I am the Senior Research Manager for the new Allen Institute of AI (AI2), managing Project Aristo, the flagship project of the new Allen Institute for AI (AI2). Project Aristo is focused on large-scale, semi-automatic acquisition of knowledge from text, and use of that knowledge for explainable question answering, in particular for answering grade-school science questions. This work is heavily leveraging natural language processing, textual entailment, and semi-formal representations.

Halo 2.0 (2013)In 2013, I created and led a new program of research, called Halo 2.0, focused on large-scale acquisition of knowledge from text and using that knowledge for grade-school level science exams. Halo 2.0 was subsequently renamed as Project Aristo, and became the flagship project of the new Allen Institute for AI (AI2).Textual Question-Answering (2011-2013) Created and led an innovative new thread of research for Project Halo to acquire knowledge and answer questions directly from natural language text, based on reasoning at the textual level (“textual entailment”), using representations derived automatically from text via natural language processing (NLP). This project laid the foundation for the current two year effort. A derivative of the underlying textual entailment software won 2nd place for Vulcan (out of 11 participating teams) in the international Question-Answering for Machine Reading (QA4MRE) challenge in 2012. AURA project (2010-2011)Designed and developed the Suggested Question mechanism for AURA, based on an early prototype from SRI, allowing AURA to propose semantically similar, answerable questions when it was unable to handle a user's question directly. This mechanism was critical for AURA's robustness and usability in trials with students in 2011.General (2010-present)With other team members, continuous involvement in defining, critiquing, and planning the evolution of the Halo Program. Substantial technical involvement in other aspects of Halo, including the “hybrid” prototype.

Principal Investigator for Boeing's work in the Vulcan Halo Project (2002-2010)Involvement in all aspects of the SRI/Boeing/UT team's work on the Halo AURA system for acquiring and reasoning with scientific knowledge bases. Developed AURA's natural language interface for asking questions, and AURA's knowledge representation and reasoning engine, KM (The Knowledge Machine).Principal Investigator for IARPA AQUAINT (Advanced QA for Intelligence) Project (2006-8)Created, won, and managed Boeing’s (Prime) project in this program, including managing work by two subcontractors (Princeton Univ and USC/ISI). Developed novel software capable of textual reasoning for answering questions. In 2007, Boeing formally recognized this work as a Breakthrough Technology. Principal Investigator for Boeing's work in DARPA's Reading to Learn Project (2005-6)Directed and conducted research into the feasibility of a machine acquiring text-book knowledge via language processing of simplified English texts.Principal Investigator for Boeing's work in DARPA's MOBIUS project (2005-8)Conducted collaborative research into the feasibility of large-scale machine reading from text. This project directly led to DARPA launching a full program in Machine Reading in 2009.Principal Investigator for Boeing's work in DARPA's Rapid Knowledge Formation project (1999-2003)Led Boeing's work (SRI team) in DARPA's RKF project. As a team, developed a system allowing experts to rapidly construct knowledge bases by combining pre-built knowledge pieces ("components") together graphically.The Technical Expert Locator (Boeing Internal) (1999-2001)Created vision for and developed Boeing’s Technical Expert Locator (TEL), an AI tool for finding experts that exploited Boeing’s Aerospace Thesaurus as a massive source of semantic knowledge. The TEL was deployed and in full commercial use from 2001-5, and was a primary vehicle for locating expertise within the company.

Developed a theory of concept composition with Prof. Bruce Porter, by which representations of complex, concepts can be assembled from modular components. This work won a AAAI'97 Best Paper award.Developed and implemented the DCE Help-Desk Assistant, a prototype knowledge-based system that inferred answers to certain types of customer queries using machine reasoning.

Created and developed Electronic Trader (ET), a prototype for automatically detecting arbitrage opportunities using AI techniques. ET was licensed and used in field trials by Canada’s Export Development Corporation.

Developed a technique for guiding inductive learning using domain knowledge, with Prof Stan Matwin, biasing the machine away from "clearly nonsensical" hypotheses. This work has been cited over 70 times.

Developed a computational model of argumentation, based on Toulmin’s work, by which an expert system and user could interact (“cooperatively argue”) to jointly solve a problem. This work was the basis of my PhD.Principal Investigator and developer of Optimist, a full-scale, commercial AI system that implements the above theory. Optimist assisted geologists in oil exploration, and was in commercial use from 1989-1994.Co-developed a new rule induction algorithm, CN2, now well-known in the machine learning community. (This work has been cited over 2000 times).