>Chris & Jessie McKinstry wrote:
>>
>> 2 - The primary purpose of GAC is to build a fitness test for humanness
in a
>> binary response domain. This will in the future allow GAC to babysit a
truly
>> evolving artificial consciousness, rewarding and punishing it as needed
at
>> machine speeds.
>>
>
>Chris, if you are actually planning to use near-brute-force techniques
>like evolution, I am curious as to what kind of computing power you
>anticipate will be required to evolve a large population of baby AIs.
>Clearly it will be much more than the amount required to run a single
>human-level AI, especially if you want to be able to evolve the population
>in any decent amount of time. Are we talking 2030 or farther before this
>amount of computing power will be available? If so, what's the point of
>hyping now what will not be technically achievable anyime soon?
>--
>Brian Atkins
Brian, I think the problem is actually much simpler than most people think,
because most of the complexity of the brain is involved in fighting the
second law of thermodynamics. Given the right training data, neural networks
much simpler than the human brain can actually out perform it (see:
http://www.wired.com/news/technology/0,1282,31644,00.html) I expect this
will be the case with training a neural network with the Mindpixel Corpus.
Aside from that, it's important to get as much data as possible. Nothing
happens without the data. Which is why nothing interesting has happened in
AI until the advent of large corpi that have allowed us to make statistical
attacks on classic problems such as machine translation and speech
recognition.
A big corpus makes a big difference!
Chris McKinstry
Director, Mindpixel Digital Mind Modeling Project
http://www.chrismckinstry.com
This archive was generated by hypermail 2b30 : Fri Oct 12 2001 - 14:39:41 MDT