The field of artificial intelligence is moving towards developing more rational and human-like agents that can effectively collaborate with humans. Recent research has focused on enhancing the ability of agents to explore, make targeted guesses, and communicate effectively with humans. This has led to the development of new methods and frameworks that enable agents to better understand human behavior and adapt to dynamic environments. Notably, the use of Bayesian Experimental Design and Rational Speech Act modeling has shown promising results in improving agent performance and human-AI collaboration.

Some noteworthy papers in this area include:

• A study on building rational agents that explore and act like people, which introduced a novel dialogue task and developed methods to benchmark and enhance agentic information-seeking.
• A paper on planning ahead with RSA, which introduced a theoretical framework for adaptive signalling in dynamic environments.
• Research on aligning large language models with procedural rules, which introduced a methodology to compel language models to make their state-tracking process explicit and verifiable.
• A study on learning and deploying proactive language models from offline logs, which introduced a general framework for learning and deploying proactive dialogue agents directly from offline expert data.