No computation without representation: Abstraction/Representation Theory applied to neuromorphic and neural wetware computers

Authors: Aslan SAtary Dizaji

Presentation type: Poster at SNUFA 2026 online workshop (4-5 Nov 2026)

Abstract

Neuromorphic engineering and neural wetware (synthetic biological intelligence, SBI) are the two standing answers to the energy asymmetry between machine learning and brains. Both are routinely described as “computing with neurons”, yet the sense in which each of them computes has never been made precise. We apply Abstraction/Representation (AR) theory—the framework of Horsman, Stepney, Wagner and Kendon for deciding when a physical system computes, and Horsman’s extension to heterotic (multi-substrate) computers—to both paradigms. AR theory asks, for any putative computer: is there a good, valid theory T of the device; a representation relation RT that can be run in both directions (encoding and decoding); at least one fundamental operation; and a commuting diagram between abstract and physical evolution that supports a compute cycle, in which the physical evolution is used to predict the outcome of an abstract one? We show that digital neuromorphic hardware satisfies these conditions by construction (its diagram commutes with ε= 0, its theory is a refinement stack that reaches the transistors, and it scales by composition), that analogue and memristive neuromorphic devices satisfy them only up to a calibrated ε with an increasingly phenomenological T, and that present neural wetware systems—DishBrain/CL1, Brainoware, and the Neuroplatform—mostly do not yet satisfy them: their device theory is phenomenological or normative rather than predictive, their adaptive state admits no instantiation relation, and the cycles they actually execute are experiment and engineering cycles rather than compute cycles. The one clear wetware compute cycle (organoid reservoir computing) obtains it by moving the adaptive part of the computation into the representation relation. The two paradigms are mirror images in the AR diagram: neuromorphic hardware is technology (abstract-first, “history”), wetware is science (physical-first, “prophecy”); and at the level of composition, neuromorphic-plus-host is hybrid while the closed-loop wetware architecture is genuinely heterotic. AR theory thereby reframes the field’s open problems—credit assignment, benchmarks, energy accounting, state persistence, and terminology—as precise questions about T, R, and ε, and yields an AR-compliant benchmark protocol for SBI. The lesson for the SNUFA community is that the substrate does not decide whether something computes; the representation does.

Are you the author and do you want to add some related links to this page? If so, check out our guide to adding related links and videos.