Artificial Intelligence

Okay, so, I just thought of something and I wasn't sure where else to put it.

As I understand it, our bodies have a natural reward system set in place by evolution. That is, the serotonin and dopamine systems.

Everything we do as humans we do because we believe it will garner us the most amount of dopamine or serotonin from our brains. Until we discover a way to artificially create dopamine and serotonin without bad side-effects, we basically (Mostly unconsciously) follow the programming in our brains that will result in the highest output of S&D. This is basically how our brains have programmed us for survival and society. Anything that is beneficial to our survival gives us more S&D. This is why food is tasty, why sex feels good, and why people feel good after exercising. Conformation with society is also rewarded with S&D often, or at least approval from our peers. They're all things that would be beneficial to our early ancestor's survival, and the programming has been passed down to us.

At least, this is how I understand it.

So, here's my proposal:

For artificial intelligence, introduce a similar reward system. Of course, instead of S&D, have it just be some signal (I'm sorry, I don't know programming jargon all that well.) that acts as a "pleasure" signal, to reproduce the pleasure we'd get from S&D. The AI is constantly seeking to acquire as many of these "pleasure" signals as possible, and will do whatever he thinks will get him the most pleasure signals, just as humans. We can then program various behaviors into the AI to produce pleasure signals. The AI will then act on these behaviors to maximize the pleasure signals he gets.

Does this make any sense? I'm sure it's mostly wrong, as I know little about biology or artificial intelligence, but I wanted to see what you guys thought.

Yes, it does. The general idea makes a lot of sense, and is used in a lot of areas in AI. For instance, neural network training is based around the system being rewarded for minimizing some error value (or punished for maximizing it). The problem is that to work, it needs to be able to recognize what needs to be rewarded so you don't have to constantly hover over it with a reward button in your hand hoping it'll know what you're praising it for when you press it. It also needs a good system for altering its behavior to get the reward without taking forever or being stuck in a local maxima.