Today’s AI sucks up a lot of power that has raised significant environmental and societal impact concerns. Organoid intelligence (OA) — intelligence powered by living cells — might help us get around the environment and impact concerns.

Grok-4 required about 310 gigawatt hours of energy to train. In Texas, a kilowatt-hour costs about 14 cents. At this rate, 310 gigawatt-hours would cost over 43 million dollars.

Data centers can impose substantial societal and environmental burdens through heavy electricity consumption and water use for cooling.  Elon Musk has proposed an ambitious workaround to the exploding energy demands of artificial intelligence by putting solar powered data centers in space.

By contrast, the human brain uses roughly the power of a dim incandescent light bulb. It can run on pizza and beer.

Hence the growing interest in a research field called low-power organoid intelligence.

Organoid intelligence is an emerging field that explores the use of lab-grown clusters of human brain cells — called brain organoids — as biological computing systems.

Human brains are alive so they adapt and change via neuroplasticity. For example, I used to be addicted to cigarettes, and quitting the addiction was difficult. Since I quit many decades ago, however, neuroplasticity has rewired my brain.  I no longer have any desire to smoke a cigarette. Neuroplasticity adapts the brain.

The error backpropagation algorithm is the system that adapts most artificial intelligence. Today’s AI systems were modeled on biological phenomena, including the McCulloch-Pitts computational model of the neuron and Hebb’s law of interacting neurons. For decades we tried to make computers think like brains. Now we’re asking brains to compute like computers.

How far have we come?

Proof-of-principle applications of organoid intelligence have thus far been limited to simple toy problems.  Substantial progress remains before the technology becomes competitive. Beyond cost and environmental considerations, successful organoid intelligence must ultimately demonstrate clear and significant performance advantages over conventional computing approaches.

Here are a few organoid experiments performed to date, 

Pong: Those old enough will remember the game of Pong  which was the first commercially successful video game. Researchers taught organoid intelligence  to play Pong.  The organoid adapted its firing patterns in response to feedback and improved gameplay performance.

Prediction & Classification.  Organoids were used to successfully perform nonlinear signal prediction and speech classification tasks demonstrating real computation using living neural tissue.

Human Neurons in Mouse Brains. Researchers transplanted human brain tissue into mouse brains  where the organoids  matured  and formed functional synaptic connections with host neural tissue. This showed that organoids can process biologically meaningful signals in living systems rather than merely surviving in a dish.  (I find such research spooky.) After decades of trying to make silicon think like the brain, we may now turn to biology to overcome silicon’s AI demands. If organoid computing succeeds, the next generation of computers may need both software updates and daily feedings.