[Ben.F.Rayfield]

http://humanai.net/thread/thought-experiment-ask-self-is-machine-on

Choose any machine you’ve seen many times which has a small light on when the machine is on. You walk into the room, look in the general direction but not straight toward the machine, and ask yourself “Is that machine on?”

What is your mental process to learn the answer, and how did you learn that process?

First you find which area of your vision the machine is in and look for the part where the light will be on or off.

Then you hook that part of your visual neurons (brain cells), which will see bright or dark, to the neurons asking the question. How do they get connected that way?

You can now choose to imagine what the machine would look like, in that specific view from where you are standing, if it were on or off. You can write bright or write dark onto those neurons. In computers we call this an Address Bus, a way of copying information between parts of the system in either direction, reading or writing. But it doesnt work exactly the same as a computer, just the general idea of choosing where information flows.

How did you learn the process that answers if the machine is on or not? How does that hook into a command to yourself like “If the machine is off turn it on”?

This is not a text based process. There arent variables like “on” and “off” in our minds. That only comes in when we get confused or we’re talking to others about it. We understand the machine being on or off visually by its light in a part of our vision that depends where we see other things normally found near it.

[ivan.moony]

I think that human intelligence is consisted of two parts. The first part would be something like instinct, a set of algorithms that fastly creates visions in our mind, but just a basic ones, like recognizing shapes. This would be unconscious part, dealing with instinct.

On top of instinct would be intentionally controlled train of thoughts, something that could be called intelligence. Like you don't recognize what you see until someone tells you what it is (or you check is it precisely something).

[Don Patrick]

It is basically a if-then rule that is most often learned through observation: whenever the machine worked, coincidentally something on the machine lights up, and when the light wasn't on, the machine wouldn't work. The consistent coincidence of these things suggests that the working of the machine depends on the light lighting up, or vica versa. Alternatively one could have been told this rule or have superimposed the rule after observing it in comparable machines.
Rules are not textual processes in a human brain, but can nevertheless accurately be stored in the form of text, as is done in an instruction manual.
Oh never mind, I overlooked the part that said "thought experiment".

[ivan.moony]

You can do a little experiment this moment with this screen. Try to change point of view inside this screen every 2 seconds. You will experience recognition of different words in short intervals. I think that that phenomena is fully automated and is stored in instinct. Recognition (I think) is realized on unconscious level, while moving eyes is the conscious part. This recognition reminds me of neural network recognition, although I know very little about neurons and didn't investigated them enough.

I think that this unconscious recognition can be applied to different levels of experience, like from shapes to letters, from letters to words, from words to sentences and finally from sentences to an image of isolated system in the Universe, all with the same algorithm.

It might be a good exercise to try to implement just unconscious recognizing as an block diagram and label it as "recognition instinct". It should operate on 1D data + time (like hearing) and 2D data + time (like eyesight). Then we would could just focus a machine sense to a bunch of data and let the instinct do the magic of learning and recognizing. Machine's focus would then be controlled by some higher behavior mechanism which should, beside this learning, implement some "action taking" mechanism too.

[ivan.moony]

A machine could also instinctively learn what is next expected from it, more learning (input), or some action (output).

Or it could learn until it recognizes what output is expected from it (then the process repeats).