Consolidation of Links on Friendly AI Thursday, Sep 7 2006
friendly ai and singularity 4:10 pm
Friendly AI on Wikipedia.
Works by the Singularity Institute/Eliezer Yudkowsky:
Why We Need Friendly AI
SIAI Guidelines on Friendly AI
Artificial Intelligence and Global Risk
Creating Friendly AI
Cognitive Biases Potentially Affecting Judgment of Global Risks
Features of Friendly AI
Coherent Extrapolated Volition
From SL4:
What is the difference between friendliness and Friendliness?
A Galilean Dialogue on Friendliness (in progress)
Friendliness not an add-on
Thinking About Friendly AI
Simplified Friendly AI
Multimedia:
Yudkowsky at the Singularity Summit (audio)
Hard AI Future Salon with Eliezer Yudkowsky (video)
How do we more greatly ensure responsible AGI? (video)
Critiques of the above:
Friendly AI is bunk by Shane Legg
Alternatives to (Yudkowskian) Friendly AI proposed on the SL4 list
Critique of the SIAI Guidelines on Friendly AI by Bill Hibbard
SIAI’s Guidelines for ‘Building’ Friendly AI by Peter Voss
By Oxford philosopher Dr. Nick Bostrom:
Ethical Issues in Advanced Artificial Intelligence
When Machines Outsmart Humans
By Dr. Ben Goertzel, Novamente LLC:
Encouraging a Positive Transcension
Thoughts on AI Morality
The All Seeing A(I)
Ben Goertzel on AI Safety (SL4 wiki)
In reading about the goals for Friendly AI, mention is often made of altruism. I can’t help but thinking though, most of the day to day things we do are motivated by selfishness. We work to sustain ourselves, make ourselves more appealing to mates, perpetuate our DNA though kids and grandkids, and so on. I don’t think this is necessarily bad.
Is this something to be concerned about? An AI which values alruism might find our current way of life completely unacceptable.
Altruism means tolerating the selfishness of other people. If selfishness is part of our identity, some people won’t want to discard it. And a true Friendly AI wouldn’t want to force them. We just want the AI to be selfless. Because the immediate result of selfish superintelligence would be our demise.
Hopefully, in the long run, making us happy will be such a minor task that regardless of the reason (altruism, nostalgia, fun), a strong AI will do it anyway. I’d still rather become the AI.
Nice list of links!
A number of the links date from the time Eliezer Yudkowsky was planning to create an humane altruistic superintelligence. In fact, all of the first set of links to Eliezer’s work but Artificial Intelligence and Global Risk, Cognitive Biases…, and Coherent Extrapolated Volition. These are still good things to read, but they represent work that has been surpassed.
Michael’s statement applies to those works: we want an AI that is an altruist, not an AI that forces all humans to be altruists
More recent developments (i.e. Coherent Extrapolated Volition: http://www.singinst.org/friendly/extrapolated-volition.html) propose AIs that cannot be considered people, so neither altruism nor selfishness apply. These are systems that bring (certain) good things to pass, but in a very different way to humans.
Tay:
AIs in general should not be thought of as people. Some AIs will be people, most will not. The former is probably more difficult and the latter is generally preferable: we don’t need to construct a sentient mind to fix our immediate problems.
AIs don’t have altruism, nostalgia, fun, unless we specifically engineer them in. To see a good argument for this, see section 1 in http://www.singinst.org/AIRisk.pdf and http://www.singinst.org/CFAI/anthro.html . So an arbitrary AI will be indifferent to the human configuration of atoms, just as most humans are indifferent to the exact configuation of sand particles on a beach. In other words, it would turn a human into a computer chip just as readily as we would turn sand into a computer chip.
So if we want a god that keeps us as intriguing pets on a whim we have to specifically design one. Given this, we can asipre to do better than such a god e.g. create a mind that truly cares about humanity, a humane altruistic superintelligence. We can do better even than that (see the Coherent Extrapolated Volition link above), but that’s another story.
Good points, Nick. I don’t think of AIs as people in any sense of the word. I think a distinction needs to be made though when an AI reaches a conscious state. It is at this point where I would say, anything can happen. Maybe they will be indifferent or nastalgic, or maybe they’ll self destruct. Given a better understanding of consciousness we could engineer around it like you say and make sure we create intelligence that only solves our problems, etc.
What about a mind upload? Certainly it is articficial, but is it a person? My answer would be no, but it is something equally if not more important than one. I guess I’m more interested in IA than AI. I want to be a part of it.
Be warned this list is incomplete, there’s more stuff out there on the subject
links for 2006-09-21
Smartspace – Smartspace – Waymarkr: Recording Life Real-time Designed for Series 60 cameraphones, Waymarkr captures images from a phone hanging in a pouch around the user’s neck and transmits the images to a Nokia Lifeblog, allowing a sort of…
[...] There is significant bias for the position that significant human brain enhancement, regardless of the means, will come before AI of human-surpassing intelligence. The reasons for bias are obvious – we as humans like control, we want control of our future, we’re biased against anything mechanical or machinelike, and it’s fun to fantasize about us personally taking advantage of the technology. We’d rather boost our intelligence “ourselves” rather than “relying upon” a “paternalistic god figure” to do it for us. All these tendencies translate to more attention and optimism surrounding IA (intelligence augmentation) progress, including predictions that it will arrive sooner – when in reality its arrival time will be contingent on a complex brew of regulatory, theoretical, and technical challenges whose ease or difficulty is orthogonal to our hopeful anticipations. Ditto with AI, of course, but because the notion of true AI is so fundamentally new relative to the idea of humans getting smarter, it has less (though still substantial) psychological baggage. The Web 2.0 crowd, in particular, seems quite enamored with the idea of boosting our intelligence through increased interconnectedness rather than centrally optimized solutions, as if every boost in interconnectedness is necessarily an improvement in quality of thought. [...]
7cThank’s.4e I compleatly agree with last post. zsy
ламинат и паркет 9e
[...] discussion is that only a small minority of the participants have even bothered to read the few documents and books in the field which exist, because the field is so new that in most places there is little [...]