Nonsense. Why shouldn't AIs be possible in completely different environments than humans? The assumption that bodily experience is required is just wrong. Why shouldn't an AI be able to live on the internet, for example?
Also, human brains are actually also just computers in a box with some wires to the outside world. We can not even be sure that the outside world exists, we only model it based on the signals that come through the wires.
I think he is arguing that we have an existence proof of exactly one type of intelligence and to assume that we can skip understanding that one and create a new kind is pretty far fetched.
Agree. The arguments put forth are nothing but romantic appeals to human supremacy. First, that you can't have intelligence without meat. Second, that you can't have intelligence unless it learns by the same bumbling process that humans use. (An assertion that early intelligent behavior is inversely proportional to later intelligent behavior.) Third, "build a body"? Really? Please email your recipe for building a body that will exhibit intelligence without Darwinian selection across the entire universe to...I don't know...quack.com.
I actually agree with you. They should be possible, and I'm sure they will exist. However in terms of building one, the only roadmap we have includes a body. To my mind it's not either/or, its now or later?
It's certainly valid to get inspiration from the human mind. Still not sure that requires a body, though. I often wonder how a human brain would react if it were born into a completely different environment (for example some virtual reality). Also aren't there humans with brain damage and hence very limited sensory input?
I suppose the human brain has some specializations hardwired into it, like face recognition, learning a language and so on (dealing with a body...). That is certainly something to think about.
Anyway, I liked Jeff Hawkins book, I am sure you are doing good work at Numenta. And dealing with a body is one of the things some AI should eventually be able to do, so it doesn't hurt to follow that route.
He joined Google as Director of Engineering. He didn't join Google as Director of AI Research. Yes, we all read that Ray's intended project was what got Larry Page to hire him, but Director of Engineering over an organization of thousands of engineers is a task all its own. I can't imagine that Ray Kurzweil shows up to work in the morning and tinkers about with his pet project all day; running Google's entire engineering team is something most of us should recognize that we are not adequately prepared to do at this point in our lives.
This entire letter just strikes me as in extremely poor taste. You disagree with Ray Kurzweil's views on AI, because they don't align with yours or your company's[1], so you write an open letter to Larry Page effectively saying that he hired the wrong guy and should have hired you? Very, very transparently self-serving. I realize the self-service was never written verbatim, but it is the subtext; what other purpose would there be to write this letter? "His AI won't work, mine will."
I also have a very strong and violent distaste for anybody who tells me something won't work before even daring to try it on their own. The contrarian engineer, "oh, that will never work," often muttered in reply to using some piece of architecture he has no experience with. He never apologizes when it works, either. Who are you to say what's possible and impossible? Maybe Ray Kurzweil knows something you don't?
Why don't we quietly commend a distinguished engineer and inventor with a decades-long career (undoubtedly, longer than this guy has been alive) for being offered a position worthy of his résumé and leave it at that, often-controversial views aside? Everybody gets caught up in Ray Kurzweil's futurist thinking and forgets that he invented some pretty pedestrian stuff that you and I rely upon daily.
There's a difference between believing in something and explaining it, which I would have loved to read, and openly criticizing Larry Page's hiring decisions because you disagree with the hire's views.
How would you feel if you got hired somewhere, it was a big enough event to make news, and I wrote an open letter saying "bummer you hired geofft; his views on database scaling will never work. Oh, by the way, I run a company that's doing cutting-edge research on database scaling. Glad my name's out there now!"
I, and by extension, you, are assuming it was a misconception at all in the first place. My jury's still out.
Could be a "his views on AI will permeate Google as a whole, too, who happen to be in a position to execute on it" sort of thing, too, knowing full well why he was hired. I agree with the tongue-in-cheek jab above, though, sounds more like a "I know the answer, why don't you?" sort of letter.
I would be humbled that news of my hiring were big enough news to attract that sort of attention, and certainly that companies think it's an opportunity for publicity. I would also hope that I had not become so self-important a person that I couldn't reply on the blog and have a nice (if brief) discussion on the merits.
"why don't these idiots recognize how brilliant I am!!!?"
We all know where this is going, he is going to prove how good his AI is by making and selling a kill-bit to the government s of the world and will go down in history as the evil POS that have indiscriminate killing robots to the world just because Google hired Ray and not him.
While Ray will go down as building the helpful AI UI in the worlds self driving cars.
Hrm. This is a valid critique, I'm sorry it came off like this.
Let me say this, if I'm wrong, and I very well might be, I won't have a moments hesitation in lauding his success. Just as you rightly point out Mr. Kurzweil should be lauded for his many past successes. I wrote this because I care deeply about this topic and want very much for Mr. Page, Mr. Kurzweil, all of us to succeed. I happen to have a bit of experience and knowledge that inform these beliefs but to me this is a goal far too important for ego, so if I'm wrong it won't stop me from cheering.
I appreciate your comment and will try to take your viewpoint into consideration in the future.
Thank you for being reasonable in your response, and I'm glad that you've taken the comment to heart.
