Bradley P. Allen

Researcher, Intelligent Data Engineering Lab (INDE Lab), Informatics Institute, University of Amsterdam

Email · CV · Google Scholar · ORCID · PhilPapers · GitHub · LinkedIn

Background

I began my career in 1982 as a Senior Research Programmer at Carnegie Mellon University's Robotics Institute. In 1984, I became one of the first knowledge engineers of the expert-systems era at Inference Corporation, where I helped build ART, one of the period's leading commercial AI tools. I went on to serve as founder and CTO of three startups (Limbex, TriVida, and Siderean) and as Chief Architect at Elsevier. After forty years in industry, I have returned to academic research at the University of Amsterdam's INDE Lab. I hold a PhD in Computer Science from the University of Amsterdam and a BS in Applied Mathematics (Computer Science) from Carnegie Mellon University.

Research

I argue that knowledge engineering is explicitation: making the practice of a community of experts explicit, so that it can be reasoned with, evaluated, and answered for. The last of these has been the hard part, a constant problem from the rule-based systems of the 1980s to today's large language models: how do we build AI applications such that we can be accountable for their commitments?

My recent work shows that large language models can do two things that make such accountability tractable. They can evaluate knowledge graphs against plain-language definitions and give reasons for their judgmentssurfacing factual errors and subtler disagreements about word meaning, laid out where people can inspect and contest them. And they can report their internal beliefs in a structured way, even when those beliefs conflict — letting people see what they are committed to and reason soundly despite the inconsistency.

This work continues on three fronts: structured human–LLM dialogue protocols for building knowledge bases collaboratively, modal and substructural logics for reasoning about what LLMs believe, and methods for probing which inferential rules an LLM actually endorses. Taken together, this points toward a knowledge engineering recast as human–LLM dialogue — one that builds on the growing capabilities of LLMs yet maintains human accountability for the outcomes.

Talks & Media

Selected Publications


Copyright © 2026 Bradley P. Allen. Last updated 7 June 2026.