[LOCKSUIT]

Remember back to the famous robot that learns to hop, and improves.

Yous obviously understand how.

But how do you imagine the machine/algorithm?

In a bad poor way?

Or like this?
http://advancessss.deviantart.com/art/New-Canvas-650853524?ga_submit_new=10%253A1481560495

Is this what you see? Or have you been seeing it as code in a crazy vertical fashion? That would fog your view of AI. To advance greatly means to clearly see it ALL!

In fact, korrelan, kei, and freddy, I advise you to create a very neat and colored-in schematic of your AI. I was astounded once I started to see all my writing as vision of a single small image. Without a image, you are very far behind from where I am. It's a must have.

[LOCKSUIT]

Do any of my advanced AI men here have a image showing neatly most or all of your AI?

I would also really like to hear Art's/korrelan's/etc's reply to the visualization above I originally asked about too, you can see a example in the link.

[Art]

Sorry for not responding sooner Locksuit.

You have it structured so that:

Input = Reward > Sense > Action = Output
Input = Generating > Action = Output
Input = Sense (sensor) > Action = Output
Input = Action > Output (bypassing other parameters like Sense, Reward, Generate)

Much like a condition / response - think Pavlov's Dog...bell ring > salivate

Simple but Interesting drawing
At least this is my observation.

[LOCKSUIT]

The way the blue image I drew works is like this:

1) Input, goes to Generator, to Action, to Output.
2) Input, goes to Reward, to Sense, to Action, to Output.

This allows it go generate and learn actions, then select senses and use actions.

[infurl]

As usual you left out the most important part. The part with the magic in it.

[LOCKSUIT]

You mean the sensing ghost the machine creates unto the universe?
Or CNN?

Just trying to help you guys cus I know you's don't have a neat schematic and can't see your AI all that easily.

[infurl]

Just trying to help you guys cus I know you's don't have a neat schematic and can't see your AI all that easily.

You don't know that at all. I use graphviz to automatically generate schematics from my work. Instead of wasting time like a bunch of schoolies competing to draw the meanest looking spaceship, some of us are actually doing something useful with our time.

[keghn]

Decomposing Tasks like Humans: Scaling Reinforcement Learning By Separation of Concerns: 
http://www.maluuba.com/blog/2016/12/9/improving-scalability-of-reinforcement-learning-by-separation-of-concerns

[infurl]

Decomposing Tasks like Humans: Scaling Reinforcement Learning By Separation of Concerns: 
http://www.maluuba.com/blog/2016/12/9/improving-scalability-of-reinforcement-learning-by-separation-of-concerns

This is just a rehash of the subsumption architecture which has been used in robotics for decades. I wonder what else they're trying to reinvent because they haven't done their research.

https://en.wikipedia.org/wiki/Subsumption_architecture

[LOCKSUIT]

Show me your image of your AI then.

I bet it is not as organized as it could be.

There is no program that will organize something particularly right for the specific project. Things don't work like that. That's like saying you have text, and it knows from the text that the BEST way to place the units is like this cus um er LOOOOOL. That itself is intelligence. Graphviz can't do it for you.

[Art]

Decomposing Tasks like Humans: Scaling Reinforcement Learning By Separation of Concerns: 
http://www.maluuba.com/blog/2016/12/9/improving-scalability-of-reinforcement-learning-by-separation-of-concerns

Isn't that what Beethoven is doing?

Decomposing! 

[LOCKSUIT]

When you understand AI is that blue pic I drew how you see it? Or do you see code/neural-connections? Or hear words and no image visualized?

When you think about how the famous stick-creatures learn to run, is the blue picture I drew basically what you see? Or code/neurons/words.

[infurl]

http://www.adampease.org/OP/images/SUMOclasses.gif

[LOCKSUIT]

Who made that?

[infurl]

Who made that?

You're kidding me. Can't you figure it out from the URL?