Monday, February 21, 2011

Formal Theory of Intelligence

A few years ago I tried my hand at a formal definition of intelligence by saying it's 'living representation'. I'd just been observing that there isn't a formal definition of intelligence anywhere. I'd also just formally defined 'life', another key concept that has no formal definition anywhere save for the one I gave it.

You see, intelligence may be just rote memorization (yes, even the Raven's progressive matrices) but there's more to rote memorization than just recording and archival of data with absolute fidelity. Real learning builds associations and is also lossy, entirely different from the way a computer stores data. And it just occurred to me now that 'living representation' is a good way to describe the 'compression unit' described in this paper describing a formal theory of creativity, fun and intrinsic motivation.

The compression unit's purpose is to ensure the knowledge base isn't drowned in an ocean of entropy (ie, waste information). So new facts are stored and then they're associated with all old facts and the whole is recompressed all over again, continuously. And that's how you get dimensional decomposition and natural language processing since they are just compression techniques!

Just a little insight inspired by thinking of idiot people incapable of coming up with proper definitions of key concepts in psych. Frankly, those psychologists who try to make definitions are so bad at it that they come up with the most ridiculous tripe unimaginable. So from that to my rather esoteric and seemingly poetic formal definition of intelligence, then to the relation between beauty and compression, and finally linking to what we know about how machine learning works.


Alrenous said...

Yes, good work.

Intelligence is a term conflating memory, reasoning, and creativity. The acquisition, expansion, and creation of information.

The life definition can be compressed.

"Living things can be assigned goals."

Any living things will defend its goal. The easiest goal to check for is simply the continued capacity to defend its goal.

Complexity considerations are unnecessary. If you have a machine that can disrupt a tornado, the tornado will never attempt to evade the machine or its effects.

Tornados will never respond to 'threats' because there's no such thing as a threat because they're not alive.

Richard Kulisz said...

Many living things CANNOT be assigned goals other than 'self-preservation'. And many things which definitely aren't living CAN be assigned goals.

That's the whole point of 'life is a complex thermodynamic system that preserves its own internal order'.

That is, life has no external goal, and its internal goal is rather trivial.

But more importantly, you're trying to "compress" my definition of life by amputating it of two of its elements. I state on the c2 page that life

* maintains itself
* uses energy
* is complex

you've amputated 2 of those 3 items, and the only item you've got left (maintains itself) is actually MORE wordy when stated your way then when stated my way.

In other words, you've got a shittier definition for the wrong thing.

Your argument against complexity is stupid. Humanity will never respond to the unexpected eruption of Yellowstone, but that doesn't mean our species isn't alive.

Humanity also wouldn't respond to time ending in our universe. But that doesn't mean it wasn't alive before it ended.

You need a lot more practice with definitions. A definition creates a border and you have to inspect BOTH sides of the border for counter-examples.