AI & Knowledge Systems Architect
James B. Quirk
I design structures that help people and machines communicate what things mean (ontologies, knowledge graphs), and the evaluation and governance that keep those structures trustworthy.
PhD in Applied Physics, Imperial College London. Currently architecting federated knowledge graph and AI infrastructure at Pearson.
Focus
What I work on
A language model has no model of the world. It has a statistical picture of how words follow one another, which is not the same as knowing what exists or what is true. Ontologies and knowledge graphs are an attempt to help machines and people agree on what things mean. We make ontological commitments.
Getting entities out of a source is not too computationally difficult. But getting theright entities out is still a challenge. Determining whether anything was missed, whether two records are truly identical, and whether we can account for how they change involves making semantic and complexity trade-offs. We make epistemological commitments.
Without a way to surface ontological and epistemological commitments as conscious choices, they are blind spots. Exposing our assumptions about how we’ve modelled some slice of the world, and how we’ve chosen to carve the evidence to fit our model, asks us to improve them over time. To do so, we make evaluations.
I write about ontology, epistemology and evals – and the practical philosophy that links all three.
Writing coming soon.
Elsewhere
Future:human
Future:human – wayfinding in an algorithmic age.
A newsletter to update your ontological priors.