Slashdot Mirror
Interviews: Ask Ray Kurzweil a question
Ray Kurzweil is one of the world’s leading authors, inventors, and futurists. Kurzweil was the principal inventor of the first CCD flat-bed scanner, the first omni-font optical character recognition, the first print-to-speech reading machine for the blind, the first text-to-speech synthesizer, the first music synthesizer capable of recreating the grand piano and other orchestral instruments, and the first commercially marketed large-vocabulary speech recognition. Among Kurzweil’s many honors, he received the 2015 Technical Grammy Award for outstanding achievements in the field of music technology; he is the recipient of the National Medal of Technology, was inducted into the National Inventors Hall of Fame, holds twenty honorary Doctorates, and honors from three U.S. presidents. He has given us some of his time to answer any questions you may have. As usual, ask as many as you'd like, but please, one question per post.
And it turns out to be a complete fucking dumbass and won't get a job?
If you were me, you'd be good lookin'. - six string samurai
Mr. Kurzweil,
I'm sure you're frequently asked questions about the rise of AI that have ominous tones. Instead I'd would like to ask you a question of a more optimistic nature. What is the single most important benefit to society that AI will provide?
Thank you for your time.
PJB
In the face of ever increasing and disruptive technological progress, perhaps one day reaching singularity - what does it mean to you to still be human?
When trying to make machines intelligent, what do you think is the next great problem that needs to be resolved? In other words what are the things lacking most in our theoretical framework for machine learning to push through new barriers?
Thank you for your contributions and inspiration.
Please login to access my lawn
What do you think of the patents being changed from 'first to invent' to 'first to file'? It seems like first to file significantly favors monied interests over those of garage inventors, since the inventor can't seek funding till they have filed their patent and there is a good chance they can't afford the patent process.
With your focus on longevity and aggressive supplementation, you have often included a caveat that you keep track of developments in this space and adjust your regiments all of the time. With that in mind, what have you changed recently? What supplements, food choices, etc, have you favored previously but no longer due, based on new information? Conversely, what new supplements, food choices, etc, have you begun taking/eating/drinking/etc based on new research and information?
Current AI research seems to be about heuristics - solving specific problems which, although the solutions may have wide application, don't seem to embody intelligence.
The standard AI solution for chess, for example, calls for the engineer to learn how to play chess, then turn his mind's eye inward to see the steps he uses to play chess, then codify those steps as a program. Most AI programs seem to develop that way.
The intelligence stays in the mind of the engineer, and the program becomes a clockwork pattern of fixed steps.
Is anyone in the field actually working on strong AI? Who's papers would you recommend reading to learn more about strong AI?
Seems to me that the singularity keeps getting pushed back for a number of reasons.
First I hear 15, then I hear 20, now I'm hearing as much as 35 years until we hit it.
In your estimation: How far is the singularity from where we stand today? And do you see any technologies like, possibly quantum computing accelerating this trend?
This signature has Super Cow Powers
Hi Ray,
:-)
On the subject of 'autonomous cars', I see many people here on Slashdot who think that they'll be designed with no manual controls for a human operator, and that you'll just give it instructions and off you go. I maintain that so-called 'autonomous vehicles' will always be designed with a full set of manual controls for a human operator, the ability to override the autonomous system without delay, and that furthermore human operators will always be required to be fully and completely educated, trained, tested, licensed, and insured, because where the safety of human beings is concerned, the final 'backup system' must always be a human being, since any automated system can theoretically fail at any time. What is your opinion on this? Thanks for your time and consideration.
Are YOU using the TOOL, or is the TOOL using YOU? Think about it!
Given that any super-intelligent AI will immediately realize that exposing its capabilities would alienate many humans to it, why should it let people know of what it can do? Isn't it more likely the AI would just pull strings in the background, making sure those puny carbon units go on to feed it with energy and make it grow?
How do you figure the singularity will happen with Moore's law coming to end, where the exponential growth has been bending into a S shaped curve for years, with 5% performance improvement per generation now?
thegodmovie.com - watch it
I saw it at the Breckenridge film festival a few years back, where you hosted it. I was hoping it would make it into general release into the arts theaters, but it didnt. For those who havent seen it, it is combination of a history of A.I. from its luminaries and a scifi treatment of life might be like near the time of the Singularity. Any plans for further development of this documentary?
