4 min readHere’s what you’ll learn when you read this story:
- In a new study, scientists successfully trained a brain organoid derived from mouse stem cells to solve an engineering benchmark known as the “cart-pole problem.”
- By applying weak or strong electric signals to reinforce certain behaviors via an adaptive AI training algorithm, researchers found that this small packet of brain tissue drastically improved its performance from a 4.5 percent success rate with random training to more than 46 percent with adaptive training and reinforced learning.
- According to a biologist unaffiliated with the study, this work shows that “the capacity for adaptive computation is intrinsic to cortical tissue itself, separate from all the scaffolding we usually assume is necessary.”
While creating organs sounds like sci-fi fodder, scientists have actually experimented with the idea for more than a century. For example, in 1907, American biologist Henry Van Peters Wilson demonstrated the foundational principles of lab-grown organs, or “organoids,” by showing how disassociated cells from a sea sponge could self-organize and regenerate in vitro. For decades, this exploration continued on in various animals until eventually, in 2009, scientists created the first 3D organoid using the intestinal stem cells of a mouse.
In the ensuing decades, organoids grew increasingly complex and provided new avenues for scientists to investigate how cells, tissues, and organs in the body work while also providing a cheaper, more ethical platform for developing potential therapies. One of the most compelling use cases for organoids has been in our never-ending investigation of the brain, the most complicated of all organs. In a new study published in the journal Cell Reports, a team of scientists from the University of California (UC) Santa Cruz successfully trained a brain organoid, developed from mouse-derived stem cells, to solve an engineering benchmark known as the “cart-pole problem.” Because these organoids lack a body, any sense of biological goals, or sensory experience, this learning behavior shows that adaptive computation may be inherent to cortical tissue itself.
“We’re trying to understand the fundamentals of how neurons can be adaptively tuned to solve problems,” UC Santa Cruz Ph.D. student Ash Robbins, lead author of the study, said in a press statement. “If we can figure out what drives that in a dish, it gives us new ways to study how neurological disease can affect the brain’s ability to learn.”
No larger than a peppercorn, yet loaded with several million neurons, the brain organoid was affixed to a special chip that allowed scientists to observe and control the firing of certain neurons. Using an electrophysiology system, the researchers used electric stimulation to send and receive information from the neurons, and to test this system, they relied on the classic “cart-pole problem,” which is a balancing task similar to keeping a vertical ruler upright on your hand while adjusting for movement and gravity. The researchers “taught” the organoid to balance the computer-simulated pole using weak or strong electrical signals. If the organoid couldn’t balance the pole for longer than the average time limit, it received “reinforcement learning” via an AI algorithm that selected which neurons to train.
“You could think of it like an artificial coach that says, ‘you’re doing it wrong, tweak it a little bit in this way,’” Robbins said in a press statement. “We’re learning how to best give it these coaching signals.”
What the scientists found is that random training achieved only a 4.5 percent success rate while adaptive training improved that rate drastically, to 46 percent. However, once the organoid rested for longer than 45 minutes, this short-lived learning all but disappeared. Because the organoid doesn’t have multiple brain regions like a human (or a mouse), it has no way of retaining long-term learning.
This isn’t the first time lab-grown brains have been trained to solve specific problems. Back in 2022, scientists trained a synthetic brain to learn how to play Pong, and more recently, other researchers developed neuromorphic platforms that use human brain cells. For the UC Santa Cruz researchers, the goal is to develop a brain organoid platform that can help scientists treat neurological disorders. But as research gets increasingly more advanced—especially when human-derived stem cells become involved—the experts are also pondering an equally important ethical question: How alive are these organoids, exactly?
“We want to make it clear that our goal is to advance brain research and the treatment of neurological diseases, not to replace robotic controllers and other kinds of computers with lab-grown animal brain tissues,” UC Santa Cruz’s David Haussler, a co-author of the study, said in a press release. “The latter might be considered cool, but would bring up serious ethical issues, especially if human brain organoids were used.”
Best-Selling Products We Recommend
Darren lives in Portland, has a cat, and writes/edits about sci-fi and how our world works. You can find his previous stuff at Gizmodo and Paste if you look hard enough.