Wiki Interview With Eliezer/Support
Support
If an individual is interested in being on the Singularity Institute's AI development team, what knowledge and understanding should that individual have all ready acquired? What fields should be studied? What important books and papers should be read?
You should not plan on learning how to program while learning about AI - you should already be an experienced programmer. You should already know about object orientation and so on. Just, in general, be up with it. Know the idioms of many languages and so on. In terms of Cognitive Science, books and papers that occur to me on a first pass are: "Godel Escher Bach" (of course) and "The Adapted Mind" are primary. Browsing through the MIT Encyclopedia of the Cognitive Sciences would be good too. Other books that I like are Dehaene's "The Number Sense" (for an in-depth examination of one ability), Deacon's "The Symbolic Species" (developmental anthropology, a good successor to "The Adapted Mind" for evolutionary theory(!)), "Levels of Organization" is obviously very important, as is the section on "Time and Linearity" in the old GISAI paper. Barsalou's "Perceptual Symbol Systems" is recommended. You should read something by Christof Koch on computing in single neurons (there are some good papers online) Lakoff's "Women, Fire, and Dangerous Things: What Categories Reveal About the Mind" and Lakoff & Johnson's "Philosophy in the Flesh" I'm currently reading Kosslyn's "Image and Brain: The Resolution of the Imagery Debate" Lenat's original papers on Eurisko are excellent reading if you can dig them up (a friend sent me Xeroxes) Hofstadter's book "Fluid Concepts and Creative Analogies" has some important essays, one of which (Hofstadter, Chalmers, and French IIRC) is online at http://www.cse.msu.edu/~cse841/papers/hofstader.pdf Read something on minicolumns by William Calvin, and find something to read on cerebella microzones. Read something on the circuitry of the visual cortex - Googling on things such as "retinal ganglion", "lateral geniculate nucleus", and "koniocellular" should get you more interesting pages than just "visual cortex". Another very interesting trick is to take some of the more prominent names from *different* fields of Cognitive Science, or the more important names, such as Lenat, Lakoff, Rosch, Hofstadter, Kosslyn, Deacon, Chalmers, Calvin, Barsalou, and look in Citeseer or Google for papers that reference various combinations of those authors, which often turns up truly fascinating stuff. Then, if you can, you look up the authors or papers which *those* authors reference. That was how I originally found out about Barsalou, but I guess I'm sort of wandering off-topic now. Key topics in Cognitive Science are: evolutionary organization, neo-Darwinian population genetics, development of the hominid family, mental imagery, sensory modalities, neural circuitry in sensory modalities, neural circuitry in the cerebellum, any neural circuitry we've studied enough to make interesting hypotheses about. Computing in single neurons, post-Aristotelian theories of category structure, such as Lakoff and Johnson, cognitive psychology of category structure, such as Rosch and Berlin and Kay, although I admit that I've been relying on secondary sources for my knowledge of that particular topic, "basic-level categories" should turn up interesting stuff on Google, cognitive psychology of decision theory - you've already seen some of that, and then there's everything else.
Oh, I forgot to mention Melanie Mitchell's Copycat program - the source code is available online. Read it. Also another key name to search on is Marr, although everyone mentions him. More literature. Functional neuroanatomy - studies of associated deficits. You can learn a lot about the mind by studying how it malfunctions. Also scary stuff such as blindsight, agnosia or anosognosia, split-brain experiments, and other phenomena that challenge our intuitive theory of identity. Also Evolutionary Psychology as-seen emotion and morality, *also Evolutionary Psychology that deals with emotion and morality, i.e., the stuff that the media distorts in print. If you're still confused about consciousness, read Chalmers's "Facing Up to the Problem Of Consciousness" so you'll know why you're confused.
(also see So You Want To Be A Seed AI Programmer)
Why do you consider those to be key topics?
The reason that all of these things are "key topics" is because they say something fundamental about the foundations of form in the mind, usually something that contradicts the dogma of traditional AI and the intuitions of non-AI programming, because they provide detail that fills in the very broad picture sketched in "Levels of Organization", because if you look up all of these things, you should be able to see something of the interacting processes on multiple levels of organization within the mind.
To help achieve SIAI's short-term and mid-term goals, what kind of activity and support is needed?
What we need, of course, is enough funding to hire a programming team, some more people to hire, volunteers to produce Singularity literature, and ideally enough funding to hire an Executive Director with previous work experience. I tend to see funding as the bottleneck.
Has SIAI's work garnered interest from academic or business research and development? If not, what do you think is needed to change the situation?
Not yet. Hopefully DGI might get more attention academically, but then again it might not. As for business research and development, I would expect Flare to be the interesting part of that. I do not think it would be a net benefit to slave the seed AI project to any design requirement that is not seed AI and that needs to be done within X months, no matter what resources become available as a result.