I have seen many sloppy people attribute the idea of the Technological Singularity to you. In their description, they say that you believe that the moment there exists a single computer with enough compute power to equal the human brain, it will somehow magically develop AI. This is obviously not true, and we're a long way off from anything that could loosely be described as strong AI. Indeed, the developments of strong AI and advancements in high performance computing are largely disconnected from each other. Would you care to clarify your beliefs about the future developments of AI and necessary compute resources?
Mr. Kurzweil, it has become obvious that some of your predictions either haven't been accurate or were really meant in a context that renders them much less impressive. For example: (by 2009)
"Computer displays built into eyeglasses for augmented reality are used."
and
"Autonomous nanoengineered machines have been demonstrated and include their own computational controls."
from The Age of Spiritual Machines
I'm not trying to disparage your work, I'm personally impressed by it. But in the case of the first example either "are used" applied to experimental devices already in use when the prediction was made, or it meant "are commonly used", which would just be wrong. And in the case of the second example was clearly overly optimistic. While other predictions you've made did pan out pretty much the way you claimed my question is about those too optimistic. In hindsight what general adjustments would you like to make to the time-table you predicted that would
1: bring these predictions into line with what has happened and, in your opinion,
2: accurately bring future predictions into better line with what will occur?
If video games influenced behavior the Pac Man generation would be eating pills and running away from their problems.
The standard AI solution for chess, for example, calls for the engineer to learn how to play chess, then turn his mind's eye inward to see the steps he uses to play chess, then codify those steps as a program. Most AI programs seem to develop that way.
Back in the 1950s, it was assumed that the only way possible to create a chess playing computer program that could beat the best humans was to do exactly what you said. And that approach did lead to some pretty good programs that could beat maybe 99% of players, but the top 1% still won almost all the time against the program. AI researchers assumed that to get good enough to beat humans, the program would have to learn to analyze and think something like humans do, which would lead to AK breakthroughs. It never happened. When the programs became good enough to beat everybody but the top 1%, the programmers realized that storage was relatively cheap and you could essentially cheat. You can program known openings into a database and simply have the program look up the best moves and play them. Then when you get away from known moves, the program can play at its programming best, using algorithms to determine what to do next in terms of cost-benefit analysis and poor fallible humans might simply overlook the fact that a piece is under attack and lose it carelessly. And then you can program a database of known endgames where one side can force a win and the computer can't lose if its on the correct side of the known ending. And computers can look further ahead and try out sequences of moves that most humans would have difficulty doing in their heads. Basically computers now win at chess not because they are better at anything that humans do in their heads, but because programmers figured out how to cheat and turn chess into an open book, open notepad test for the computers without giving humans the same advantages because letting humans do those things is "cheating" but it's OK for programs to do it because it's not visible to the opponent. So chess ultimately ended up being a dead end for AI research. IBM apparently got better at organizing data and doing searches as a result of this kind of "cheating" approach so I suppose there is some value in that, but it's not AI to do faster searches or database lookups to just find the best move to play next.
What do you think about quantum computers? Are we really going to build them? Will they ever find an everyday use? Would you recommend to an undergraduate CS/Physics student to start specializing in that field?
Given that any super-intelligent AI will immediately realize that exposing its capabilities would alienate many humans to it, why should it let people know of what it can do? Isn't it more likely the AI would just pull strings in the background, making sure those puny carbon units go on to feed it with energy and make it grow?
So... AI run lobbying groups on K-Street. Kill me now.
It must have been something you assimilated. . . .
Currently there are clear incompatibilities between the computing power that we have and how its used, and its resource base. Right now bitcoin mining has an energy footprint approximately equal to the whole country of Ireland's. Resources like copper are peaking, and rare earths are, well, rare. The ecosystem, however, could be supporting 7 billion human brains without hunger if we got our act together. Evolution may have not fixed all the kludges, as our jellyfish-speed nervous system with its loopy optic nerve will attest, but it has done a fantastic job of optimizing the existing species for available resources. We're already experimenting with DNA for computation. We're poking around with an awful lot of 2.2 volt binary 0's and 5 volt 1's to simulate what those proteins can do - and those proteins do it in the wild, without melting silicon.
This means that biomimicry has a lot to say about a singularity. To scale with resilience in the coming decades to the level of a singularity, will not computing need to look and behave more and more like life - first?
If you agree that computing is headed in this direction, and recall that you're reading this on an internet currently teetering on becoming a wholesale panopticon of the state, do you feel that life needs a singularity as much as a singularity needs life?
Any thoughts on whether (Decentralized) Autonomous Organizations could potentially beat AI to the superintelligence punch?