Problem Solving in Machines

[fluffy]

Extensively one would need a problem to solve.

Also I think that you would need to understand it.

Those two things don't match well in today's machines.

Instead I came up with this:

A combination module (planning)
A memory module (resources)
A implementation module (a body)

Putting a round peg in a round hole is simple for kids.

Math is harder.

So an algorithm for all three modules must have a digital environment.

Most ideas we have are high dimensional. In laungauge and visual spatial - it's recursive.

The body sets limitations.

Memory is so you don't back track.

In practice it is like the mind stone from avengers.

Memories connect to each other multiple ways.

Like the pegs and holes.

Either they fit or don't.

Thinking is about connecting them without physically doing anything, then you remember.

A big network of memories exist giving lots of things to try.

A reward function might be needed but simply to gain solutions by socializing. Helping people.

I have a picture. Need to find it.

 

[fluffy]

Visualizing it would be different in 3D.

Its just a hierarchal graph.

 

[fluffy]

Memory self organization depends on model constraints in an environment.

Thinking without combinatorics explosion.

Only partially hardwired.

-

Need some base metrics.

 

[fluffy]

This is the new design I came up with in the hospital. It uses selective attention and working memory.

It looks like a balloon.

 

[birdsnestfern]

Is it about selecting between two choices or what does it do?

 

[fluffy]

Is it about selecting between two choices or what does it do?

You can only pay attention to a limited number of things at once.

So what it does is simulate what might work by referencing the outside world with its internal memory.

So if a choice is to be made it has to think about what each option represents in relation to their values.