Michael John Witbrock is a computer scientist in the field of artificial intelligence, associated with Carnegie Mellon University. A native of Christchurch, New Zealand, he is the former vice president of Research at Cycorp, which is carrying out the Cyc project in an effort to produce a genuine artificial intelligence. His background also encompasses work with Common Lisp and ObjectStore.

Witbrock was born in Christchurch, New Zealand, and holds a Ph.D. in computer science from Carnegie Mellon University. Before joining Cycorp, he was a principal scientist at Terra Lycos, working on integrating statistical and knowledge-based approaches to understanding Web user behavior; he has also been associated with Just Systems Pittsburgh Research Center and the Informedia Digital Library at Carnegie Mellon. In 2016, Witbrock joined and led the Reasoning Lab at IBM Watson.^[1] In 2019, Witbrock was recruited by the government and returned to his home country to establish and lead AI research initiatives in New Zealand.^[2]

Witbrock's dissertation work was on speaker modeling, and before joining Cycorp, he published in a broad range of areas, including neural networks, multimedia information retrieval, natural language understanding, genetic design, computational linguistics, speech recognition, and web-browser design. His work at Cycorp has focused on improving its knowledge formation efforts, particularly dialogue processing, machine reasoning, and on improving accessibility to the Cyc project. At the University of Auckland, Witbrock's research spans natural language processing, multi-hop reasoning, causal inference, and graph neural networks, focusing on advancing AI's interpretability, robustness, and real-world applications.^[3]

Michael Witbrock's early publications in the 1990s include "Rapid Connectionist Speaker Adaptation," co-authored with Patrick Haffner and presented at the IEEE 1992 International Conference on Acoustics, Speech, and Signal Processing in March 1992. In 1993, he and Marco Zagha contributed the book chapter "Back-Propagation Learning on the IBM GF11" to Parallel Digital Implementations of Neural Networks, edited by K.W. Przytula and V.K. Prasanna Kumar and published by PTR Prentice Hall in Englewood Cliffs. In October 1994, he published "User Extensibility in Amiga Mosaic" alongside Michael Fischer and Michael Meyer in the Proceedings of the Second International World Wide Web Conference (WWW) '94: Mosaic and the Web. He also co-authored "News-on-Demand: An Application of Informedia Technology" with A. Hauptman and M. Christel, appearing in the September 1995 issue of d-lib Magazine, The Magazine of the Digital Library Forum.

Around the turn of the millennium, Witbrock conducted research on statistical text summarization and the Cyc knowledge base. With Vibhu O. Mittal, he presented "Ultra-Summarization: A Statistical Approach to Generating Highly Condensed Non-Extractive Summaries" at SIGIR’99 in Berkeley, California, in August 1999. In October 2000, Michele Banko, Vibhu O. Mittal, and Witbrock published "Headline Generation Based on Statistical Translation" at ACL-2000 in Hong Kong. Mittal and Witbrock also authored "Language Modeling Experiments in Non-Extractive Summarization," published in 2003 as Chapter 10 of Language Modeling for Information Retrieval by Kluwer Academic Publishers. That same year, Witbrock co-authored two Cyc-related papers presented in Acapulco in August 2003: "An Interactive Dialogue System for Knowledge Acquisition in Cyc" with David Baxter, Jon Curtis, and others at the IJCAI-2003 Workshop on Mixed-Initiative Intelligent Systems, and "Inducing Criteria of Lexicalization of Parts of Speech using the Cyc KB" with Tom O’Hara, Bjorn Aldag, Stefano Bertolo, Nancy Salay, Jon Curtis, and Kathy Panton at IJCAI-03.

In the 2020s, his work has focused on modern artificial intelligence, machine learning, and reasoning. In 2021, he published "A Deep Reinforcement Learning Approach to First-Order Logic Theorem Proving" with K. Srinivas, V. Thost, and others in the Proceedings of the AAAI Conference on Artificial Intelligence. In 2023, he co-authored the arXiv preprint "Large Language Models Are Not Strong Abstract Reasoners Yet" alongside G. Gendron, Q. Bao, and G. Dobbie. He followed this in 2024 with "Robust Node Classification on Graph Data with Graph and Label Noise," co-authored with Z. Deng, Y. Chen, and R. Amor and published in the Proceedings of the AAAI Conference on Artificial Intelligence.

External resources include Michael Witbrock's Cycorp home page, his CMU publications page, and his video bibliography. Additionally, Cycorp publications provide recent publications with downloadable versions in PDF and/or PostScript format.