On Value Functions, Theory of Life and Bellmann Equation Gymnastics
Does reinforcement learning have what it takes to bring us to artificial general intelligence? Are value functions universal in describing any problem? Is the problem of intelligence describable? Questions, questions…
Value functions, the powerhouse of reinforcement learning. Value functions, the cornerstone of intelligence? In the reinforcement learning classic, “Introduction to Reinforcement Learning” even Sutton and Barto state the following:
We take the position that value functions are important for efficient search in the space of policies.
And what are policies? Policies are the solution to our sequential decision problem. Now I am going to state something bold here. Sequential decision problems are all around you, your life is a sequential decision problem. Look at it this way, every day from when you get up you have to decide (sequentially) how your day is going to look like based on the new data that you receive. If you can filter out the noise from the data, most certainly you can learn something, useful. Summarized in this ambiguous image:
