While much of the world debates the limits of silicon-based AI, a small group of scientists is taking a very different path: they grow real human neurons and train them to perform computations. Early biocomputers can already recognize speech, play Pong, and solve the simplest tasks. It sounds like science fiction, but it is part of scientific reality in 2025. Although the field is still far from creating a full “brain in a dish,” progress is so rapid that ethical questions are emerging already today.

From Organoids to “Organoid Intelligence”

The story began with brain organoids — three-dimensional structures that human stem cells naturally form in a Petri dish. Since 2013, they have been widely used in pharmacology to test Alzheimer’s treatments and model fetal brain development. In 2022, however, the Australian company Cortical Labs took the concept further: it connected an organoid of around 800,000 neurons to a silicon platform and succeeded in training it to play Pong, the classic arcade game from 1972.

The cells received electrical signals describing the position of the paddle and the ball, and in return learned to generate impulses to hit the ball back. Within five to ten minutes of training, accuracy improved, demonstrating that the organoid was indeed learning. This marked the first instance of living neural tissue showing targeted, adaptive behavior in response to external stimuli.

Since then, the term “organoid intelligence” has become more common, even though many neuroscientists consider it an exaggeration: the observed behavior is still primitive adaptation rather than anything resembling cognition.

How It Works in Practice

Modern biocomputers function as hybrids. An organoid—typically between 500,000 and 2 million neurons—is grown on a multi-electrode array (MEA) that simultaneously stimulates the cells and records their activity. Signals are translated into digital form, processed by a conventional computer, and then sent back as electrical impulses.

Current capabilities remain modest: basic sound recognition, simple reinforcement learning, and logic tasks at the “if A, then B” level. Their energy efficiency, however, is remarkable. Two million neurons consume about 20 milliwatts, which is millions of times less than the power draw of GPUs handling similar tasks.

Why This Technology Matters

  1. Disease modeling: studying conditions such as epilepsy, schizophrenia, and autism on living human neural tissue, which is more accurate than any computer simulation.
  2. Advanced toxicology: evaluating how chemicals affect the developing brain.
  3. Fundamental research: understanding how structured behavior emerges from seemingly chaotic electrical activity.
  4. Long-term vision: creating computational systems that learn like biological brains rather than like artificial neural networks.

Ethics Lag Behind

For now, regulators classify organoids as biomedical models rather than potentially sentient entities. The pace of progress, however, forces researchers to consider future implications. Some scientists argue that within 10 to 15 years it may become possible to grow structures containing billions of neurons, capable of complex responses or even rudimentary emotional states. This raises the question of where the boundary lies between tissue and a being that might require rights or protection.

A number of researchers are already calling for a temporary halt to large-scale expansion until clear ethical guidelines are established.

In Brief

Scientists have trained lab-grown human neurons to play Pong, recognize simple sounds, and perform basic reasoning tasks, marking the emergence of the new field of “organoid intelligence.” The capabilities remain extremely limited, but the energy efficiency and potential for brain modeling are enormous. The key question is no longer whether the technology will advance, but how society will respond when it advances too far — ethics is falling far behind technology.