On genetic algorithms, decision trees and future
I’ll start with the future.
It’s unknown, it can’t be totally predicted. But, and it’s known for sure it could be changed by actions of people and natural events. Huge computers, then and now, are trying to forecast the weather. Neural-network-based applications are guessing the stock prices. What’ll happen - inevitable, and we don’t know exactly what’ll happen. Yes, we can guess and build plans for the future, but in real life, there are always surprises. And I can’t say that I dislike them :)
Soon enough I’ll say goodbye to Haifa University, and it’s totally going to happen. The question is what to do next: Master degree, finding a job, marriage. Or it’s better to stay where I am and in the meanwhile work on some project? Or to found a start-up company?
When I tried to decide what’s better and what to do, I used the technique called “Decision trees”. It’s usually used in business to determine if risk worths the profit or where to invest the capitals. And, yes, it works based on the prediction - the chances of success are predicted by human from his experiences, risks are guessed roughly, but nevertheless, this technique shows good enough results.
Well, even on this small amount of choices I worked for a day and still didn’t got it - what’ll be the best thing for me. May be I could use friends’ experience? There is only question: If what’s good for them will be good for me too?
Let’s pretend that I’m not so special, and what works for most of us is good for me. Computers are known to build decision trees well. Let’s take data on a lot of people and let the algorithm to make this tree. Good enough? Looks right.
Not so, because everyone is different. Two persons that seems very similar to each other may have points that’ll make them kind of opposites. Actually, no one can predict(again!) what’ll be good to person. Also, we can stuck in cases of overfitting and underfitting, and it’s like saying that you have to be exactly Steve Jobs to be successful, or that only Jewish people are successful.
And now let’s go to the Genetic Algorithms. It could be pure fun to create several copies of this universe and myself in the center of it, and let them live for some time. And to see what’s going if everyone of them chooses different path :)
I like genetic algorithms, they are pretty good for the job, but the situation needs the criteria of success to be defined well, and so the fitness function. Looks, like it can’t be done too, in any means. It’s like to find the meaning of life, universe and everything, and I don’t think, that 42 will be enough answer.
Looks like I’m getting myself in endless recursion.
Anyway, we still don’t have enough calculation power to simulate all this :)
P.S. It leads to building computer that simulates another computer, that in it’s turn…