A Doctor in a Cell

This week, in a conference hall on Pennsylvania Avenue, researchers gather for the first International Conference on Organoid Intelligence. Ask a search engine about “eukaryotic AI” today and an obliging chatbot will describe an emerging field in which living cells are programmed to think. No recognised field goes by that name. The real fields behind the label, including the one meeting in Washington, are more modest and more interesting.

[https://organoidintelligence.org/icoi/]

Start with the cell. A eukaryote is any organism whose cells keep their DNA inside a nucleus: yeast, oak trees, octopuses, us. What makes the eukaryotic cell remarkable is less its size than its history. Well over a billion years ago, one simple cell engulfed another and, rather than digesting it, kept it. The captive became the mitochondrion, the power plant inside nearly every cell in your body. When Lynn Margulis, then publishing as Lynn Sagan, made the case in 1967, journals turned her paper away repeatedly. Today it is in every textbook. The most consequential innovation in the history of life was a merger.

That matters, because what is now being attempted is another merger: living cells wired into machines.

The best known effort is Australian. In 2022, Cortical Labs in Melbourne grew neurons on a chip and taught them to play Pong. In March 2025 it launched the CL1, billed as the first commercial biological computer, at roughly US$35,000 a unit. In Switzerland, FinalSpark rents remote access to human brain organoids, small spheres of neural tissue grown from stem cells, by the month. The pitch is energy. A human brain runs on about twenty watts, a dim bulb’s worth, while the data centres behind modern AI now use about as much electricity as France.

[https://corticallabs.com/]. [https://finalspark.com/]

Here the honest answer to “could this take shape?” is: not in the form the energy pitch suggests. The organoids die. FinalSpark’s typically last around a hundred days on the electrodes, its best so far seven months; the CL1’s cells survive up to six. A processor that must be fed, kept warm and replaced every season is no rival to silicon. Madeline Lancaster of Cambridge, who grew the first brain organoids to study developmental disorders, called the grander claims science fiction: intriguing, but not yet backed by the science. Last November, several brain organoid pioneers warned that inflated talk of “intelligence” could provoke sweeping laws that choke off the medical research that actually helps patients.

But the more interesting version of the idea was never about a better chatbot. It was about placing a small, dim, reliable computer inside the body, where silicon cannot follow. That idea first took working form in Rehovot.

In 2004 Ehud Shapiro’s team at the Weizmann Institute described in Nature a computer built entirely from DNA and enzymes. Its program checked a sample for the molecular signature of prostate cancer and, only if every marker agreed, released a strand of DNA designed to disrupt the cancer cell. If even one marker read normal, it released a second strand that neutralised the first. An earlier version was listed by Guinness as the smallest biological computing device. Everything happened in a test tube. Shapiro, a software pioneer who had taught himself molecular biology, called his goal a doctor in a cell, and was candid that it lay decades away.

[https://en.wikipedia.org/wiki/Ehud_Shapiro]

The decades are passing. One of his students, Yaakov Benenson, went on via Harvard to ETH Zurich and in 2011, with Ron Weiss of MIT, placed a logic circuit inside living human cells. It read five molecular markers and set off the cell’s own suicide programme only when the profile matched a cervical cancer line. Other cell types were left alone. Cancer therapies built on the same logic, though with different molecular machinery, immune cells engineered to strike only when two conditions are met, have since entered clinical trials. Benenson’s verdict on conventional medicine is blunt: we still tend to treat disease with a single drug, “regardless of how complex the biology and the causes of diseases may be.”

[https://radar.gesda.global/people/yaakov-benenson]

This is eukaryotic computing as it really exists. It does not think or converse. When Benenson was still a doctoral student, the Harvard geneticist George Church praised the approach for avoiding “an abstract race with existing computers on their own turf.” It performs AND and NOT with molecules, in the one setting where a single correct decision is worth more than a billion fast ones. That future is far more plausible than racks of brain tissue answering email, and far more valuable.

[https://www.kcl.ac.uk/people/george-church]

There is a lesson in this for Israel, and it is not a comfortable one. Nobody in 2001, when Shapiro’s first molecular computer appeared, could have written a business case for it. The state has since discovered the field he helped open. Its National Bio-Convergence Program has put NIS 276 million into research infrastructure between 2023 and 2025, and reported in April that the sector has grown to 186 companies. That is the harvest. The seed bed is another matter. In June 2025 Iranian missiles struck the Weizmann campus, reducing laboratories to rubble and destroying irreplaceable samples. Israeli researchers won 30 early career grants from the European Research Council in 2024; this year they won 11, a fall only partly explained by Switzerland’s return to the scheme and changed criteria, and Weizmann won one. A country can fund the applications of curiosity only for as long as it keeps funding the curiosity.

The second lesson concerns language. “Eukaryotic AI” is what emerges when a machine assembles plausible words without understanding them. The irony is hard to miss: the genuine article, a cell that computes, is being explained to the public by a computer that cannot tell real science from a convincing phrase.

Reports in 2004 said a single drop of water could hold around a trillion of Shapiro’s devices. That drop has long since dried and the tube been emptied. The hall in Washington empties on Tuesday, after two days of talk about intelligence. What should remain from both rooms is the smaller idea from Rehovot: not a cell that thinks, but a cell that knows exactly when to act.