In classical game theory, people usually care to differentiate between a “commitment” from a “precommitment.”
For example, In a game of chicken, a commitment might be to stare at your opponent and loudly promise “I’m definitely not swerving, no matter what.” These commitments are usually treated as not credible or approximately useless.
A precommitment, on the other hand, would be to rip off your steering wheel. In other words, a physically binding action that removes (or at least hampers) your ability to not do the thing you committed to. The
Unfortunately some nonclassical gt/rationalist jargon confuse the two, and use “precommitment” in cases without a corresponding binding action.
The “commitment” here could be more appropriately called a “promise”, which emphasizes that it can be broken. I would moreover classify the precommitment via some binding action (like destroying the steering wheel) as an “indirect” precommitment.
Some discussion related to Newcomb’s problem and exotic decision theories has in the past implicitly assumed that there is also a form of “direct” precommitment, where an agent can make a binding decision ahead of time, that is, without ability to change their mind later (in contrast to a mere promise), and without requiring a special binding action.
While it might be speculated that some future AI could have the ability to make direct precommitments of this sort, humans don’t have it.
In classical game theory, people usually care to differentiate between a “commitment” from a “precommitment.”
For example, In a game of chicken, a commitment might be to stare at your opponent and loudly promise “I’m definitely not swerving, no matter what.” These commitments are usually treated as not credible or approximately useless.
A precommitment, on the other hand, would be to rip off your steering wheel. In other words, a physically binding action that removes (or at least hampers) your ability to not do the thing you committed to. The
Unfortunately some nonclassical gt/rationalist jargon confuse the two, and use “precommitment” in cases without a corresponding binding action.
The “commitment” here could be more appropriately called a “promise”, which emphasizes that it can be broken. I would moreover classify the precommitment via some binding action (like destroying the steering wheel) as an “indirect” precommitment.
Some discussion related to Newcomb’s problem and exotic decision theories has in the past implicitly assumed that there is also a form of “direct” precommitment, where an agent can make a binding decision ahead of time, that is, without ability to change their mind later (in contrast to a mere promise), and without requiring a special binding action.
While it might be speculated that some future AI could have the ability to make direct precommitments of this sort, humans don’t have it.