An Introduction to Multi-Agent Systems

[Clayton]

Tonight's watch-list...

[Atruepatriot]

Thank you.

[Clayton]

Thank you.

I got a few videos in and realized that this guy is super-clear in his explanations, so I thought I'd share...

[Atruepatriot]

Thank you.

I got a few videos in and realized that this guy is super-clear in his explanations, so I thought I'd share...

Cool, check it out later tonight! I always learn from this stuff you share.

[Clayton]

I finished the series (survey level only, not taking notes lol). I am doing research for some planned posts which will touch on this and related topics. By way of teaser: (logical) proofs = algorithms = best strategies in solved (or solvable) games (of sufficient generality). The first equality is not my own concept, but the second equality is what I want to show (or, if it already exists in the literature, find).

The reason I am interested in the second equality has to do with sorting out the long-run tradeoffs between force and truth in an entirely abstract sense. In other words, suppose the universe is filled with beings who do not believe in truth and you are one of a tiny group of people who do. Is that the end of truth? Or is there a larger framework in which, even starting from a purely hostile situation, the truth becomes inevitable (even if the truth-haters never believe/accept it). If the forces opposing truth can win by sheer numbers, then any time our abstract universe has significantly more than 50% of agents who reject and oppose truth, then peaceful truth-belief is doomed. You can think of it a bit like rational pacifists[1] trying to persuade aggressive cretins not to attack them, using only rational discussion (in which the cretins refuse to participate...) This is the (game-theoretic / computational) problem I spend most of my free time thinking about in one way or another.

I am convinced it can be shown that, in a universe in which the rules are sufficiently general that a universal Turing machine can be constructed, you can also show that every strategy which denies truth ultimately fails. As stated, this is not sufficiently precise, so that's why I'm studying about multi-agent systems. We have to first define some "game" which is being played, and I found a video a couple years ago that perfectly described the game I have in mind but I have completely lost that video and I have searched for it again and again in vain. It would be immensely helpful to find it again, so I keep looking. The little that I can recall from it is that we basically have a software program that is required to satisfy some property, such as, "never crashes without giving an error message." Then we have some hardware on which that software is running. There are two players in the game. The first player must write the software to satisfy the given constraint. The second player is allowed to alter the hardware in any way they like. This is just a high-level recollection and there were a lot of technicalities surrounding what these players were allowed/required to do, but the conclusion of the talk was that it is indeed possible for the first player to win even despite the enormous advantages of the second player.

"Arbitrarily altering hardware" is a very good abstract description of what force/violence is. And so the position of the rational pacifist who wants to say that "peace/truth/liberty wins (sooner or later)" is like the player who is trying to write a piece of software that obeys some kind of constraint - such as throwing an exception and failing gracefully no matter how contorted the hardware this software will turn out to be run on. Anyway, the search continues...

[1] - I don't mean absolute pacifism (excluding even self-defense), I just mean people who are committed to peaceful cooperation if it is at all possible, other than by completely relinquishing their rights (which isn't peace anyway)...