
Tech • AI • Robotics • Game
⏱ ≈ 3 min read · 12-min video — 9 min saved
Iris, a humanoid robot in San Francisco, uses rented networks of living human neurons in Australia and Switzerland to influence its social responses, offering an early but limited test of hybrid biological and silicon robotics.
Iris was developed by Yann LeCun, a Stanford bioengineer and founder of robotics company Open Mind. The robot remains largely conventional in its vision, language and control systems, but living neuron cultures are used to shape how it responds to people. Those cultures help determine the robot’s mood and whether it reacts warmly or coolly during interaction.
The biological component is not inside the robot. Open Mind rented access to commercial platforms holding human cortical neurons grown from reprogrammed adult cells. One platform used flat neuron cultures from Cortical Labs in Australia; another used three-dimensional neuron spheroids from a Swiss provider. Each culture sits on electrode arrays that both stimulate the cells and record their activity.
Cameras on Iris capture visual input, which software converts into electrical pulse patterns and sends over the internet to the neuron platforms overseas. The neurons respond with electrical spikes, and those signals are translated back into outputs that affect the robot’s behavior. The biological layer acts less like a full brain than a reflexive social module layered on top of standard computing.
In nearly 500 stimulation rounds, fresh flat cultures produced correct readings up to 94% of the time and responded in less than 0.25 seconds. The spheroids reached up to 91% accuracy, but some responses took as long as 2 seconds. Outside researchers called the work an important first step, while noting that natural human sensorimotor loops typically close in 10 to 100 milliseconds.
Robots controlled by living neurons are not new. In 2000, researchers linked lamprey cells to a robot. In 2002, Steve Potter’s lab at Georgia Tech demonstrated a “hybrod,” a small robot controlled by rat neurons on a chip, with loop times under 0.1 seconds. Other projects used rat neurons to control a remote drawing arm, fly an F-22 simulator, and steer a wheeled robot called Gordon.
The novelty lies in combining human neuron cultures, a humanoid body, internet-linked biological computing across continents, and commercial access to the cells as a service. That market is expanding. Swiss startup FinalSpark opened a remote Neuroplatform in 2024, with reported access priced around $500 a month, offering tiny human brain-cell organoids connected to electrodes and monitored continuously online.
The biological systems are fragile and temporary. The flat cultures used with Iris survived more than 100 days, while some spheroids lasted only a few days. As the cultures aged, response lag increased and the robot’s temperament shifted. That means the same robot could react differently over time, and once the cells die, that part of the system is lost because it cannot be straightforwardly copied or backed up.
Rather than treating biological drift as a flaw, LeCun has argued that mortality, inconsistency and individual variation may make robots feel more relatable. He has suggested that a partly living home robot could be harder to duplicate and better for privacy because information stored in neurons is difficult to extract. The broader aim is to test whether living neural tissue can do useful work in robots, not to replace conventional computing.
Experts say the system’s capabilities remain narrow and heavily dependent on ordinary electronics. Some see potential in lower-energy computation and learning from less data, but doubt there is yet a realistic path to scalable real-world deployment. On ethics, researchers say simple dish-grown neuron cultures do not appear to have the circuitry needed for suffering, though donor consent for stem-cell-derived tissue remains essential.
Iris shows that living human neurons can influence a real robot’s social behavior, but only in a modest and highly experimental way. The central challenge now is reliability: keeping the cells alive longer, increasing electrodes, and determining whether such hybrid systems can progress from curiosity to practical robotics.
Ask a question