Understanding Human Mind before Creating Super AI
There are many Artificial Intelligence classes all over the country that have a curriculum that simply teaches various algorithms for solving puzzles and different problems. These are solutions to problems with fixed domains, they are brute forcing the answer through calculation, not solving it like humans. While the human brain is very unpredictable and behaves as one entity with different neurons shouting over each other, electronic AI can simulate the human mind extremely accurately.
The first step that many AI enthusiasts miss, is understanding the natural algorithms of the brain itself.
When a computer is programmed to play a chess game, it makes the best decision based on cold calculation and an estimation of the score after several moves ahead assuming the human is the best player.
When a human is playing a chess game, he/she decides the best next move based on the current situation, perhaps to better position itself for later stages of the game, perhaps a plan is formed to capture a certain valuable piece through sacrifice or trap, and perhaps a plan is formed for a quick checkmate. The computer has the advantage because of the amount of calculation and the database full of expert patterns and expected responses. Yet, humans on the internet team up together and defeat extremely powerful chess AI programs, proving once again that the human mind can be better at chess if trained properly.
The natural algorithm of the brain seems to apply various problem solving procedures to situations that have large amounts of unknown variables. Programmers have attempted to mimic this and call it a general problem solver, but have so far been overwhelmed by the amount of unknown variables when applying the same process to different situations.
The reason human minds are more powerful than computer AI, is because they've had decades of training from childhood and a natural knowledge and pattern database that has been collected since childhood. This is why AI has not become as advanced as in the movies.
In order to do this, we must first understand the steps our own minds take when approaching a problem. Firstly, how do we recognize it's a problem and begin the process of solving it? How do we arrive at the solution, do we search for the solution, do we use our past experience, or do we experiment?
The idea must be that the AI program itself must be able to modify it's own sort of algorithm based on experience and experimentation.
Therefore, it is evident that mimicking the human's capability to learn, understand, and develop it's own algorithms within itself is what AI must focus on. One solution would be for AI to understand natural English language and then surf the internet to learn at an incredible fast rate.
The problem is, the saying "Don't believe everything you read on the internet." How does a program decide which contradictory statements are true. Does it take a statistics of all the statements and their results? Does it interpret which sources are more credible? Does that mean an AI in the future would be susceptible to cults or conspiracy theories?
The Other Problem and It's Solution
The other problem would be that mimicking the human mind becomes close to impossible and even if successfully mimicked, the flaws will also be repeated.
Therefore, the best solution must be a hybrid in which the computational power and accuracy of the computer is used alongside the unpredictability, creativity, and learning capabilities of the human.
As in one X-Files episode, AI must first be adapted to behave like small animals, like spiders, then development must be made into understanding human concepts.