A curated, regularly updated library of 50 papers on computing with living neurons, from the foundational studies to this year's preprints. Every card links to the original paper, with a plain-English summary of what the authors did.
The papers that defined the field, in the order they appeared.
The authors developed cerebral organoids, 3D cultures from human pluripotent stem cells that form several interdependent brain regions. Using patient-derived stem cells, they modeled microcephaly and observed premature neuronal differentiation in patient organoids.
Human cortical organoids showed steadily increasing electrical activity over several months. Their network activity developed from periodic, regular oscillations into more irregular patterns, with synchronous events that resembled features of preterm human electroencephalography.
Using high-density CMOS microelectrode arrays and shank electrodes, the authors recorded spontaneous activity from human brain organoids. They found many weak connections around a skeleton of fewer strong ones, and field potentials that phase-lock to spiking.
The authors built DishBrain, which connects human or rodent neurons on a high-density multielectrode array to a simulated game of Pong. Guided by the free energy principle, cultures showed apparent learning within five minutes of gameplay that was not seen in control conditions.
This paper coins the term organoid intelligence and sets out a collaborative program for the field, covering standardized brain organoids, microfluidic perfusion, 3D microelectrode arrays, organoid-computer interfaces and an embedded ethics approach.
Participants of the first Organoid Intelligence Workshop (2022) call on scientists to explore brain organoid cultures for understanding the brain and for biocomputing, while addressing the ethical implications. They list advances needed in stem cells, interfaces and machine learning.
Stimulating rat cortical neurons in vitro with inputs generated by mixing two hidden sources, the authors found the neurons self-organized to encode each source. Variational free energy minimization quantitatively predicted the networks' responses and plasticity.
The authors present Brainoware, which uses a brain organoid on a high-density multielectrode array as an adaptive reservoir computer. Spatiotemporal electrical stimulation gave nonlinear dynamics and fading memory, and the system was applied to speech recognition and nonlinear equation prediction.
The authors describe the Neuroplatform, a hardware and software system for round-the-clock recording and stimulation of neural organoids, with automated microfluidics and a Python API for remote use. Over three years it was used with over 1,000 organoids.
41 papers
Analyzing local field potentials rather than single units, the authors found slow delta-range oscillations arising within organoids on microelectrode arrays. Oscillation strength tracked neuronal maturity, changed with stimulation, and shifted in an oxygen-glucose deprivation stroke model.
The authors report PULSE, a foundation model trained on over 900 hours of high-density microelectrode array recordings. It detected accelerated network maturation in 16p11.2-deletion organoids, and closed-loop control of a virtual inverted pendulum engaged existing spontaneous activity patterns.
The authors report INFORM, a flexible interface that wraps organoids and combines electrophysiology, two-photon calcium imaging, microstimulation and chemical delivery. It recorded over 371 days and resolved functional subnetworks, including evidence of inhibitory signaling.
The authors report a generative model that represents high-density microelectrode array spiking as a shared vocabulary of spatiotemporal motifs. Tested on 31 assays of brain organoids and human hippocampal tissue, the motifs were broadly reused across tissue types.
The authors report Brainobot, a system that uses brain organoid reservoir computing as a high-level decision layer for a humanoid robot. In proof-of-concept tests, it performed object grasping and laser chasing in a closed loop.
The authors report a task-based framework that measures how well cortical organoids on a microelectrode array represent and classify distinct stimulation patterns. This ability emerged with maturation and was reduced by drugs, an APP mutation and monocytes from Alzheimer's patients.
The authors report a reinforcement learning policy trained on a biophysically calibrated digital twin and transferred to cultured neurons, where it controlled network bursting better than heuristic controls while limiting stimulation. Calcium imaging showed how the policy used network topology.
This review traces brain-inspired computing from symbolic logic through artificial neural networks and neuromorphic processors to biohybrid computers built with living neural structures, and discusses the technical, biological and ethical challenges facing organoid intelligence.
The authors present an event-triggered workflow that uses an IoT messaging protocol to coordinate data transfer, storage and cloud computation for high-density microelectrode array recordings, demonstrated with cortical organoids and ex vivo brain slices.
The authors developed a multilayered microelectrode array of stacked mesh electrode layers separated by soft spacers, recording from multiple depths inside cerebral organoids. Integration lasted over four weeks and captured depth-dependent neural activity.
This review identifies four barriers to using human brain organoids to assess chemical risks to learning and memory, including limited maturity and standardization, and outlines scientific, technical and regulatory efforts to address them, using heavy metals as an example.
The authors built low-cost carbon nanotube 3D microelectrode arrays that fit standard culture plates. They recorded from 74 human cortical organoids, including neurotypical and Angelman syndrome organoids, which they describe as the largest scale reported in organoid electrophysiology.
This review examines how electrode and optical materials shape input stimulation and output recording in neural biocomputing, covering metals, conductive polymers, nanostructures and soft, transparent materials, plus multimodal platforms that stimulate and record at once.
In a nationally representative experiment, the authors studied public attitudes toward brain organoids linked to computer hardware and software. Attributing consciousness to these systems correlated with more ethical concern but also more perceived benefit and support for research.
The authors combined cultured cortical neurons, microfluidic devices and high-density microelectrode arrays in a closed loop. Training a linear decoder with fixed feedback let the networks generate periodic and chaotic signals, and modular microfluidic connectivity was key to learning.
Recording human forebrain organoids on high-density microelectrode arrays, the authors report that acute ketamine abolished population bursting and disconnected a subset of burst-driving backbone units, while chronic exposure produced tolerance to this effect.
