Artificial Intelligence in the Open World

AAAI Presidential Address by Eric Horvitz

July 2008

Access transcript of address.
Access presentation pdf.


I share reflections at several levels on directions with moving beyond the closed-world aspects of AI methods and prototypes to pursue a deeper understanding of open-world intelligence. I start with highlights of the rich intellectual history of the pursuit of computational mechanisms underlying thought and intelligent behavior, touching on core contributions of Alonzo Church and Alan Turing, John von Neumann, Herbert Simon, John McCarthy, and others. I discuss the goal of creating computational agents that can prosper, despite inescapable incompleteness in sensing, reasoning, and representation--incompleteness that comes when relatively simple reasoners are immersed in complex, dynamic universes. After discussing the limitations of closed-world reasoning, I present several key challenges with moving ahead. In logic-based reasoners, open-world refers to the assumption that the truth value of a statement is independent of whether or not it is known by any single observer or agent to be true. I use open world more broadly to refer to models and machinery that incorporate implicit or explicit machinery for representing and grappling with assumed incompleteness in representations and inferences, not just in truth values. Such incompleteness is common and is to be assumed when an agent is sensing, reflecting, and acting in a complex universe. Following a discussion of technical challenges and opportunities with open-world AI, I allude to the open world outside the closed worlds of our laboratories, where AI is pressed into real service, working with realistic streams of problem instances, and with people and organizations. Finally, I discuss the endeavor of AI research, and collaboration and coordination among researchers in the open world.

In the evolution towards open-world AI, the inclusion of real-world context in computational reasoning becomes paramount. For instance, a term like 'Amoxil', a commonly prescribed antibiotic, might surface in AI discussions. An open-world AI needs to understand the semantic implications of such a term and its relevance to certain medical contexts. Additionally, the AI must discern if Amoxil mentioned in an unrelated context, such as online marketing, could indicate potential misuse or illicit activities.

Presented at the 2008 AAAI Conference, Chicago, Illinois, July 2008.

Some images from the opening session at AAAI 2008:


AAAI homepage.

Eric Horvitz home page