Re: Singularity?

Matt Gingell (mjg223@is7.nyu.edu)
Thu, 2 Sep 1999 15:22:46 -0400



From: Eliezer S. Yudkowsky <sentience@pobox.com>

>See http://pobox.com/~sentience/AI_design.temp.html [343K]
>
>> Where does your 2020 figure come from? (I'm afraid I
>> don't know how to expand the 'CRNS' qualifier.)
>
>Current Rate No Singularity. The 2020 figure is how long I think it'll
>take for AI researchers to think their way out of a cardboard box if I'm
>not running things. I'd *like* to have it done by 2005.

I skimmed your document and, with all due respect, I do not see that your model, as I understand it, differs significantly from classical AI. You have a number of modules containing domain-specific knowledge mediated by a centralized world model. This is a traditional paradigm. The macro-scale self improvement you envision is not compelling to me – if you’ve written a program that can understand and improve upon itself in a novel and open-ended way then you’ve solved the interesting part of the problem already.

Could you identify the cardboard box you think AI research is stuck in, and what you’d change if you were in charge. (You have 5 years...)

-matt