[agi] Purpose problem.

Wed, 27 May 2020 07:26:00 -0700

Speaking of Architecture,
speaking to Immortal D.

What is in your architecture to deal with the purpose problem?  Note: I am not asking to start a discussion about "friendly AI."  What I want to know is how the heavily neural net construction of Immortal Discoveries deals with the task of picking a purpose.  

A reminder/review of why I consider there to be a purpose problem:

Purpose problem

There is a “purpose” problem with AI. The problem is not a lack of purpose, but rather the discernment to select a purpose. What shall the AI “aim” at? Where will it be going? How often will it change goals? Will the purpose be noble? Will it be practical? Will it be self serving?


What is the units purpose?

to be nice?

to be helpful?

to make money?

to build an empire?

to conquer?

to assist humans?

to babysit the world?

to make big decisions?

to correct disparity?

to show the better way?

to cook up specialties?

to advance technology?

to relieve suffering?

to demonstrate the difficulty of being alive?

to advance humor?

to contribute to poetry?

to write the best book?

to create virtual realities?

to build Dyson spheres?

This purpose problem could also be called the “will” problem. What will the AGI aim at?

Matt recently said

"By free will, I mean what it feels like to make decisions. The feelings train us to believe that our actions are somehow distinguishable from the output of a deterministic neural network."

I submit that "will" be it free or not, is the underlying logic that determines what we do.  And, that it has little to do with what we feel.  Our logic is likely subconscious.  The effort we make to police and clean up our logic has much to do with "owning" our will and owning our actions.

Stan


Reply via email to