Have any attempts ever been made at having an evolving evolution algorithm?
The problem though I guess is that evolving a program would be hard, how are you supposed to evolve something which has to be very carefully constructed to not crash and burn?
>>58702676
Do you mean life an AI life sim? There are quite a few, but they're toy programs. It's not going to evolve into the matrix.
>>58702676
>evolution evolving
>algorithm
Just keep breeding until it leads to the longest lifetime, remembering all errors of past generations.
>>58702737
BR Ratio of 1
Welcome to Humanity. Are you enjoying purpose/being alive?
>>58702737
Evolution only requires the ability to reproduce. Long lifespan isn't a necessity, and neither is intelligence.
Maybe it'd be good to try evolution by making algorithms which are allowed to modify themselves duel, in some manner that forces them to be neither stagnant nor purely random.
What if they were each given a pixel they could place/colour, and they could bet points on how closely their pixel would match the other's?
It wouldn't be iterative, the pixels would always be there and free to change, and as long as the bet was there the computer would keep gaining/losing points.
It's an interesting game, because it's so simple, but it's extremely complicated to play.
I can hardly think of a better game to train a neural net on.
>>58702676
Shark blowjobs
>>58702676
Create a set of code which can be used to suite new types of functionality for new situations then program the algorithm to utilize and modify them as the situation demands, allow it to grow(store memory from past actions) until it is aware of what it is.
Then you *could* have something which resembles intelligence on a smaller scale.
>>58702752
back to >>>/sci/, simon
yes it's called a genetic algorithm or genetic programming
>>58703221
Quaternary code>binary code
>>58702676
Evolution algorithms exist, but you have to make a heuristic of what desired behavior is and what to modify.
Take humans:
Our heuristic of a good human is how many children we have.
We have many parameters, so we can solve a lot of tasks that doesn't involve breeding.
But computer programs are very hard to make.
You cannot say you want to solve any problem using any method and then use any parameters to solve it with.
It gets very complex.
What you can do is take a specified problem, make a heuristic for that and then optimize for that heuristic.
Say you want to play a game. There you can evolve strategies and stuff as the goal of the game is easily translated into a heuristic.
But if you don't know what you want to solve, you don't know what is good and then you can't just crunch numbers until you solve something.
>>58703391
But the whole point of evolving an evolution algorithm, or having the program evolve itself, is that if you suddenly changed the problem it was being tested on it could possibly adapt far faster than something being evolved fresh from the start with a standard algorithm.
Especially if the first problem it was tested on was one with unclear and varying goals, such as >>58702872