Crackpot Layman Theories on Cognition and Artificial Intelligence: Part I

I was watching a show on PBS (Charlie Rose) and a gentleman was being interviewed about his thoughts on physics, the brain and they touched upon artificial intelligence a bit, and as usual my mind wandered on a tangent about those subjects.

He said that our knowledge of the brain was like taking a computer, smashing it with a hammer then trying to figure out how it works by studying each component individually. That made me think of how Japaneses kanji can express an entire concept with a single character.

Perhaps we’re thinking about cognition and intelligence as if they can be developed through combining ones and zeroes, or by putting together individual letters to form words. What if a single neuron can contain a complex thought all on its own depending on when and how it was “triggered” in conjunction with other parts of the brain?

Then I thought about how we recognize objects or sounds. We only know from previous experience. How do I know that object over there in the living room is a coffee table? I can recognize it by its shape, size, color, texture and other attributes. I can render a model of it in my mind, not only of its physical properties but also of its significance based on my experience.

I’ve heard the term “learning computer” tossed around in a few different contexts. I’m thinking that a program could be taught about shapes, then through sensors be able to recognize those shapes. Then it could render a physical model of an object that it sees based on its proximity and determine its function or significance based on that physical model and its experience, just like we do.

From what I’ve seen of current computer cognition/AI work, it’s as if most research is taking a brute-force approach to recognizing objects rather than giving the program a “mind” that can take sensory input and create a model and derive function/significance from that.

For instance, if you feed a computer a flat picture of a coffee table, current programs can do a decent job of recognizing the same coffee table or similar tables – as long as the picture is taken from a similar angle. What if you instead gave the program a 3-dimensional model of the table? What if you then showed various objects being placed on it? You would be providing a mental model of both the object and its purpose. And recognition could be based on matching shapes with the mental model. Of course, the program would need to be able to render a mental model of the table by using perspective, then match that to the pre-loaded or “learned” model.

Of course these aren’t ground-breaking observations or thoughts, but I just had to get them saved lest I forget. Stay tuned for more miscellany.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: