From traffic lights to household cleaning robots, automation is a ubiquitous and growing aspect of our lives. However, the transition from automation to agency, where a software or robotic agent conducts a role and has the responsibility that might have once been only trusted to a human, is still lagging. Currently, the scope of most intelligent agent systems is limited to specific domains, specific configurations, and specific operating conditions. To move ahead, we must address several gaps in order to advance to human-agent collaboration. These include:

• Understanding Objectives: An agent may not know the desires of the user, because (a) there isn’t a language to convey it, (b) a language would be too complex or (c) the user may not know what they want until they see a solution. In addition, the agent may not be able to convey its objectives to the human. For example, consider the barriers to creating an autonomous agent that would book flights instead of simply gathering and displaying flight data.

• Adapting to Context: An agent may be required to understand and adapt to general aspects of the state of the world outside the basic scope of its task. In addition, it should adapt given the objectives and goals mentioned above. For example, consider a route-planning agent that would not just consider current traffic conditions but anticipate how sporting events or upcoming weather phenomena would affect solutions.

• Collaborative Computation: There are many problems, such as coordination of rescue and recovery efforts after a disaster, where generating a solution is too complex for an intelligent agent to generate a reasonable solution. There may be collaborations where, with human guidance, intelligent agents can be created to improve performance when partnered with human actors.

The aim of this symposium is to begin to form a community of researchers working on different aspects of the problem. The Agents and AI communities have tackled this problem from the agents’ point of view, while the Intelligent User Interfaces community has begun to address this problem from the human’s point of view. By bringing together researchers from these and other communities, we can foster collaborations to share accomplishments so far, identify common issues and challenges, compare architectures and propose directions to move forward.