Brendan Duncan, PhD serves as the Head of AI at Feeling Great.

Brendan specializes in recommender systems, natural language processing, and machine learning.

Selected Work Experience

Head of AI

Researcher in AI

Head of Data Science at Fliptop (Acquired by LinkedIn)

Education

M.S. in Computer Science

Ph.D. in Computer Science

A metaphor to illustrate the benefits of intent signals. We should create intent-aligned AI systems that shield us from a large number of unintended and harmful behaviors.

Thesis and Philosophy

Brendan’s PhD thesis, “Leveraging Behavioral Intent Signals in Recommendation and Natural Language Generation” (publication on hold for two years because of proprietary research), demonstrates that including explicit intent signals in AI models significantly improves their performance. Intent signals can be obtained in several ways: from human annotation, from an additional dataset that is correlated with underlying intent, and from the output of a preliminary model trained to recognize intent structure in a dataset.

Brendan feels strongly that safe AI systems require either training models with explicit intent signals or alignment with human annotators for well-defined tasks. It is potentially dangerous, in his view, to assume that any LLMs can be used as an “Artificial General Intelligence” (AGI) to make important decisions that the system was not explicitly trained to make. Where possible, we should either use an “Artificial Specific Intelligence” that has been explicitly trained to perform a given task, or have annotators audit a system to ensure it’s properly aligned for a given task.

Publications

  • Duncan, B., Kallumadi, S., Berg-Kirkpatrick, T., and McAuley, J., 2024. STOIQ Rec: Leveraging Customer Service Conversations for Chatbot Recommendation. Submitted to RecSys 2024. Bari, Italy.

  • Duncan, B., Kallumadi, S., Berg-Kirkpatrick, T., and McAuley, J., 2023, October. MAWI Rec: Leveraging Severe Weather Data in Recommendation. Submitted to RecSys 2024. Bari, Italy.

  • Duncan, B., Kallumadi, S., Berg-Kirkpatrick, T., and McAuley, J., 2023, May. Jointly modeling products and resource pages for task-oriented recommendation. In Companion Proceedings of the ACM Web Conference 2023 (WWW ’23 Companion). Austin, Texas, USA.

  • Duncan, B., 2015, October. Modeling a Sales Funnel and Lead Behavior for Predictive Lead Scor- ing. Invited talk at Big Data for B2B Marketing and Sales Workshop, IEEE International Confer- ence on Big Data. Santa Clara, California.

  • Duncan, B.A. and Elkan, C.P., 2015, August. Probabilistic modeling of a sales funnel to prioritize leads. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1751-1758). Sydney, Australia.

  • Duncan, B. and Elkan, C., 2014, September. Nowcasting with numerous candidate predictors. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases (pp. 370-385). Springer. Nancy, France.

Brendan Presenting research to Lowe’s CIO Seemantini Godbole in June 2023 (Seemantini was awarded CIO of the year in 2020 and 2023)

Filed Patents

  • Brendan Duncan: System and Method for Full Funnel Modeling for Sales Lead Prioritization: March 16, 2015.

  • Brendan Duncan: System and Method for Using Marketing Automation Activity Data for Lead Prioritization and Marketing Campaign Optimization: June 30, 2015.

  • Dan Chiao, Brendan Duncan, Liang-Yu: System and Method for Lead Prioritization Based on Results from Multiple Modeling Methods: June 30, 2015.

Interests

Brendan Duncan lives in San Diego near UC San Diego’s campus, where he received his undergraduate degree and Ph.D.

When he gets the chance, he likes to go out on the ocean—either surfing the waves near campus or crewing for sailboat races in the area.