1
$\begingroup$

The more problems an agent can solve, the higher the probability of natural selection.

Isn't this the process by which humans have become general-purpose intelligence by solving many problems in nature?

Isn't this article I wrote similar to this study?

deepmind-Generally capable agents emerge from open-ended play

they said "The agent playing a variety of test tasks. The agent was trained across a vast variety of games and as a result is able to generalise to test games never seen before in training."

I think this is similar to my argument

$\endgroup$
11
  • $\begingroup$ I think I get what you're asking. I took the liberty of editing and reopening, since so reasons were given for closure. $\endgroup$
    – DukeZhou
    Jun 3 at 22:57
  • 1
    $\begingroup$ What do you mean by "missions", "entites" and "objects" here? You have asked many similar questions in the past. Basically, you're asking whether evolution or "building skills on top of other skills" can lead to AGI. What answers are you expecting? Yes or no? You already have those answers, so what are you really expecting? What have you found by searching the web? Are you aware of "evolutionary algorithms"? If yes, how does it related to your question? $\endgroup$
    – nbro
    Jun 5 at 12:15
  • $\begingroup$ @nbro A mission is a problem. For example, winning a game or playing mini-games to win. And the object is the agent. The more problems the agent solves, the more likely it is to be selected by natural selection, and the agent will evolve to solve as many problems as possible. If you choose an agent who can solve multiple problems and can solve new problems in a short time, that agent may have applied the abilities used in the problems it has solved in the past. $\endgroup$
    – Dimer
    Jun 5 at 14:07
  • $\begingroup$ @nbro Natural selection of agents that solve multiple problems in this way is the way to create general-purpose intelligence. I may be wrong, but historically, I know that the human evolutionary process has also evolved to know how to solve various problems and thus acquired general intelligence. $\endgroup$
    – Dimer
    Jun 5 at 14:08
  • 1
    $\begingroup$ Closed pending clarification—the question is too loose and is unlikely to yield a productive answer. Yes, performing a large number of tasks across different domains is an attribute of AGI, but the way the question is phrased is not constructive or rigorous. $\endgroup$
    – DukeZhou
    Jun 17 at 22:58