The authors embodied mouse cortical organoids in a pole-balancing task through closed-loop electrophysiology. Training signals chosen by reinforcement learning improved performance for most organoids, though gains did not persist after a 45-minute rest and required glutamatergic transmission.
Using shell microelectrode arrays that wrap around neural organoids, the authors found that stimulation currents of 20 to 30 µA significantly increased firing rates, and they generated 3D spatiotemporal maps of neuromodulation across the organoid surface.
The authors combine a microfluidic chamber holding 10 brain organoids with a ten-shank 3D microelectrode array to deliver varying drug doses and record each organoid's activity, testing it with potassium chloride and SCN2A epileptic organoids.
The authors developed flexible, self-folding 3D microelectrode arrays that rely on strain differences between polymer layers. The devices recorded spontaneous action potentials and local field potentials from cortical organoids 800 to 1,500 µm across and from cardiac gastruloids.
Using CMOS microelectrode arrays with 236,880 electrodes, the authors recorded brain organoids from over 46,000 electrodes at once, resolving single-cell firing. They introduced propagation velocity and propagation area as network endpoints and tested responses to several drugs.
This perspective contrasts organoid intelligence, a top-down approach that aims for physiological realism, with a proposed bioengineered intelligence pathway that assembles modular, engineered neural circuits. It weighs the advantages and challenges of each for information processing.
This review summarizes how brain-on-a-chip systems for wetware computing are built from brain organoids, microelectrode arrays, electrophysiology interfaces and microfluidics, and surveys the encoding and decoding methods used to process their data.
This systematic review analyzed 24 ethical and philosophical publications on consciousness in human brain organoids. It finds that uncertainty about consciousness complicates the debate and suggests organoid intelligence may be a more tractable focus.
The authors report an embedded electrode array supporting pairs of connected human brain organoids. Trained organoids showed network-level long-term potentiation that NMDA receptor blockade prevented, and paired organoids improved at a closed-loop maze game with reinforcement feedback.
The authors built an automated IoT platform combining microfluidic feeding, electrophysiology and imaging to maintain cortical organoids. In 7-day studies of mouse organoids it enabled hourly recordings, and media exchange showed no acute effect on firing rate.
This review examines evidence that dissociated neuronal cultures show predictive coding, adaptive learning, goal-directed behavior and deviance detection. It links these to self-organized critical dynamics and the free energy principle, and discusses implications for neuromorphic and reservoir computing.
Borrowing from semiconductor 3D stacking, the authors vertically assembled neural organoids in a PDMS chamber. Stacked arrays showed inter-organoid connectivity and enhanced neural dynamics compared with single organoids and 2D arrays, and longer culture increased functional complexity.
The authors introduce virtual white matter, a system that detects activity in one microelectrode array dish and delivers precisely timed stimulation to another, creating real-time, bidirectional digital links between separate neural cultures.
This review covers flexible and stretchable electronics for stable, long-term interfaces across the 3D volume of organoids, multimodal charting of cell activity during development, and prospects for combining embedded electronics with organoid intelligence.
The authors report that networks of one, two or three connected organoids on high-density microelectrode arrays received daily repetitive stimulation. After two weeks, three-organoid networks decoded stimulus location better than simpler networks and showed changed connectivity.
This review maps ethical, legal and social issues in human brain organoid research across in vitro work, transplantation into animals and biocomputing, identifying many issues beyond consciousness and calling for a multidisciplinary approach.
The authors argue that if consciousness requires a unified field, possibly arising from the brainstem, the range of neural organoids of ethical concern narrows, with implications for the ethics of transplanting organoids into animals.
This paper proposes adding organoid intelligence, the use of brain organoids to study neuroplasticity, to developmental neurotoxicity testing so that chemical effects on plasticity can be measured, with AI methods to analyze the complex data.
The authors fused human thalamic and cortical organoids into assembloids that formed reciprocal axonal projections and synapses. Both connections showed short-term plasticity, and long-term potentiation and depression could be induced, through mechanisms that differed from rodents.
This piece argues that organoid intelligence could offer insight into the neuroscience of learning and memory and a biohybrid form of information processing, and names region-specific organoids, sensors, signal processing, AI and miniaturization as key advances.
This protocol describes a semi-guided method for growing cortical organoids that uses shorter induction steps and neurotrophic factors. The authors report that the organoids recapitulate the maturation of neural oscillations seen in the developing human brain.
The authors discuss the scientific status of brain organoids and organoid intelligence alongside consciousness, moral status, chimeras, informed consent and oversight, and argue for a balanced governance framework developed with scientists, ethicists, policymakers and the public.
This review summarizes how functional neural networks form in human brain organoids, how they are characterized with microelectrode arrays and calcium imaging, and how they have been used to study conditions such as Rett syndrome and Alzheimer's disease.
The authors set out three sets of moral concerns about human brain organoids: the potential emergence of sentience or consciousness, an analogy with artificial womb technology, and issues raised by biocomputing and human-animal chimeras.
Inspired by EEG caps, the authors made self-folding, optically transparent shell microelectrode arrays that wrap around organoids. They recorded from 400 to 600 µm organoids for up to four weeks, including responses to glutamate stimulation.
No papers match those filters.
Our guides explain these papers in context, without the jargon.
Organoid intelligence (OI) uses lab-grown human brain organoids on electrode arrays as living computers. Learn how it works, key results, limits and ethics.
Read article →
Cultured neurons can change behavior through closed-loop feedback. Learn how DishBrain taught neurons Pong, the free energy principle and the sentience debate.
Read article →
Reservoir computing pairs a fixed dynamical system with a trained readout. Learn how Brainoware used a brain organoid as the reservoir, its results and limits.
Read article →