Our research goal is to develop methodologies for building persistent, adaptive, collaborative, and believable agents. The agents must be persistent because we expect them to populate a world that is itself long-lived. The world is dynamic and constantly evolving, requiring our agents to be adaptive. Because the worlds we imagine are complex, and populated by a community of many other agents, our agents must form evolving collaborative partnerships with other agents in order to accomplish their goals. Finally, the community of agents will involve humans, so the agents must communicate and act in a believable manner. These realities give rise to a number of technical issues:

• Machine Learning. Agents must be able to adapt their behavior over time. In focusing on machine learning, we seek to understand: What algorithms are best suited to non-stationary, non-Markovian worlds? How can we make those algorithms scale?

• State and Activity Discovery. Adaptive agents must be able to recognize activities and discover new patterns in both the behavior of other agents and in the dynamics of the environment. In focusing on discovery, we seek to understand: When is a new state or activity worth discovering?

• Collaboration. Agent should be able to identify potential mutually beneficial (if transient) partnerships. In focusing on collaboration, we seek to understand: How can an agent detect that another agent can help it to accomplish its goals? How can the agents build a representation of their shared goals that does not also require a centralized "super mind"?

• Partial Programming. It should be possible for human developers to program new agents and build new environments rapidly, leading to the need for a programming environment that integrates adaptation directly. In focusing on partial programming, we seek to understand: How is it possible to allow non-machine learning experts to take advantage of machine learning algorithms in order to construct autonomous, adaptive systems? How can an agent build models that represent operationalizable and executable code?