This award supports the organization and the participation of approximately 35 experts in a two-day workshop to be held July 12-13, 2007 at MIT Endicott House in Dedham, Massachusetts. The purpose of this workshop is to bring together leading researchers in Artificial Intelligence to assemble a research agenda for developing AI systems capable of human level performance (Human Level AI or HLAI). HLAI includes the following constellation of capabilities exhibited by human beings: (1) The ability to adapt and survive for an extended period of time in a rich, complex physical and social environment; (2) The ability to learn throughout this extended period of existence and, in particular, to improve performance on a fixed set of tasks, learn to perform new tasks, transfer knowledge and skill from one task to another, the ability to execute a new task after being told (or shown) how to do it; (3) The ability to reason about their own capabilities and limitations, to formulate new tasks, and to gather information needed for learning how to perform those tasks; (4) The ability to communicate and interact (e.g., cooperate) with other agents, to form teams, to recognize the goals, beliefs, and intentions of other agents, to persuade and negotiate with other agents; and (5) The ability to design and execute changes in the environment (physical and social) to improve their ability to survive and achieve their goals. In other words, an HLAI system is characterized by the potential to perform a great breadth of tasks--including those not pre-envisioned by the system designer--and by the ability to reflect on and improve all of its own capabilities, and to do so in a robust, flexible manner. This workshop will review the current state of the art, propose new integrated architectures for robust integrated intelligent systems, and suggest ways for researchers to leverage results across the many subspecialties in AI.
