Artificial Irreverence: Improvised Theatre with Robots
summary
Built two stage partners called Pyggy and A.L.Ex: A.L.Ex trained on 100,000 movies (≈1 billion tokens) and generates lines with a LSTM recurrent network conditioned by topic models, candidate ranking and turn‑taking logic. Voice I/O runs on a laptop (Lua/Torch for NN, JavaScript/C# for UI), the robot body is an off‑the‑shelf 16‑servo kit, and the project plays 30+ live shows including Edinburgh Fringe. It yields funny moments but suffers real failures—speech‑recognition errors, crashed motors and a burned transatlantic setup.
AI-written summary · about
chapters (14)
- 0:00Start
- 15:36Long Distance Relationship (London - Edmonton, Alberta)
- 23:02Statistical Language Models
- 24:48Recurrent Neural Networks
- 26:30Long Short-Term Memory (LSTM)
- 26:58Additional inputs: topic models enabling the chatbot to stay on topic
- 28:22Physical avatar
- 28:37A fun, personal, DIY project
- 32:18Exquisite corpses
- 33:22Improv is about doing the obvious thing
- 35:42Exploring (limited) artificial creativity
- 39:25Avoiding the uncanny valley
- 43:40Technology as a creative constraint
- 44:41Take aways
related talks
38mMaking Art with Artificial Intelligence: Artists in Conversation
44mIn the Brink of Consciousness
06mInterview with Jong Lee
28mThe Machine as Alien Ethnographer
39mChoreorobotics: Teaching Robots How To Dance
18mRobotics: why now?
49mMartin Förtsch, Thomas Endres & Jonas Mayer
55mEveryday AI: Practical Applications Beyond the Chatbot