Two projects I worked on recreated moments no camera caught: Pelé’s “lost” 1959 goal at the Pelé Museum in Santos, and Liz Garbus’s Telluride short Love, Rendered. I also contributed to Gemini Omni while Flow added image and audio generation, and wrote about giving agents the drudgework to protect human attention.

I served as Technical Director on ANCESTRA for its Tribeca premiere (essay, making-of film), and launched Flow, a filmmaking tool built alongside directors.

We launched Google Veo and VideoFX, partnering with filmmakers including Donald Glover and his studio Gilga. Two papers measured divergent creativity in humans and LLMs and tested how LLM humour aligns with working comedians.

Screenwriters and playwrights tested Dramatron (code, paper, patent), our tool for co-writing full scripts with language models. In parallel, we published on cultural general intelligence and communicative capital.

I wrote a field guide to designing and evaluating human-centred interactive ML. With colleagues, I studied emergent communication across agent populations and at scale, and helped build BIG-bench.

We built StoryCentaur (code, paper) so writers could steer language models with a few examples mid-draft. Other work explored actors performing with AI narrators, remote stage improv, synthesizing realistic laughter, and symbolic behaviour in AI.

Information theory gave us a way to shape narrative arcs through dialogue, alongside agents trained to imitate interactive partners. On stage, Rosetta Code let casts improvise with AI in any language.

I defended Humour-in-the-Loop, my PhD thesis on improvised theatre with interactive machine learning, at the University of Alberta. With Thomas Winters, I built TalkGenerator.com to generate slide decks on the fly for live speakers.

Improbotics launched with casts in Edmonton, London, and Stockholm, putting human ensembles on stage to improvise live comedy alongside an AI scene partner.

Piotr Mirowski and I debuted HumanMachine and A.L.Ex. at the Edinburgh Fringe. Those shows led to our first papers on performing theatre alongside AI and improvised comedy as a Turing test.

I put Pyggy, a chatbot trained on movie dialogue and paired with a projected puppet, on stage at Rapid Fire Theatre to improvise live scenes in front of an audience.