Avi Pfeffer

Using emotions to enhance decision-making

We present a novel methodology for decision-making by computer agents that leverages a computational concept of emotions. It is believed that emotions help living organisms perform well in complex environments. Can we use them to improve the decision-making performance of computer agents? We explore this possibility by formulating emotions as mathematical operators that serve to update the relative priorities of the agents goals. The agent uses rudimentary domain knowledge to monitor the expectation that its goals are going to be accomplished in the future, and reacts to changes in this expectation by experiencing emotions. The end result is a projection of the agents long-run utility function, which might be too complex to optimize or even represent, to a time-varying valuation function that is being myopically maximized by selecting appropriate actions. Our methodology provides a systematic way to incorporate emotion into a decision-theoretic framework, and also provides a principled, domain-independent methodology for generating heuristics in novel situations. We test our agents in simulation in two domains: restless bandits and a simple foraging environment. Our results indicate that emotion-based agents outperform other reasonable heuristics for such difficult domains, and closely approach computationally expensive near-optimal solutions, whenever these are computable, yet requiring only a fraction of the cost.

Probabilistic programming for assessment of capability and capacity

Answering the questions What can the adversary do? and What will the adversary do? are critical functions of intelligence analysis. These questions require processing many sources of information, which is currently performed manually by analysts, leading to missed opportunities and potential mistakes. We have developed a system for Assessment of Capability and Capacity via Intelligence Analysis (ACACIA) to help analysts assess the capability, capacity, and intention of a nation state or non-state actor. ACACIA constructs a Bayesian network (BN) to model the objectives and means of an actor in a situation. However, a straightforward BN implementation is insufficient, since objectives and means are different in every situation. Additionally, we wish to apply knowledge about an element gained from one situation to another situation containing the same element. Furthermore, different elements of the same kind usually share the same model structure with different parameters. We use the probabilistic programming language Figaro, which allows models to be constructed using the power of programming languages, to address these issues, generating BNs for diverse situations while maximizing sharing. We learn the parameters of a program from training instances. Experiments show ACACIA is capable of making accurate inferences and that learning effectively improves ACACIAs performance.