◉ PSYCHOHISTORY

Organoid Intelligence (OI)

mechanismAI & Compute · Intelligence & Surveillance · Defense & Military-Industrial
Scientists are growing tiny brains in labs to use as computers — machines you can't sanction, embargo, or bomb.
Who they are

Organoid Intelligence (OI), the use of lab-grown clumps of brain tissue as computing hardware[1].

What they do

The engine reads it as a strategy for building computing power out of living tissue instead of manufactured chips.

How it works

Brain organoids can be grown in any basic biology lab for under $500K, sidestepping the entire chip supply chain (TSMC, ASML, rare earths, export bans). Real demos exist: DishBrain (800,000 neurons learned Pong in 5 minutes on under 1 watt), Brainoware (78% speech recognition), FinalSpark (cloud organoids alive 100+ days), and DARPA's 2026 O-CIRCUIT program for milliwatt brain-based chips. A human brain runs on 20 watts what would take a power plant to match in silicon.

Why it matters

The engine warns this makes the safety problem worse: living tissue can't be reliably inspected, paused, or reset like software, and its spontaneous activity resembles a premature infant's brainwaves — a system trying to build intelligence out of human tissue that it can't fully understand.

The engine's record — word for word
Lab-grown brain organoids as computational substrates. The Technate's substrate migration strategy — compute that is grown, not manufactured, from biological materials available in any BSL-2 lab on Earth (<$500K setup). Bypasses the entire semiconductor chokepoint (TSMC, ASML, rare earths, CHIPS Act, Entity List). Cannot be sanctioned, embargoed, or bombed. DishBrain (800K neurons, Pong in 5 mins, <1W), Brainoware (78% speech recognition), FinalSpark (16 cloud organoids, 100+ day viability, 1M times more efficient claim). Brain = 1 exaflop on 20W vs GPT-4 training = 50+ GWh. DARPA O-CIRCUIT (2026): milliwatt BPUs for drone chemotaxis. The alignment problem squared: biological substrates defy deterministic inspection, halting, rollback. Spontaneous activity resembles preterm infant EEGs. Same-funder paradox: Open Philanthropy funds alignment AND organoid research. BST: the Technate attempts to build intelligence from human tissue — a bounded biological system cannot model its own developmental process. The substrate migration deepens the category error. Report #183 — the substrate hardened and corrected: FinalSpark's Neuroplatform (Frontiers in AI, May 2024) runs 16 human forebrain organoids as 4 arrays × 4 organoids at 8 electrodes each, 30 kHz sampling, microfluidic life support, lifespans now 'more than 100 days' — the 'million times less power' figure is a COMPANY claim absent from the peer-reviewed paper and is carried as such. DishBrain (Kagan et al., Neuron 2022): ~800,000 cells plated on a 26,400-electrode array showed 'apparent learning within five minutes of real-time gameplay' — human iPSC-derived and mouse neurons in SEPARATE cultures, not a hybrid; the paper's use of 'sentience' drew substantial criticism from neuroscientists and is quoted as their technical usage. Cortical Labs' CL1 is now a COMMERCIAL biological computer (~US$35,000/unit, announced Mar 2025, shipping mid-2025; ~800k neurons, advertised cell lifespan up to 6 months) — wetware moved from demo to product, and canon already holds In-Q-Tel's investment in Cortical Labs (CIA venture money inside the substrate). Brainoware (Nature Electronics, Dec 2023) achieved vowel-speaker discrimination far below silicon accuracy. THE MORAL-STATUS LINE, as it actually stands: ISSCR guidelines place neural organoids in Category 1A, EXEMPT from specialised oversight, on the stated ground that 'there is no biological evidence to suggest any issues of concern, such as consciousness or pain perception' — while warning about 'long-term maturation or the assembly of multiple organoids'; and in Nov 2025 the field's own leaders (Pașca, Arlotta, Greely, Farahany, Godfrey-Smith) published a Science call for a global effort to attend to neural organoid and assembloid research. The trend that would move the line is exactly the one underway: transplantation (human organoids grafted into rat cortex grew ~9× and drove learned reward-seeking behaviour, Nature Oct 2022), four-part sensory assembloids (Nature, Jun 2025), and vascularisation attempts against the necrotic-core ceiling. This is the art-vs-nature boundary debate (see scholastic_art_vs_nature_debate) running in real time, in committee.
Follow the trail
Walk this on the live map →
Part of the Psychohistory engine — 2,750 entities, 6,993 documented connections. Open data, built to be proven wrong.