On Monday, December 09, 2024, at 1:17 PM, James Bowery wrote: > Is this inadequate to prevent the random agent strategy? In addition to what you quoted, another point that I forgot to add is that in Matching Pennies you can play randomly strategically to observe your opponent without taking more than a 50% loss, and also hide your strategy from your opponent. You can do the same in PtB, but hiding your current strategy is less advantageous because your choice is just one of potentially many (depending on how many agents are in the game), so your strategy is not being observed directly and you are not taking advantage of non-randomness that could be occurring in current rounds.
On Monday, December 09, 2024, at 11:30 AM, Matt Mahoney wrote: > It is also a proof of Wolpert's theorem, that two computers cannot mutually > predict the other's actions. Imagine a variation of the game where each > player receives the source code and initial state of their opponent as input > before the start of the game. Who wins? > > Wolpert's theorem is the reason AI is dangerous. We measure intelligence by > prediction accuracy. If an AI is more intelligent than you, then it can > predict your actions, but you can't predict (and therefore can't control) its > actions. Wolpert's Theorem just emphasizes the need for meta-learning in AGI systems. No single agent can dominate every game of PtB, but if the system is fed the game results from current and past games, it should learn how to create a better agent for the next PtB games, especially taking meta-game information into account. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T705ed500a1a7e589-M53c2481015b2cb75900c6f7a Delivery options: https://agi.topicbox.com/groups/agi/subscription
