An overview of computational methods for goal modeling
Keywords: goal modeling, bayesian inference, inverse reinforcement learning
Abstract: Goals play an crucial role in driving humans for their daily lives. They can be as abstract as a long-term blueprint or as specific as an immediate directive. For sophisticated AI agents, goals are essential as they not only define the objectives to be achieved but also mimic human decision-making processes, enabling agents to understand and interact in human-centric environments. In this essay, we reflect on the psychological understanding of goal perception and attribution with behavioural experiments and proposed mechanisms. Based on this point, we delve into the contemporary computational techniques for goal modeling. We specifically focus on Bayesian models and inverse reinforcement learning. In the end, we conclude with the pros and cons of existing methods and discuss potential future directions.
Submission Number: 145
Loading