

Discussion on Mortal Computations with Alexander Ororbia, Karl Friston, and Chris Fields
May 25, 2024
In this engaging discussion, Alexander Ororbia, a computational neuroscientist, teams up with theoretical neuroscientist Karl Friston and researcher Chris Fields to explore the groundbreaking concept of Mortal Computations. They tackle how morphology influences computation and the challenges of programmability in biological systems. With fascinating examples like two-headed planaria, they delve into the uniqueness of biological systems versus machine replicability. The trio also addresses the intricacies of memory storage, inferential dynamics, and the implications of design within Markov blankets.
AI Snips
Chapters
Transcript
Episode notes
Morphology Shapes Computation
- Structure and substrate jointly determine a system's computational capacities and should not be divorced in AI design.
- Mortal computation emphasizes morphology and changing substrates as central to what a system can compute.
Two‑Headed Worm Reprogramming
- Planarian flatworms retain a body-wide electrical pattern that encodes target anatomy and can be rewritten experimentally.
- Michael Levin describes changing that pattern to make worms regenerate with two heads despite normal genetics.
Programmability vs Uniqueness Tradeoffs
- Programmability trades off against uniqueness and replicability across systems from laptops to biological entities.
- Resource and energetic costs correlate with where a system lies between replicable von Neumann devices and unique biological morphologies.