The creation of AI has been one of the biggest achievements in the history of humanity, and it could help unlock secrets that are fundamental to our very existence.
Google DeepMind CEO Demis Hassabis has said that the progress of AI has been enabled by technical advancements inspired by the workings of the human brain, and it will in turn help us understand our brains even better. “Originally, with the field of AI, there was a lot of inspiration taken from architectures of the brain, including neural networks and an algorithm called reinforcement learning,” he said at an event. “Then we’ve gone into a kind of engineering phase now, where we’re scaling these systems up to massive size (to create) all of these large foundation models or language models. There’s many leading models now,” the Nobel prize winner added.
“And I think we’ll end up in the next phase where we’ll start using these AI models to analyze our own brains, and to help with neuroscience as one of the sciences that AI helps with,” Hassabis predicted. “So actually, I think it’s going to come sort of full circle. Neuroscience has sort of inspired modern AI. And then AI will come back and help us, I think, understand what’s special about the brain,” he said.
Hassabis seemed to be referring to neural networks, which are the fundamental building blocks of modern LLMs like ChatGPT and Gemini. These are computer networks that were modeled after neurons in a brain, and with interconnections between them. The theoretical base for neural networks was proposed in the late 1800s, but it was until much later that computer systems were developed to be able to implement them. In the early 2000s, neural networks went out of fashion because they weren’t able to replicate results that researchers had anticipated. But the use of GPUs to train these networks in the 2010s through deep learning led to some interesting outcomes, and the discovery of add-on technologies like the transformer have turned them into the powerhouse they are today. Today, neural networks underpin programs like ChatGPT, and help create “intelligence” through thin air.
And Demis Hassabis believes that this intelligence will help push forward field like neuroscience, and help us understand the human brain. It will be a nice circular moment for AI — the human brain inspired much of the modern progress in AI, and it will help us understand the human brain a lot better. And if this progress recursively continues, there’s no saying what we might end up creating with these neural-network inspired architectures.