Look out gamers, there’s a clump of neurons in a dish that wants to take you on.
A Melbourne-led team of scientists have grown a network of neurons capable of playing the computer game Pong, in an extraordinary step towards understanding our brains and creating synthetic ones.
And in a next step that would make their undergraduate selves proud, they want to get “DishBrain” drunk and see how well it plays.
Scientists at Cortical Labs, CEOed by Dr Hon Weng Chong, grew a neural culture on a multi-electrode array out of mouse embryonic cells and human induced pluripotent stem cells differentiated into cortical cells, forming a “biological neuronal network” of about 800,000 neurons.
They simulated the arcade game by delivering inputs to specific electrodes and using activity by defined regions of the network to move a paddle up or down. Incorrect responses that did not result in interception of the ball by the paddle, i.e. misses, were “punished” with an unpredictable stimulus – which is about the only way to incentivise a system like this. Under the free energy principle, developed by UCL theoretical neuroscientist Professor Karl Friston, a system will “learn” behaviour that reduces unpredictability within it.
“One interpretation,” the authors write, “is that playing Pong generates more predictable outcomes than not playing Pong by reducing uncertainty. Note that a ‘miss’ results in unpredictable outcomes because the ball resets and its subsequent motion is unpredictable.”
It took the system only minutes to learn the game. It had to learn all over again each session, but this did not surprise the team since the network was made of cortical cells, which are not specialised for long-term memory.
Professor Friston himself has called the Cortical Labs work “a quantum leap forward” that brings us a step closer to creating synthetic brains. “The authors have managed to get a neural network to make sense of data from the world and act on the world at the same time.”
The authors conclude: “Ultimately, although substantial hardware, software, and wetware engineering are still required to improve the DishBrain system, this work does evince the computational power of living neurons to learn adaptively in active exchange with their sensorium. This represents the largest step to date of achieving SBI [synthetic biological intelligence] that responds with externally defined goal-directed behaviour.”
The way the system learned was distinct from silicon-based neural networks, the team found. Synthetic biological intelligence, they say, may arrive before artificial general intelligence, because of the inherent efficiency and evolutionary advantage, fine-tuned over billions of years, of biological systems.
In the meantime, they plan to introduce ethanol to the culture and see how it affects the game. Based on the evidence of your Back Page correspondent’s misspent youth, three schooners will be the sweet spot.
If you fancy a game of Beer Pong, drop penny@medicalrepublic.com.au a line.
