Concluding Thoughts

We see three primary contributions of this formalization of novelty that will spur further research. First, formalization forces one to specify (or intentionally disregard) the required items in the theory. This can lead to insights about the problem and fill in knowledge gaps. For example, when applying the theory to the CartPole domain, numerous unanticipated issues were highlighted, new predictions made, and new experiments validated the new insights. Second, formalization provides a common language to define and compare models of novelty across problems. The precision of terms reduces confusion, while the flexibility allows it to be applied to a wide range of problems. Third, the formalization allows one to make predictions about where or why experiments incorporating some form of novelty might run into difficulties. For example, when the world-level and perceptual-level dissimilarity assessments disagree, we predict novelty problems will be more difficult. One example of difficulty is world-disparity using variables not represented in perceptual space.