Computer chips from Cortisol Labs with human neurons9. April 2020
Computer chips from Cortisol Labs with human neurons
Sidney, April 9, 2020
The Australian start-up company Cortisol Labs builds miniature brains with real, biological neurons embedded on a specialized computer chip. The Melbourne-based company hopes to be able to teach the hybrid mini-brains to perform the same tasks as software-based artificial intelligence, but at a fraction of their energy consumption. Currently, Cortisol Labs is working to get their mini-brains to play the old Atari arcade game Pong.
Cortical Labs used two methods to make the hardware: either it extracts mouse neurons from embryos, or it uses a technique in which human skin cells are transformed back into stem cells and then made to grow into human neurons. The neurons are then embedded in a nutritious liquid medium on a special metal oxide chip. This chip contains a grid of 22,000 tiny electrodes that allow programmers to provide electrical inputs to the neurons and also to record their outputs.
“We want to show that we can shape the behavior of these neurons,” said Hon Weng Chong, the company’s co-founder and Chief Executive Officer. Although it starts with Pong, Cortical Labs will be able to master this by the end of the year. Chong added that the company’s hybrid chips could eventually be the key to complex reasoning and conceptual understanding that today’s artificial intelligence cannot provide. The company’s method, if it proves to be scalable, also offers a potential solution to one of the most annoying problems that requires deep learning: It is incredibly energy intensive.
Neuroscientist Karl Friston of University College London, known for his work on brain imaging and theoretical underpinning of the functioning of biological systems, saw a demonstration of Cortical Labs’ technology earlier this year and was impressed with the company’s work. Aspects of Cortical Labs’ system are based on Friston’s work, but the neuroscientist has no connection to the Australian start-up company. According to Friston, the idea of trying to integrate biological neurons with semiconductors is not an idea he had expected. “But to my surprise and delight, they went straight to the real thing,” he said about Cortical Labs’ use of real biological neurons. “In my opinion, what this group is capable of is the right way to put ideas into practice.
The use of real neurons leads to the avoidance of several other difficulties that software-based neural networks present. For example, in order to tune artificial neural networks for learning, their programmers must go through a complicated process of manually adjusting the initial coefficients or weights applied to each type of data point that the network processes. Another challenge they face is getting the software to weigh up how hard it should try to explore new solutions to a problem, rather than relying on solutions that the network has already discovered to work well.
Chong, a former medical doctor, began looking for ways to develop hybrid biological computer intelligence systems about two years ago, along with his co-founder and Chief Technology Officer, Andy Kitchen. Chong said the two were interested in the idea of an artificial intelligence that would have the flexibility to perform almost any task as well or better than humans. “Everyone is running to build the AGI, but the only true AGI we know is biological intelligence, human intelligence,” Chong said. He noted that the two believed that the only way to achieve intelligence at the human level was to use human neurons.
Cortisol Labs works with mouse neurons and has long been used by neuroscientists as a surrogate for human neurons because long-established methods for extracting and cultivating them existed. Recently, scientists at the Allen Institute for Brain Science in Seattle found differences in the proteins that envelope the neurons of mice and humans. This means that they have different electrical properties, and these mouse neurons may not be good substitutes for human neurons.