Brain organoids, sentience and AI

What this gives us: a real case of 'sentience' being redefined as 'responds to input'. If that definition were applied to AI, the word would mean nothing.

2026-10-03

Checked by our internal skeptic review; corrections welcome.

How to read the tags

[REPORT] a finding reported in a study. [THEORY] an argument or framework. [OUR REASONING] our own inference, not a finding. Tags like [abstract], [full text] or [search snippet] say how much of a source we actually read.

Brain organoids are small 3D clumps of brain-like tissue grown from stem cells. Brackets show how each source was read.

Trujillo, Revah and Brainoware were read as abstracts only. The ISSCR and National Academies quotes come from search snippets that matched across several copies, but the full documents were not read.

1. What is measured

Electrical activity in organoids (Trujillo et al. 2019, Cell Stem Cell). The authors report that network events resembled some features of preterm human EEG [abstract]. [OUR REASONING] Partly resembling preterm brain waves is not the same as having experiences.
Organoids transplanted into rats (Revah et al. 2022, Nature). Human organoids in newborn rats' brains matured and received sensory input, and switching them on with light could drive reward-seeking in the rats [abstract]. The study does not test whether the human cells feel anything. [2026 Author Correction not read]
DishBrain (Kagan et al. 2022, Neuron). It is not an organoid but a flat layer of mouse or human-derived neurons on an electrode chip that improved at a simple Pong game [full text]; Balci et al. dispute the controls and effect sizes (see below). The paper states: "It is proposed that these neural cultures would meet the formal definition of sentience as being “responsive to sensory impressions” through adaptive internal processes (Friston et al., 2020)." [full text] Its declaration says Kagan is an employee and shareholder of Cortical Labs, that he and a co-author hold related patent interests, and two other authors had Cortical Labs funding for this work [full text].
Critiques of DishBrain. In a Neuron letter, Balci et al. (2023) write: "This application to neurons in vitro is in our view even more inappropriate and is not justified by the data presented in the paper." [full text] Birch (2024, ch. 1, p. 14) writes: "I see it as a mistake to define sentience in this way." He notes that by this definition all living cells, even those of brain-dead humans, might count. His "strongly recommend against" wording is in ch. 2, p. 23 [full text].
Brainoware (Cai et al. 2023, Nature Electronics). An organoid on an electrode array helped with speech recognition [abstract]. Its abstract claims nothing about sentience, and its authors declare no competing interests.
We checked commercial ties only where a paper declared them. We did not search for other authors' ties, for example in the Trujillo team.

2. Ethics guidance

ISSCR Guidelines 2021: "At this time, there is no biological evidence to suggest any issues of concern, such as consciousness or pain perception with organoids corresponding to CNS tissues, that would warrant review through the specialized oversight process." [search snippet]
National Academies 2021: "It is extremely unlikely that in the foreseeable future they would possess capacities that, given current understanding, would be recognized as awareness, consciousness, emotion, or the experience of pain." The finding adds that this should be revisited as organoids grow more complex [search snippet, Finding III.5].
Nuffield Council on Bioethics: It published an organoid report on May 12, 2026. Its press release states: "Although our report concludes that neural organoids do not currently possess the biological complexity associated with sentience, we note that a key part of guidance should be to help researchers identify features of organoids that may warrant heightened ethical scrutiny." [full text of press release; report not read]

3. Proven versus debated

Shown: Each measured effect in section 1 is reported in a primary paper.
Not shown: No study shows that any organoid or culture has experiences. None of the three bodies finds current organoids conscious or sentient (Nuffield from its press release only), and each expects to revisit the question.
Debated: The dispute is whether "sentience" applies to DishBrain at all. Balci et al. also question its controls and effect sizes [full text].

4. What this means for AI

[OUR REASONING] DishBrain illustrates the first word-swap in our glossary, where "sentience" is used to mean mere sensing. Its definition counts responsiveness to inputs, which AI systems already have, so a DishBrain-style claim about AI would empty the word. Living human brain tissue is not judged conscious just for being biological, so biology alone is not enough. Whether it is needed is open: substrate-neutral theories look at the computation, while some theories tied to biology could favor organoids over digital systems. Doing better at Pong or speech recognition shows learning, which does not by itself show sentience.