TheDailyCurrents.com

Cortical Labs, a startup based in Australia, has developed what it describes as a “code-deployable biological computer.” Called CL1, the technology is a type of synthetic biological intelligence consisting of a combination of real neural networks and computer chips.

Human neurons are grown on a silicon chip, creating a fusion of brain cells and silicon. The silicon chip sends and receives signals to and from the neurons in a kind of feedback loop, similar to the way neurons send signals back and forth to each other. By integrating silicon and live tissue, computer code can be sent directly to the neurons.

Cells Grown on a Computer Chip

This sounds a bit like brain-on-a-chip technology, and there are some similarities. In fact, the immediate predecessor to CL1 was DishBrain, a network of brain cells in a dish. Brett Kagan, Chief Scientific Officer and Chief Operations Officer at Cortical Labs, led the team that designed DishBrain and taught it to play the classic video game Pong.

DishBrain’s cell cultures learned to track the ball and control the paddle in the popular tennis-style video game. The team published the results of their study on DishBrain in 2022 in the journal Neuron. CL1, Kagan says, “is the next evolution of brain-on-a-chip technology.”

The 800,000 neurons that make up CL1 are engineered from human skin cells and blood cells that have been turned back into stem cells, then reprogrammed to become brain cells. The cells are grown directly on a computer chip, with electrical contacts connecting the biological and the digital. Or as the company’s website puts it, “What digital AI models spend tremendous resources trying to emulate, we begin with.”

The cells are kept alive by a life support system that filters waste, provides nutrients, and regulates gases, acidity, and temperature.

Not So Traditional Training

The methods used to train CL1 are also innovative. Traditional machine learning was designed for silicon computing systems, says Kagan. The methods used here “take inspiration from neurocomputational theories on how the brain works,” he says. These theories include Karl Friston’s Free Energy Principle and Active Inference frameworks.

Nabil Imam, a researcher at Georgia Tech University who works on biological computation, describes this as a “niche technique” in machine learning, but adds that there is nothing about it that requires it to be in a dish.

“Whatever they did in the dish can actually be done in just a regular computer using neural networks,” he says.

What are the Advantages of CL1?

However, there are some distinct advantages to CL1. The system requires a fraction of the power used by conventional AI data centers, an important consideration in these days of rapidly worsening global warming. Also, it could potentially reduce the need for animal models in certain kinds of research.

Other advantages of CL1, Kagan adds, are the ability to learn with very limited data, the ability to generalize, and the ability to deal with fuzzy data and changing dynamic environments in real time. After all, as the company’s website states, “The neuron is self-programming, infinitely flexible, and the result of four billion years of evolution.”

At the moment, the most likely uses for CL1 are for drug development, personalized medicine, and neuroscience research, though Kagan says that people are exploring CL1 as an alternative to traditional AI and robotics technologies.

“The goal of this technology is not to replace current computing methods, but to provide a better tool for where current methods fail or require huge amounts of data or power to train,” he adds.

Is it a better approach than just silicon? Not necessarily, according to Imam. However, AI is still in its early days, he says.

“We don't know the best way of solving many of these problems, and we do know that the brain solves many, many things very efficiently. It’s the most efficient intelligence system that's out there.” And now it has a silicon partner.

Article Sources

Our writers at Discovermagazine.com use peer-reviewed studies and high-quality sources for our articles, and our editors review for scientific accuracy and editorial standards. Review the sources used below for this article: