Organoid intelligence

• 1. Introduction

  ›

  Thomas Hartung of Johns Hopkins University drew a sharp line in 2023. "While silicon-based computers are certainly better with numbers," he said, "brains are better at learning." The field taking that argument seriously is organoid intelligence. It grows three-dimensional cultures of human brain cells in laboratories and connects them to computing systems. The cell cultures are called brain organoids. The systems they form go by the shorthand OI, and the phrase "nervous filesystem" has also been used. Hartung pointed to a concrete limit: transistor density in computer chips may be approaching its physical ceiling. The brain, wired in an entirely different way, may not face the same constraint. Whether living tissue can actually step into computing roles that silicon is struggling to fill is what the field is trying to find out.

  126

• 2. How Brains Differ

  ›

  Organoid intelligent computer systems fall into a class researchers call biohybrid systems, blending biological material with engineered hardware. The biological side of that combination brings unusual computational properties. Human brains are slower than machines at processing simple information. That speed disadvantage disappears when tasks become more complex. Brains can handle fewer and more uncertain data without breaking down. They perform sequential and parallel processing at the same time. Their internal structure is highly heterogeneous. Incomplete datasets, which tend to trip up conventional algorithms, are something brains navigate without difficulty. Some researchers argue that this suite of properties means brains outperform non-organic machines in decision-making. Brains are also substantially more energy-efficient than silicon hardware and can hold large quantities of information. Brain-inspired silicon chips exist to close some of this gap. Most of them are designed on digital electronic principles, which limits how closely they can replicate what biological tissue does natively. The most direct test of what an organoid itself could do came from a study the researchers named Brainoware.

  162

• 3. What Brainoware Computed

  ›

  A high-density multielectrode array connected the Brainoware organoid to its computing environment. The stimulation sent to the tissue was spatiotemporal: calibrated across both location and time within the electrode grid. Nonlinear dynamics in the organoid were not a problem to overcome. Fading memory, another property of biological tissue, was put to use rather than filtered out. The organoid also showed unsupervised learning: repeated exposure to training data reshaped its functional connectivity without any explicit programming. The framework used to interpret these behaviors was reservoir computing. That approach treats a complex physical or biological system as a computational resource in its own right. Within that framework, the system performed speech recognition. It also predicted outcomes in nonlinear equations. Those two tasks, drawn from fields as different as language and mathematics, showed the range of what a biological computing system could attempt. Infrastructure for running experiments like these at scale was built separately, under the name Neuroplatform.

  53

• 4. Neuroplatform

  ›

  The Neuroplatform allows researchers to experiment with brain organoids from a distance, using an application programming interface. Physical proximity to the biological material is no longer required. OI generates complex biological data, and the tools needed to decode it come from bioinformatics. Those tools uncover patterns and insights that raw output alone cannot reveal. Training an OI system follows a different logic from training a conventional artificial intelligence. AI systems learn through machine learning, an algorithmic process applied to data. OI systems use biological learning instead. In biological learning, the tissue physically restructures itself in response to experience rather than having parameters adjusted by code. OI research may also yield insights into how the brain develops and how learning and memory work at a cellular level.

  41

• 5. Consciousness and Consent

  ›

  Two ethical concerns have been named in OI research. One is whether an organoid could develop sentience or consciousness. The other involves the relationship between a stem cell donor and the OI system grown from their cells. The first concern has not been resolved. Researchers take it seriously while noting that current organoids are far from the kind of complexity associated with conscious experience. The second raises questions about what obligation, if any, connects a donor to the system that emerged from their biological material. Those are among the harder questions. Dementia is one of the neurological disorders the field names as a target. The same organoids at the center of its ethical debates may also be the tools that eventually contribute to answers there.

  162