As AI has become a huge industry, to an extent it has lost its way. What is needed to get us back on track to true intelligence? Most of all we need agents that learn continually from their first-person experience. We also need world models and planning. We need knowledge that is high-level and learnable. We need to meta-learn how to generalize. The Oak architecture is an approach based on reinforcement learning that seeks to satisfy all these needs and which has three special features. First, all of its components learn continually. Second, each learned weight has a dedicated step-size parameter that is meta-learned using online cross-validation. And third, abstractions in state and time are continually created in a five-step progression:
1) create a new feature
2) pose a subproblems to achieve the feature
3) learn to solve the subproblem
4) learn a transition (world) model of the solution
5) plan using the transition model.
The Oak architecture is rather meaty; in this talk I will give an outline and point to the many works, prior and contemporaneous, that are contributing to its overall vision of how superintelligence can arise from an agent’s experience.
