On Sun, Oct 26, 2014 at 6:06 PM, Tim Tyler via AGI <[email protected]> wrote:
> Maybe this indicates irrational behaviour - but I think it is more
> likely that software really is the problem.

As I said, software is one part of the problem. The other parts are
hardware and knowledge collection. According to my cost estimate
(which I think everyone has seen), software is the least expensive
component of automating the labor force. At 300 million lines, it can
be written for $30 billion. But if you are trying to just simulate a
single human brain (or a robot, perhaps), it is the most expensive
component.

My estimate is based on the information content of the human genome.
The interesting thing about this is that the human genome is not much
bigger than that of an insect. Thus, the major distinction between
"insect level intelligence" and "human level intelligence" is
computing power.

A lot of people think that there ought to be a really simple general
learning algorithm. Perhaps it is a deep, hierarchical neural network
or something like it that could be applied to vision, hearing,
language, motor control, game playing, or whatever else you throw at
it. First, Hutter proved that the optimal solution (AIXI) is not
computable, and that approximations like AIXI^tl and Goedel machines
have exponential time complexity. Second, Legg proved (
http://arxiv.org/abs/cs/0606070 ) that good predictors (how we
normally measure intelligence) are necessarily complex.

I don't expect that everyone will accept these proofs. After all, we
still have people trying to recursively compress random data, build
perpetual motions, and prove P = NP. But that doesn't mean we should
give up on AGI. The payoff would be enormous. I do claim this is not
something you are going to solve at home on your laptop. Big companies
like Google and Facebook have the resources to solve this, and they
are doing so. What Hutter and Legg essentially proved is that your
brilliant insight is not going to beat hard work.

-- 
-- Matt Mahoney, [email protected]


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to