By now, many people will have seen my (possibly infamous) posts written for the Future Fund worldview prize, where I argue that AGI won’t happen any time soon.
To prove that life is stranger than fiction, this morning I saw an article which could change everything. Researchers in Melbourne, Australia have built a small network of real neurons which can learn the game of Pong through simple reinforcement learning. This configuration might some day be able to overcome the limitations I identify in artificial neural networks. For example, networks of real neurons might be able to learn real symbolic processes.
Of course it is way too early to make predictions about what these networks will be capable of in the future. I’m just very excited to see where this goes.
A strange twist on the road to AGI
By now, many people will have seen my (possibly infamous) posts written for the Future Fund worldview prize, where I argue that AGI won’t happen any time soon.
To prove that life is stranger than fiction, this morning I saw an article which could change everything. Researchers in Melbourne, Australia have built a small network of real neurons which can learn the game of Pong through simple reinforcement learning. This configuration might some day be able to overcome the limitations I identify in artificial neural networks. For example, networks of real neurons might be able to learn real symbolic processes.
Of course it is way too early to make predictions about what these networks will be capable of in the future. I’m just very excited to see where this goes.
https://www.smh.com.au/national/scientists-teach-brain-cells-in-a-dish-to-play-pong-opening-potential-path-to-powerful-ai-20221012-p5bp32.html
https://www.cell.com/neuron/fulltext/S0896-6273(22)00806-6?_returnURL=https%3A%2F%2Flinkinghub.elsevier.com%2Fretrieve%2Fpii%2FS0896627322008066%3Fshowall%3Dtrue