I know little about machine learning, so that may be a dumb question: in linguistic there is an argument called the poverty of stimulus. The claim is that children must figure out the rules of language using only a limited number of unlabeled examples. This is taken as evidence that the brain has some kind of hard-wired grammar framework, that serves as a canvas for further learning while growing up.
Is it possible that tools like EfficientZero help find the fundamental limits for how much training data you need to figure out a set of rules? If an artificial neural network ever manages to reconstruct the rules of English by using only the stimulus that the average children is exposed too, that would be a strong counter-argument against poverty of stimulus.
I would be careful using reinforcement learning to check for theoretical maximization of training data, given that plenty of agents generally do not start out with 0 bits of information about the environment. The shape of input data/action space is still useful information.
Even in designing the agent itself, it seems to me that general knowledge of human-related systems could be introduced into the architecture.
Selecting the architecture that gives us highest upper-bound for information utilization in a system is also, in some sense, inserting extra data.
At the start, I thought you were going to in a different direction:
Given that we are giving AIs frameworks, just how far can we go with that? And will they be useful outside of this or that set of games?
Is it possible that tools like EfficientZero help find the fundamental limits for how much training data you need to figure out a set of rules? If an artificial neural network ever manages to reconstruct the rules of English by using only the stimulus that the average children is exposed too, that would be a strong counter-argument against poverty of stimulus.
Unless, that artificial network comes with a grammar framework?
I know little about machine learning, so that may be a dumb question: in linguistic there is an argument called the poverty of stimulus. The claim is that children must figure out the rules of language using only a limited number of unlabeled examples. This is taken as evidence that the brain has some kind of hard-wired grammar framework, that serves as a canvas for further learning while growing up.
Is it possible that tools like EfficientZero help find the fundamental limits for how much training data you need to figure out a set of rules? If an artificial neural network ever manages to reconstruct the rules of English by using only the stimulus that the average children is exposed too, that would be a strong counter-argument against poverty of stimulus.
I would be careful using reinforcement learning to check for theoretical maximization of training data, given that plenty of agents generally do not start out with 0 bits of information about the environment. The shape of input data/action space is still useful information.
Even in designing the agent itself, it seems to me that general knowledge of human-related systems could be introduced into the architecture.
Selecting the architecture that gives us highest upper-bound for information utilization in a system is also, in some sense, inserting extra data.
At the start, I thought you were going to in a different direction:
Given that we are giving AIs frameworks, just how far can we go with that? And will they be useful outside of this or that set of games?
Unless, that artificial network comes with a grammar framework?