Artificial General Intelligence via the CogPrime Architecture (2012) [pdf]

goertzel.org

51 points·amirouche··7 comments

Nice, this is a work in progress of a book on building general purpose AIs. I worked with Ben at Intelligenesis/Webmind and years later I brought him in as a consultant on a medical AI project. To say that he is full of interesting ideas is an understatement.

A little off topic, but after the recent news of a deep neural network trained in a day to play expert level chess, I am starting to believe that Hinton's hack to train many level networks actually has some promise for general AI. Another few clever optimization hacks and faster/cheaper hardware might be enough.

> a deep neural network trained in a day to play expert level chess,

> Hinton's hack to train many level networks

Could you please post the links to that?

Probably this:

http://www.technologyreview.com/view/541276/deep-learning-ma...

http://arxiv.org/abs/1509.01549

Links to discussion here https://news.ycombinator.com/item?id=10239987

Buiding a strong evaluation function is only a small portion of what is needed to build a competitive Chess Engine. It is necessary to have a fast (and reasonably accurate) evaluation function paired with some form of tree search.

Not to belittle the achievement, but the neural network is still very far from being able to play `expert level chess`.

Looks like they have a LaTeX macro or something that's putting a newline after every instance of the word CogPrime in the intro, and a space after each in the main text. It's actually kind of distracting to read. I think the xspace package might help

https://www.ctan.org/pkg/xspace?lang=en

Ben's work isn't new to me, but I'm thrilled to see it get upvoted because it deserves a broader audience.

Reply on news.ycombinator.com