3 ms·

The problem with this, specifically the assumption that RL gives an equivilance to natural selection and evolution, is that RL typically assumes a computational

The problem with this, specifically the assumption that RL gives an equivilance to natural selection and evolution, is that RL typically assumes a computational environment it interacts in while natural selection and evolution assumes the physical world as the environment. The important difference here is that in order for RL to translate to solving real world problems, you need to faithfully and computafionally simulate the real world's physical processes and rules, or at least enough that n-th order processes exist accurately. I've done various types of computational modeling and simulation work at different scales throughout my career with all sorts of scientists and engineers and I can tell you, pretty much no domain is there where you have good enough representative models RL can be used in. Some narrow special cases exist but nothing to the degree of a massive environment full of well coupled expert domain models. Some of the best cases are going to be so computationally bound that it would be quicker to do things for real vs simulate. If you want RL to work and learn, it's likely possible under the connection you point out, but has to do this using physical machines and sensors interacting with the physical world like life as we know it does. Your AGI won't be able to cheat and run through the evolution process quicker using faulty reductionist models we use in most simulations (which is what everyone implicitly is hoping for), IMHO. If you try this, your AGI is going to learn all sorts of flaws within those environments or at the very least, have so many narrow scoped bounds it won't be that "general." A lot of simulated models are frankly garbage (they have some useful narrow scope but are typically littered with caveats) and they've been in development pretty much since digital computing began.

tsimionescu 5y ago
You are absolutely right about RL in practice. In fact, if we were to look at actual RL algorithms, I believe the paper's claims fall flat in many other ways. This is actually my criticism of it: its arguments are only convincing when RL is defined only as the extremely general notion of an agent seeking to maximize some reward function by interacting with an environment. This is so comically general that the only alternative I can think of is to posit a transcendental god. Once we get into the details, their claims stop being iron clad. Even worse, some of their claims become actually hard or impossible to accept if applied to actual RL algorithms we have today. You give one good example with the difficulty of modeling the world. The implicit claim they make that this would be realizable in reasonable time (say, less than a billion years) is also not well supported. The idea that humans or mammals learn their social behaviors through RL rather than a good deal of reasoning from evolutionarily-trained first principles pretty clearly fails in the face of the poverty of the stimulus argument[0]. Overall, the claims in the paper tend to switch between obvious (if taken to talk about the general idea of maximizing reward) to almost certainly wrong (if taken to talk about known RL algorithms, reasonable time frames, and specific examples of what is supposed to be learned). [0] the poverty of the stimulus argument may be controversial in linguistics where it was first formulated. Still, if applied to mammal or insect socialization, the extremely low time frames in which individuals of a species start exhibiting typical behaviors basically proves in my opinion that they are instincts, trained at the population level through evolution, not individual learning through RL. The extreme similarity of behavior between individuals of the same species, VS the variety of behaviors between different species, also suggests an important component of species-level rather than individual level learning.

tsimionescu 5y ago

You are absolutely right about RL in practice. In fact, if we were to look at actual RL algorithms, I believe the paper's claims fall flat in many other ways. This is actually my criticism of it: its arguments are only convincing when RL is defined only as the extremely general notion of an agent seeking to maximize some reward function by interacting with an environment. This is so comically general that the only alternative I can think of is to posit a transcendental god. Once we get into the details, their claims stop being iron clad. Even worse, some of their claims become actually hard or impossible to accept if applied to actual RL algorithms we have today. You give one good example with the difficulty of modeling the world. The implicit claim they make that this would be realizable in reasonable time (say, less than a billion years) is also not well supported. The idea that humans or mammals learn their social behaviors through RL rather than a good deal of reasoning from evolutionarily-trained first principles pretty clearly fails in the face of the poverty of the stimulus argument[0]. Overall, the claims in the paper tend to switch between obvious (if taken to talk about the general idea of maximizing reward) to almost certainly wrong (if taken to talk about known RL algorithms, reasonable time frames, and specific examples of what is supposed to be learned). [0] the poverty of the stimulus argument may be controversial in linguistics where it was first formulated. Still, if applied to mammal or insect socialization, the extremely low time frames in which individuals of a species start exhibiting typical behaviors basically proves in my opinion that they are instincts, trained at the population level through evolution, not individual learning through RL. The extreme similarity of behavior between individuals of the same species, VS the variety of behaviors between different species, also suggests an important component of species-level rather than individual level learning.