Can a Machine Learn From Another Machine?

 Artificial Intelligence is a new frontier for machines. There is some hope that machines will eventually learn for themselves. There is a dancing robot, and a dog that walks. Atheists can fall to the trap of thinking that this proves that man is just a complicated machine. He might be, but this isn't sufficient to prove it. The hope is that Artificial Intelligence can learn to teach itself. For some people this is actually a fear. That would be a big step in arguing that machines are like humans. Because Christians doubt that humans are machines, they are uniquely suited to question this hypothesis.

For an experiment, take two Chess engines; start them at a reasonable level, and watch them play each other. If neither has been "trained," by playing against a human, they will play to a stalemate. Repeating the process does not improve their play.

This is because Chess engines are designed to start learning from scratch. Some Chess engines are programmed to never excel the human they are playing against, although the human can still lose, through bad play. But good Chess engines are programmed to learn the opponent, and eventually apply all the rules of the opponent more consistently than the human opponent can himself duplicate. Then the human "teacher" eventually loses. By adding teachers or letting the engine learn different opponents, the chess engine, on the steroid of high speed processors, can eventually beat almost any human. Only a human with new ingenuity can beat the machine. 

By substituting a well trained chess engine for the human, we can speculate if the "student" engine would ever surpass the "teaching" engine. I would postulate that this would only be possible by improving the CPU power, or changing the learning algorithm of the "student" engine.

What does this mean about Learning Algorithms and Artificial Intelligence? I postulate that it means that Artificial Intelligence can only reach a certain level of sophistication. It can only learn new things from a mind that can give it new problems to solve. It may even need a human to solve them.

My best suggestion for an avenue of advancement is to automate the rules from the book, "How to Solve It," by G. Polya, teach it an axiomatic set of algorithms, and see if a machine can prove a new theorem.

As a proof of concept, you could provide a set of axioms to the artificial engine; Counting, Addition and Subtraction, And, Not, OR and XOR, and see if it can learn multiplication from addition. Multiplication is not axiomatic.

Machines may be able to learn a great many things and complete remarkable tasks, but one learning machine may never be able to outperform human ingenuity, by learning from each other machines.

This is important to address the fear that machines will one day take over from humanity and exterminate them from malice. Only human malice can do that!

Comments

Post a Comment

Popular posts from this blog

A Question About Erasthmus' Sieve

Students of Mathematics are familiar with Erasthmus' Sieve. 2     x 3          x                4       x 5               x 6     x     x 7 8     x 9          x 10       x          x 11 12     x     x 13 14     x 15          x     x 16     x 17 18     x     x 19 20     x          x 21          x 22     x 23 First, we strike all multiples of 2 up to the square root (5.) Then all multiples of 3, then all multiples of 5, etc. I wrote a digital implementation of the Sieve, and used the index cr...
Read more

Notice of corrupted results: Vigenere may yet be found to be a "group."

Retraction: It is with regret that I do not delete the previous post. I am an individual, not an organization, and I failed to adequately double-check my results, but I am leaving the post to maintain some accountability. I am able to definitively say that the script for 2chars_from_file.py DOES NOT yield an exhaustive list. It develops the ciphertext for all 2 Char combinations of the FIRST 10, three character ciphertexts. This is not due to a logic error, but rather there is a virus that corrupts the indentation of the two write statements.  This may be verified by downloading the relevant code, and examining it, and running it if needed, against the relevant datafile. It is difficult to be exhaustively correct about ALL of the scripts. This calls ALL of my results in question, with regard to the qualification of a mathematical group. If Vigenere IS a group, the resulting substitution is still valid, but there exist 6 Char passwords that will accomplish the same encryption in ONE...
Read more

A Simple Rule for the Stock Market

The Stock Market is where the wealth of America is stored. As such, it is best when the stock market faithfully reflects the ups and downs of the economy. Stock market corrections are how the stock market reflects setbacks to the economy. This is good, as far as it goes, but in fact, when a stock market correction occurs, it is not well understood where the value of labor, goods and services stored in the stock market goes.  The whole idea that rising stock prices reflect improvements in the economy, is that this is the result of good investments. Investing is an act of devoting time, effort, or energy to a particular undertaking with the expectation of a worthwhile result. An investment is proven to be a good one, when it pays off after a duration of time. If you buy a stock, you wait for it to rise in price, and sell it at the appropriate time. "Buy low, Sell high," is how stock brokers explain their craft. How long should you hold a stock? This has always been up to the ow...
Read more