AGI Workshop Videos online + briefs Wednesday, Oct 4 2006
Information about the last AGI Workshop (the one I was talking about earlier) is now online at AGIRI’s website here. The first two segments, “AGI versus Narrow AI”, and “AGI – What are the Risks?” If someone would transcribe the panel discussion to text, it would be helpful. We had an interesting conversation on the risks of AGI, whether or not a kind human would be a good model for a Friendly AI, and so on.
Briefs:
Netflix is holding a $1 million AI contest. It is strongly reminiscent of the Hutter Prize. Instead of predicting Wikipedia text, this time the AI attempts to generate recommendations that have a high correlation to the user’s actual likes and dislikes.
CNN is reporting on a musical robot that composes, performs, and teaches.
The White House has their nuclearphilic energy plan online for you to see, including links to Bush explaining why we need more “nuk-yuh-lur” power plants to push the economy forward. We need to build thorium nuclear reactors to eliminate the impact on our environment.
Twelve teams have been selected for the DARPA Urban Challenge.
Word on the street is that mobile will be big in 2007. Watch that space.
Over at Lifeboat we’ve changed around the site design significantly… let us know what you think.
Brian Wang, and others, are starting to think about the risk of space rocks being used as kinetic weapons. Using accelerated rocks as weapons has a remarkably high damage-to-effort ratio – not sure if this will be a risk in the timeframe-that-matters (pre-Friendly AI) zone, but maybe. In the post-Friendly AI-zone, the problem’s already solved, because you’ve got a self-improving human-friendly superintelligence that can take care of it.
The Amish are pretty forgiving… I guess it’s possible for non-vengeful intelligences to theoretically exist, how amazing.
Iran is being kinda weird, asking France to oversee enrichment. Basically, here’s their deal: they want nuclear power, and we don’t trust them not to nuke Israel. Everyone should have the right to nuclear power, but no one should possess nuclear weapons. The only way to ensure this is to have a completely transparent program, which many leaders are reluctant to grant, but it must be done, including for the USA!
Americans are ignorant about science, big surprise. This time the survey was about water. From the page: “Fewer than half of those surveyed answered correctly when asked how much of the planet is covered in water (about three-fourths). Hardly anyone knew how much of that water is drinkable (about 1 percent).” If you ever wondered why there weren’t more transhumanists already, there’s your answer.
People all over the Bay Area are protesting against Bush. For those in other parts of the US, take note: one of the most economically productive metropolitan areas in the world is also that most against Bush.
Humorous article over at WIRED on how to dismantle an atomic bomb.
If you are unhappy about Bush then try moving over here to Europe. You would love Sweden (where I live) or Norway. Half socialistic, no jobs, no innovation, endless sensitivity training, just talk, talk, talk without action. Just hate for everything American and most of all Bush. It is ugly.
I think you Americans should be happy for having Bush, even if some of his policys are not to my liking either, he’s overall a good man with back bone and trust in his own people.
Many people all over the world, especially I, would give up a lot to become American citizens.
Sorry for being political but you started. ;)
Eliezer said “I’m not trying to create god…I don’t want to create something that will tell us what to do…just something that will set up better background conditions for us to work and play.” This quote was a revelation for me. Eventhough I have read essentially every document linked to from the SingInst website (albeit, perhaps not as thoroughly as I should have) I always thought that an AGI as conceptualized by Eliezer would go beyond ridding the world of obvious evils to creating a utopia. This was the one sticking point I had with Eliezer’s theories; I was uncomfortable with the prospect of a friendly AGI creating a utopia, because I thought it would be based on what the majority consider to be utopian, rather than what I personally consider to be utopian. Perhaps Eliezer should emphasize that the AGI won’t impose itself too much on humanity. It will simply optimize the conditions in which we live (e.g., no more death or aging, no poverty, etc.).
Have you read Coherent Extrapolated Volition? That document merely describes creating an output from a synthesis of extracted human input, and does not imply the forcing of a utopia or invasiveness in any way.
Unfortunately, the difference between eliminating the bad and boosting the good is academic and impossible to define. No more death or aging, no poverty, etc., WOULD BE a utopia. If you can’t define the difference in terms of math, it doesn’t exist to a Friendly AGI.
And believe me, if we built a Friendly AGI correctly, it would be perfectly capable of creating a utopia that everyone enjoys, even if you are skeptical today. Just because humans use the prospect of creating ‘utopia’ as a poltical ruse to gain power, doesn’t mean that 1) it isn’t actually possible, and 2) it isn’t desireable anyway.
Problems with conceptualizing Friendly AI appear when we introduce anthropomorphism and also overestimate the conditions necessary to please humans. With limited resources and intelligence, pleasing humans is difficult. With unlimited resources and intelligence, pleasing all humans becomes much easier – the problem is selecting among the billions of possible ways in which you can do it.
I recommend reading CFAI again and noticing that even with all the talk of Sysop, a pure altruist boosted to superintelligence would be perfectly capable of pro-actively creating a better world without being annoying about it.
http://singinst.org/CFAI
Unfortunately CFAI is obsolete, but it’s worth reading because it’s considerably more advanced than what most other people have to say about AI safety.