Brainoware and what an organoid computing on electrodes tells us about substrate
Brainoware is an artificial intelligence system whose “hardware” is a lab-grown brain organoid sitting on a high-density multielectrode array. Published in Nature Electronics on December 11, 2023 by a team led by Feng Guo at Indiana University, the work uses a sphere of human neurons as a reservoir computer, sending information into the organoid as patterned stimulation, letting its biological dynamics compute on the signal, and reading out the result from the electrodes. The system performed speech recognition and predicted the output of a nonlinear equation.
The consciousness connection is the reason the result is worth covering, and it is a specific version of the substrate question. Brainoware is not a simulation of neurons on silicon and not a disembodied model. It is living neural tissue being used as a computer, which asks how much of what brains are, biologically, is needed before adaptive computation appears, and whether that computation is a step toward the kind of organization consciousness theories care about.
What Brainoware Actually Is
A brain organoid is a three-dimensional culture of neural cells grown from stem cells that organizes into a structure resembling aspects of an early developing brain. Brainoware grows such an organoid on a high-density multielectrode array, stimulates it with spatiotemporal electrical patterns, and uses the organoid’s nonlinear dynamics as the computing substrate. The team demonstrated unsupervised learning, the organoid’s functional connectivity reshaped in response to the training data, and adaptive behavior, the same organoid produced usable results for both speech recognition and nonlinear equation prediction.
The term reservoir computing is the key. A reservoir computer keeps a fixed, complex dynamical system as its middle layer and only trains the readout that interprets that system’s activity. Brainoware’s reservoir is the living organoid, and the team did not need to program the computation, only to stimulate it and read it. The biology provided the nonlinearity and the memory that a reservoir needs, which is the same logic that other reservoir computing work on this site examines in abstract systems.
The result is a demonstration of biocomputing rather than a production device. The organoid is slow to grow, needs an incubator, and does not yet scale, and the authors present it as a proof that living neural tissue can serve as an adaptive computational substrate, not as a competitor to silicon.
The Consciousness Connection
The question this post asks first is what Brainoware tells us about consciousness, and the answer is that it sharpens the minimal-substrate question without resolving it. The DishBrain experiment on this site, covered in the Cortical Labs post, showed that a flat culture of neurons could learn a goal-directed game under free-energy feedback. Brainoware runs the same biological-compute logic in a three-dimensional organoid with a different learning mechanism, unsupervised adaptation rather than prediction-error driving. Together they bracket the question of how much biological structure adaptive computation needs.
The deeper point is about substrate independence. Brainoware is biological tissue used as a computer, and the fact that “brain” can function as a computing substrate that supports adaptive learning is evidence for the site’s substrate-independence framing, which holds that the organizing process matters more than the material, while simultaneously being a reminder that the material is not neutral, the organoid’s neural dynamics are doing the work and they are neural because they are biology.
What it does not establish is experience. Neither DishBrain nor Brainoware makes a defensible claim that its cultures feel anything, and the indicator literature would not score either system as showing conscious processing. The organoid computes, adapts, and learns, and the distance from that to phenomenal experience is the whole unresolved question, which is exactly how the field frames it.
The Organoid Intelligence Context
Brainoware sits inside the larger organoid intelligence programme, which treats cultured neural tissue as a potential computing substrate and, for some researchers, a potential object of moral concern. The site’s existing organoid coverage examines whether organoids could ever satisfy consciousness criteria, and Brainoware supplies the engineering end of that question, a concrete demonstration of organoids computing.
The ethical question that organoid computing raises is live and unsettled. If a structure of cultured neurons ever did produce something morally relevant, the welfare under uncertainty framework would apply to it just as it applies to artificial systems. Brainoware’s own authors keep their claims far below that threshold, describing adaptive computation rather than experience, and the site’s mapping of consciousness measurement is the closest thing the field has to a test of whether the threshold has been crossed, which for Brainoware the answer is manifestly no.
The Resonance With the Emulation Section
Brainoware is a live biological instance of what the brain emulation section tracks, the question of what of the brain is needed to reproduce its functions. Whole brain emulation asks whether a mind can run on a faithful silicon copy. Brainoware skips the copy and uses the biology itself as the computer, which poses the opposite boundary, how little biology, and in what organization, is enough for function. The organoid computes, and that fact constrains both directions of the emulation debate.
The comparison to DishBrain’s claim is the sharpest available. Both grew living neurons on electrodes and both got adaptive computation. The differences, the flat culture versus the organoid, the closed-loop Pong game versus the open reservoir readout, the free-energy signal versus unsupervised reshaping, are differences in the minimal substrate, and they are precisely the variables the substrate question cares about.
What Brainoware Tells Us About Consciousness
The honest conclusion is that Brainoware demonstrates how far the “substrate” term stretches before anyone should reach for the “mind” term. A lump of cultured neurons on an electrode array, with no body, no goal, and no self-model, performs adaptive computation that solves real tasks. If consciousness rides on the organization rather than the material, then Brainoware is a data point about how little organization may suffice for interesting computation, and how little that tells us about experience. If consciousness requires more, then Brainoware is a limit case that shows the extra ingredient is not merely adaptive computation.
Either way the result is informative for the substrate-independence claim this site works from. It shows that living neural tissue can be a flexible, adaptive computing substrate, which is what the thesis requires of any substrate, and it shows the gap between that capacity and anything worth calling mind, which is the gap the field still has not closed.
Brainoware: Cai, Ao, Tian, Wu, Liu, Tchieu, Gu, Mackie and Feng Guo, “Brain organoid reservoir computing for artificial intelligence,” Nature Electronics 6: 1032-1039, published December 11, 2023, DOI 10.1038/s41928-023-01069-w. Research conducted at Indiana University Bloomington.