I agree with another commenter: I think this letter would have been better left closed. I had the privilege of meeting Ray Kurzweil last year in New York at an intimate talk (he was annoyed he had to cut his preso down to an hour), and he's a genuinely smart guy and a real pleasure to talk with. That being said, he has some genuinely bananas-crazy ideas, and they might work or they might not. I'm happy to see him try on the off chance they will, though, and I'm glad Google is paying him to explore some.
I know a lot about Paxos after spending years studying it and its implementations, but I'm not going to openly and publicly criticize Mike Burrows for his work on Google's Chubby. He knows a lot of things that I don't, and I'm smart enough to realize that. I'm more game to learn from his work than criticize Google for hiring him. I do take who a person is into account when thinking about their body of work, but only in the positive.
That's where I'm coming from. You care about all of us working together and succeeding, yet try to divide with your letter, so it's a bit strange. That's all.
"Director of Engineering" does not mean they are in charge of the engineering organization. At Google there are many people with the title "Director of Engineering". It's a 2nd level management position, leading perhaps 50-100 people.
Really? I know less about the GOOG organization than I thought, and I stand corrected. Thank you for correcting me.
Who is in charge of the engineering organization? A SVP? I would have assumed a Director of Engineering would have reported to Vic Gundotra, and overseen engineering as a whole.
That undermines half of my comment, so I'm glad you said something.
Well, it did rather obvious to me that Google, one of the most brilliant engineering companies in the world, wouldn't put Ray Kurzweil in charge of its engineering operations, a field he is known not to specialize in. Just as much, being such an engineering focused company, Google is going to have a lot of management people with titles related to engineering.
I'm not sure what you're trying to tell me, aside from I was too stupid to realize that the word Director doesn't translate among different organization charts. Is that accurate?
I'm glad it was obvious for you? I'm not sure how to reply to this one. In my org, a director is at the top of a chart.
Also, quite the contrary, Kurzweil was quite the engineer before becoming involved in futurism. A cursory Google will reveal several of his inventions. There is a Kurzweil K2600 sitting about six feet from me on display.
I'm very skeptical of this recent emphasis on the theory that our complete sensory experience has to be simulated (or replicated via robotics) in order for a strong AI to learn.
All you need to do is read a bit about Helen Keller to know that language should be a sufficient window into our world:
I make the argument that 'some' of the sensory experience must be available and that interaction with the real world is crucial. To your point about Ms. Keller, I have used this example myself as to why robotics should focus more on the sensation of touch and it's why I commissioned the development of a low-cost etextile robotic skin from a brilliant designer/engineer at MIT. That project (rSkin) can be found here (http://www.instructables.com/id/rSkin-Open-Source-Robot-Skin...)
Where did you make the argument? I see you making the claim, but there's nothing arguing it, no evidence. There's just the vague thought experiment about extracting signal from noise.
To point it out another way: humans can't be intelligent because they are lacking in six senses that Martians have. It's that even richer experience that enables true intelligence; without these qualia and the vital marrow of the foundation of the core of the heart of really truly scotsmany work, humans can't be intellegent. Can you knock down this claim without knocking down yours as well?
Most importantly, building a machine that can really communicate with us without giving it a body is obviously a dead end. If the point for Ray is mastering language in AI, then I think Ian is right.
Reading "Methaphors we live by" By George Lakoff really opened my mind on this subject. Mastering not only the forms of language, but the meaning, and so the concepts, requires an extensive amount of interaction with the world through a human body.
Just think bout the development of a child that can't touch, hear, see, smell, or can't feel any emotion. What about his communication skills ? We do have exemples.
It is an engineers dream that communication would be devoid of emotions, shared experience, and many other things fundamentally born from experience. We all know this to be false if we care to study a little about communication outside of the realm of engineering.
AI may be possible without a human body, but if there is to be communication, then there is to be a body whatever its form, or a simple form of what we call a "self". In my (lacking) knowledge, embodied AI is the only branch of AI to have considered this deeply.
Ian may be right. Maybe we rush to the conclusion without taking many things into account. Maybe this is the curse of a branch of CS that gave to much promises from the start.
Then AI is probably too broad of a term. Some things in intelligence may be in our reach, some others not.
29 comments
[ 4.4 ms ] story [ 71.3 ms ] threadAlso, human brains are actually also just computers in a box with some wires to the outside world. We can not even be sure that the outside world exists, we only model it based on the signals that come through the wires.
I would say 'more difficult than it needs to be.' But essentially correct.
I suppose the human brain has some specializations hardwired into it, like face recognition, learning a language and so on (dealing with a body...). That is certainly something to think about.
Anyway, I liked Jeff Hawkins book, I am sure you are doing good work at Numenta. And dealing with a body is one of the things some AI should eventually be able to do, so it doesn't hurt to follow that route.
This entire letter just strikes me as in extremely poor taste. You disagree with Ray Kurzweil's views on AI, because they don't align with yours or your company's[1], so you write an open letter to Larry Page effectively saying that he hired the wrong guy and should have hired you? Very, very transparently self-serving. I realize the self-service was never written verbatim, but it is the subtext; what other purpose would there be to write this letter? "His AI won't work, mine will."
