Re: rambling question on AI

From: Eliezer S. Yudkowsky (
Date: Fri Jan 04 2002 - 22:50:52 MST

Reason wrote:
> Why? It takes four months, five developers and a quarter of a million
> dollars to produce software that retails for half a million, including
> support. In most cases, this is the compromise version -- the full (and much
> more complex) vision isn't realized until a few years of new versions down
> the line. So the argument would be, if someone can dedicate resources to do
> that for another field, what is it about AI development that means you can't
> produce something that can be sold in the same timeframe?

If AI development were not qualitatively more difficult than a
*programming* task, the Singularity would have occurred long since (the
early nineties would be my guess).

> With a broader view of the term "product," alternatively, you could produce,
> say:
> a) knowledgable consultants to farm out to other AI ventures
> b) a single useful component that can be licensed to other AI ventures
> c) a book on AI development
> d) the material and contacts for a subscription-based newsletter for
> development companies
> And so forth. Not that I'm convince there's a vast market of AI ventures out
> there, but I don't think that there's anything wrong with taking those
> directions.

If we'd wanted to go that route, we could have done Flare as a for-profit
venture. And if we'd wanted to go *that* route, I have a half-dozen good
ideas for starting a for-profit corporation; they just happen to be
totally unrelated to AI.

But there is a saying: "Life is what happens while you are making other
plans." And life has taught me that if you want to avoid this curse
coming true of you, you had better avoid doing ANYTHING that does not
DIRECTLY advance your long-term goal (in my case, the Singularity). If
I'd known this when I was sixteen, I would have been able to put in an
additional two years or so of Singularity work by now.

And furthermore, we don't dare run the risk of building a *good*,
*profitable* product in a for-profit company. This could easily be worse
than building a bad product; "life is what happens while you are making
other plans" would very quickly happen to us, and the "interim product"
would become the company. A spinoff of a real, direct-to-Singularity
development effort is one matter; a genuine spinoff can be handed over to
new programmers while the core development continues. Building something
easy and safe that isn't what you actually want to build, as your core
development effort, is much more dangerous a distraction.

Furthermore, no matter how good an idea I have for a for-profit
corporation, it still isn't going to have better than a fifty percent
chance of success, because the ordinary failure rate for startups is
ninety-five percent. If I still thought that the Singularity was located
in 2030, the way I did when I first got into all this, then I might be
more inclined to try and start a for-profit corporation to get the funding
et cetera, because there would actually be the time to take the company
public and sell the stock and turn it into a private operating foundation,
or to try again with a nonprofit if the for-profit route failed. As it
is, the possible horrendous time crunch only confirms the recommendation
of my experience so far, and that experience is this: working directly on
the Singularity is the only thing that got me anywhere in life, while
working on other things in the name of "practicality" and "doing something
for the short term" just wasted my time.

-- -- -- -- --
Eliezer S. Yudkowsky
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 13:37:32 MST