AI
Artificial Intelligence, my take on a vastly misunderstood subject.
Our goal is to produce a living thinking robot with AI or maybe just "I."
Everyone...most everyone that is, is going about making a human, a robot, a replicate, a digital thinker with sight, feelings, knowledge and sound all backwards. First building a shell, making it look human then trying to make it act human is the last of the major steps in this creation. You have to start at the other end.
Even Mr. IBM intelligence is just a parsing and look-up machine. He makes great on-the-fly associations but lacks in the creative department. He does some learning but it's limited and probably doesn't correct itself.
To make an artificial intelligence you must first have an organism that responds to its environment. To respond it must first learn. To learn, it must have desires and anti-desires. An aversion to pain, starving, death, extremes of pressure, temperature, etc. is a good first step. As important as negative aversions are positive rewards. The first step is for our unit to know good and bad.
Currently, there is not a way to make a computer "Feel" threatened, in danger, hungry or cold enough to be uncomfortable. Until there is this feeling, there is no self-learning.
Take a single celled protozoa. Why does it tend to move away from stimuli that are injurious (too hot) to its well being and attracted to (food for instance) that which is supportive? Is instinct the hard wiring that we need to get sensory systems information to mean something? We can't program the learning (well not at first, we'll get to that later) but we have to program in the instincts. Darwinian theory shows us naturally honed instincts quite effectively. Those with "wrong" or organism non-proliferation concepts perished while those that had it right or close enough went on to reproduce.
While our goal is not to produce offspring its basic learning is second only to the hard wiring. There has to be a way for our AI unit to "feel" like something is to be avoided or that something else is "good." We don't have the time to first invent a way for computers to reproduce with sufficient firmware to warrant calling it instinct. If we could do that, the instinct to survive would be properly instilled. We may not recognize the code but through occasional random code alterations and billions of generations we might have something. Maybe some massively distributed computing power could perform that and create that initial spark of "something."
This something would be the beginning of the ability of these units to learn. They would instinctively know what would kill them because the ones that couldn't make this distinction have died off generations ago. The remaining units just know that there is good and bad and can start to learn to associate some things with good and others with bad. This is how they can learn. With sufficient sensory input and also sufficient ability to manipulate their environment they will advance asymptotically to their capacity.
Humans did something like that.
At first they will learn to tell what hurts and what feels good, warm from cold, pleasant from unpleasant. Associations will be made; warm is good, hot is not, cold is not. Hungry is bad, warm might mean food, pain hurts and is bad, curiosity an exploration might bring food which is good. Being still means nothing, making noise makes hunger go away...sometimes. The bottom line is that some things are good and some are bad. How do you reward an algorithm? If I had that answer I could create intelligent life out of sufficient computing power, sensory input, feedback and output. A robotic unit could teach itself how to walk because it would know that what it just did either worked or did not and that one was good and one was bad. Fine tuning would be a matter of finding out the limits of the successful / unsuccessful procedures and usually finding the optimum somewhere between those limits.
The unit must be able to feel pain in order to establish many of the physical limitations. Pain would have to be sensed and deemed bad in the form of electrical shock, heat, pressure, abrasion, radiation, etc. (anything that does damage, permanent or temporary). Pleasure would have to be known as well, having batteries charged would have to "Feel" good and be desired. Being in a non-destructive or optimally discovered temperature for greatest longevity and being in an environment conducive to further learning would have to be good and desired as well. Some of these would be instinct and also be modified by learned knowledge to be passed to future generations as improved instinct.
The programmed learning I mentioned before would be that knowledge, part instinct, part learned that is passed along. The degree of sensory sensitivity would have to be learned a new for each generation because their sensors, strengths and other capabilities would likely be optimum at a different point. We could pass on more than the instincts that mammals pass on.
So, how do we take that first step? Any ideas? Nano evolution? Where does the survival instinct come from to begin with? No wonder all of the plants and animals from 1 cell to the multi-cellular menagerie of earth are so similar, they started with the same cell that wanted to keep being, wanted to stay alive, wanted to survive while others didn't care.
I just have to collaborate with a few PhD's in AI and biology and physics and robotics and computer scientists and...and...and I hope I live long enough to see enough resources put to this task even though it may not see fruition in that lifetime.
I'm not talking about science fiction. Although this arena has been traversed often, its fundamentals have been set too high; so high that they are unattainable. Azimov's 3 laws of robotics should have started with:
0. A robot must be able to distinguish what is good and what is bad on all relative levels of existence and know to cherish one and avoid the other.
Then we have made artificial (would it really be so artificial?) intelligence.
Just my 2%.Wes Shaw
100186213
