One of my favorite topics is artificial intelligence, or – more specifically – what we can learn from neuroscience about artificial intelligence. So, when I was gifted the book “Life 3.0: Being Human in the Age of Artificial Intelligence” by Max Tegmark I enjoyed the read thoroughly. But, several scenarios envisioned in the book as paths to human-like artificial intelligence didn’t make sense to me, as a neuroscientist. So a bestseller book on artificial intelligence completely ignored the views of neuroscience.
This is why invited Dr. Grace Lindsay, host of the podcast “Unsupervised Thinking” about computational neuroscience and artificial intelligence. Grace is a postdoc at University College London, and she is currently writing a popular book about computational neuroscience.
Neuroscience inspired the technology that is currently leading the field in artificial intelligence: artificial neural networks (ANNs); now better known as ‘deep networks’ as in ‘deep learning’. The inventors of ANNs were the first to implement the basic idea of distributing computations across a large number of small processing units – neurons. For decades this method suffered from it’s need for large amounts of data and a lack of appropriate hardware. As soon as these prerequisites were met, ANNs really took off. Today, some people are thinking about how progress in neuroscience can further inform the structure of ANNs to improve on their performance – because they still are far behind what a brain can do.
Referring to Tegmark’s book we discuss scenarios that he writes are proposed to lead toward human-like artificial intelligence. We discuss whether modelling a human brain on different levels, from the molecules of every brain cell up to the behavior of an individual human, would work out – or would even count as intelligence.
Could we upload our minds? Would human-level AI be conscious? Will the “singularity” kill us all? We try to answer these questions form the viewpoint of neuroscience.