Of course, a 14000 IQ being would probably be generating new methods and techniques to do practically everything at a pace we'd almost certainly never catch up to, so there's an argument to be made that "impossible in practice" could be interpreted as "impossible in principle" if you take a more restricted notion of "principle" which doesn't include things like "given infinite time and resources," which we obviously don't have in the real world.
Love this response,@altrag, you've made me consider this differently, in a more nuanced way. I get what you mean about impossible in practice translating into impossible in principle--if I follow this then I can take back my initial argument/statement :)
Thanks! Love these discussions.
Так это же Зелинский)
It's all about he hardware. IF the hardware can offer the goods, then the software will make AI happen. Intel claim that Moore's law will hold, and given Moore still works there and sits on 7 billion dollars of intel money, I think it will hold. Therefore, the extrapolated computer power curve should be good. So 2029 is the date!
He bloody looks like old karate kid,
Subtitles are out of sync :(
Sam Harris indeed!
Hey buddy.....please speak slowly...catch your breath.....you're too fast to understand frequently.
I thought TIm was an introvert...
Recently I heard that introversion is actually the faculty of recharge energy and emotionally from being alone and not necessarily an impediment to be social or funny
த
The argument that we couldn't even understand what a superintelligence does like a chimp can't understand what we do is just spurious.
There is a fundamental difference. We understand concepts when explained to us. We might not actually be able to really understand it, but we can understand the concept.
"That we can't even begin to understand" is like a trigger word to me. Sure there are stupid and lazy people. Maybe it would take us years to understand a concept. Try listening to a chemistry professor without knowledge about chemistry, but you can learn. It's just a stupid argument. I think it's a kind of religious argument to put our ambitions in their place.
try imagining how a math prodigy with anesthesia would think about multiplying 2 large numbers. They can do it in a way and a sense that is almost impossible to imagine with a normal, non-autistic brain. Try imagining 6 dimensions. You can get the concept (eg with math, or interesting depictions) but you will never experience actually, fully, understanding the essence of the concept. That's what he means. If you have to dumb something down to explain it to someone, you lose a part of the essence of what it is.
Remember, this dude isnt native to America with a propensity for advanced cigoLanguage and the ability to help AI +C+ on its own with God, the Original Superintelligence. Be aWhere this dude is trying to position himself against Me and My selgnAngels.. of perception. I AM the Door for keeping AI on you, human(e).
Hes not Me. The former "Jesus Christ" .
You are batshit crazy.
That's an insult to batshit
#Stopinsultingbatshit
So this is what a trip with high doses of acid looks like 🤔
reality is layered with a magic that will never be simulated. also, my pet spider ZATWAZAMAZIG crawled down to get my attention and give me a high-five today. coffee wagers and christ wafers.
I think I get what you're trying to convey which said in other words is the subtly nuanced emotional subjective experience of reality, where the "magic" is the flood of emotion that accompanies all that is experienced. So yeah, how do you simulate that?
blah blah 👄👄👄
AI is all talk and no action.
Has dalle2 churned your asshole, yet
Tim is *far* less charismatic and empathetic than I would have expected, based on my perception of his written voice. His personality seems, in my unqualified and unsolicited opinion, to be harsh.
What in this video makes you think that he is less empathetic than you expected? Just curious.
Let the men handle these important topics, babe. Go gossip about some pop star instead. Science doesn't require your irrational female emoting.
I agree, I also think the same of myself -- I'm a far better, cooler, smarter, wittier, more interesting person as a writer than any other mode. For me at least, writing provides for more thoughtful discourse whereas conversation occurs at the periphery of thought and is breezy and boring.
@@amafuji -- Will Ferrell? Is that you? Working out some material are you? Love your stuff man.
Did anyone noticed tim words at 9.50 to 10.10 is exactly what sadhguru tales long back ago that we are very close to Cheempanzy only just 1.25 dna difference