I also have a very strong and violent distaste for anybody who tells me something won't work before even daring to try it on their own. The contrarian engineer, "oh, that will never work," often muttered in reply to using some piece of architecture he has no experience with. He never apologizes when it works, either. Who are you to say what's possible and impossible? Maybe Ray Kurzweil knows something you don't?
Why don't we quietly commend a distinguished engineer and inventor with a decades-long career (undoubtedly, longer than this guy has been alive) for being offered a position worthy of his résumé and leave it at that, often-controversial views aside? Everybody gets caught up in Ray Kurzweil's futurist thinking and forgets that he invented some pretty pedestrian stuff that you and I rely upon daily.
[1]: https://embodiedai.pbworks.com/w/page/29940748/Welcome%20to%...
EDIT: Scott Draves corrected me on my assumption of what a GOOG Director of Engineering is, below. I'm quite wrong.
How would you feel if you got hired somewhere, it was a big enough event to make news, and I wrote an open letter saying "bummer you hired geofft; his views on database scaling will never work. Oh, by the way, I run a company that's doing cutting-edge research on database scaling. Glad my name's out there now!"
It's just shitty and low-brow.
Not forgetting that the letter is based on the misconception that he was hired to work solely on his AI project, rather than Director of Engineering.
Could be a "his views on AI will permeate Google as a whole, too, who happen to be in a position to execute on it" sort of thing, too, knowing full well why he was hired. I agree with the tongue-in-cheek jab above, though, sounds more like a "I know the answer, why don't you?" sort of letter.
We all know where this is going, he is going to prove how good his AI is by making and selling a kill-bit to the government s of the world and will go down in history as the evil POS that have indiscriminate killing robots to the world just because Google hired Ray and not him.
While Ray will go down as building the helpful AI UI in the worlds self driving cars.
Let me say this, if I'm wrong, and I very well might be, I won't have a moments hesitation in lauding his success. Just as you rightly point out Mr. Kurzweil should be lauded for his many past successes. I wrote this because I care deeply about this topic and want very much for Mr. Page, Mr. Kurzweil, all of us to succeed. I happen to have a bit of experience and knowledge that inform these beliefs but to me this is a goal far too important for ego, so if I'm wrong it won't stop me from cheering.
I appreciate your comment and will try to take your viewpoint into consideration in the future.
Thanks.
I agree with another commenter: I think this letter would have been better left closed. I had the privilege of meeting Ray Kurzweil last year in New York at an intimate talk (he was annoyed he had to cut his preso down to an hour), and he's a genuinely smart guy and a real pleasure to talk with. That being said, he has some genuinely bananas-crazy ideas, and they might work or they might not. I'm happy to see him try on the off chance they will, though, and I'm glad Google is paying him to explore some.
I know a lot about Paxos after spending years studying it and its implementations, but I'm not going to openly and publicly criticize Mike Burrows for his work on Google's Chubby. He knows a lot of things that I don't, and I'm smart enough to realize that. I'm more game to learn from his work than criticize Google for hiring him. I do take who a person is into account when thinking about their body of work, but only in the positive.
That's where I'm coming from. You care about all of us working together and succeeding, yet try to divide with your letter, so it's a bit strange. That's all.
Who is in charge of the engineering organization? A SVP? I would have assumed a Director of Engineering would have reported to Vic Gundotra, and overseen engineering as a whole.
That undermines half of my comment, so I'm glad you said something.
I'm glad it was obvious for you? I'm not sure how to reply to this one. In my org, a director is at the top of a chart.
Also, quite the contrary, Kurzweil was quite the engineer before becoming involved in futurism. A cursory Google will reveal several of his inventions. There is a Kurzweil K2600 sitting about six feet from me on display.
All you need to do is read a bit about Helen Keller to know that language should be a sufficient window into our world:
http://en.wikipedia.org/wiki/Helen_Keller
To point it out another way: humans can't be intelligent because they are lacking in six senses that Martians have. It's that even richer experience that enables true intelligence; without these qualia and the vital marrow of the foundation of the core of the heart of really truly scotsmany work, humans can't be intellegent. Can you knock down this claim without knocking down yours as well?
Reading "Methaphors we live by" By George Lakoff really opened my mind on this subject. Mastering not only the forms of language, but the meaning, and so the concepts, requires an extensive amount of interaction with the world through a human body.
Just think bout the development of a child that can't touch, hear, see, smell, or can't feel any emotion. What about his communication skills ? We do have exemples.
It is an engineers dream that communication would be devoid of emotions, shared experience, and many other things fundamentally born from experience. We all know this to be false if we care to study a little about communication outside of the realm of engineering.
AI may be possible without a human body, but if there is to be communication, then there is to be a body whatever its form, or a simple form of what we call a "self". In my (lacking) knowledge, embodied AI is the only branch of AI to have considered this deeply.
Ian may be right. Maybe we rush to the conclusion without taking many things into account. Maybe this is the curse of a branch of CS that gave to much promises from the start.
Then AI is probably too broad of a term. Some things in intelligence may be in our reach, some others not.
I haven't ready Metaphors we live by, I'll pick it up!