And since determinism, can’t play mixed strategies,
It can try. It can cycle. Look up ‘Iocaine powder rock paper scissors’, for material around how computer programs play rock paper scissors.
Basically a comparison operator for Turing machines where you can say a machine is more intelligent than another machine, without specifying which problem they are meant to handle. I don’t think this exists.
You can look at which program tends to beat which other programs. (Though hardware does come up sometimes.)
Explicitly calling it a Turing machine makes it less likely we bring in such undefined terms into the discussion.
How do you tell what program is running?
an agent can’t be “perfect” at all problems imo. It has play better on some problems and worse on others.
There are differences between the problems. If an agent can’t tell the differences, then it’s got some limitation, like being shorter than all the problems to solve. (The benefit of formalizing this, is the source of the problem determines what you go about doing about it.)
And since determinism, can’t play mixed strategies, you’re either 100% perfect or 100% imperfect for a given situation.
You can learn. You can come up with better algorithms.
While the question of ‘random’ often makes reference to having some source of it, in practice, some effort may be required to achieve this—whether via a quantum or other source of real randomness, or cryptographically secure random number generation.
(In practice, part of how some ‘random number generators’ work is by mixing in some randomness from user input (timing I think), and combined with the old seed to get the new seed. A simple, isolated system may be predictable.* Pseudorandomness isn’t confined to ‘one program’ - it often uses code that already exists somewhere else. (And often, a seed, or a state.) (A predetermined random seeds that’s the default, or a new one based on time is commonly used in some languages.)
*However, if I don’t have a quantum random number generator, then the question of getting that data in a way that isn’t intercepted, would be a bit tricky.)
It can try. It can cycle. Look up ‘Iocaine powder rock paper scissors’, for material around how computer programs play rock paper scissors.
You can look at which program tends to beat which other programs. (Though hardware does come up sometimes.)
How do you tell what program is running?
There are differences between the problems. If an agent can’t tell the differences, then it’s got some limitation, like being shorter than all the problems to solve. (The benefit of formalizing this, is the source of the problem determines what you go about doing about it.)
You can learn. You can come up with better algorithms.
While the question of ‘random’ often makes reference to having some source of it, in practice, some effort may be required to achieve this—whether via a quantum or other source of real randomness, or cryptographically secure random number generation.
(In practice, part of how some ‘random number generators’ work is by mixing in some randomness from user input (timing I think), and combined with the old seed to get the new seed. A simple, isolated system may be predictable.* Pseudorandomness isn’t confined to ‘one program’ - it often uses code that already exists somewhere else. (And often, a seed, or a state.) (A predetermined random seeds that’s the default, or a new one based on time is commonly used in some languages.)
*However, if I don’t have a quantum random number generator, then the question of getting that data in a way that isn’t intercepted, would be a bit tricky.)
ETA: