Just reading Max Tegmark’s Life 3.0. Its a great overview of where we are with AI and our world. Not so detailed on how AI algorithms work, and certainly not as technically informative as his awesome Mathematical Universe (see Books of 2015), but broad, conversational and thought-provoking on the impacts on society. He is sure a thoughtful fellow.
Tegmark, like his mates Hawking and Musk, worries about how a super-intelligent AGI (artificial general intelligence) will eventually escape from its creators air-gapped servers “into the wild”. I agree - and will go further. Tegmark et al. are calling for an ethics of AI research, protocols to prevent the escape. But I suspect its probably inevitable. There’s going to be too much competition and ambition, too many resources available online, too many things that can go wrong.
Tegmark outlines a couple of examples that brought to mind Alex Garland’s beautifully disturbing movie Ex Machina. In Garland’s future dystopia, the AGI-robot Ava escapes because she learns how to exploit Nathan’s arrogance, Caleb’s need for love (Caleb Smith, hint hint!!) and how to tell when a human is lying. We’re too easy, even the smartest of us. While Tegmark is talking about a code-based AGI, rather than one hosted in a super-cute robot frame, the problems are similar. Here’s Ex - if you haven’t seen it, do.
By the way, Garland has figured out that ultimately the real Turing Test is whether an AI can make you fall in love with it.
But the point of this post was to highlight this great image, on the top of this page, via Tegmark, from robotocist Hans Moravec on how our intelligence landscape is slowing being ‘flooded’ (i.e. learned) by artificial intelligence. I hadn’t seen it before, but I really liked it. Intelligence is a bunch of different skills, working together - some easier to learn than others. Gradually, capabilities like playing chess, understanding images, doing calculations, reading emotions are being learned via ever-more complex machine learning algorithms. And so the water rises. Drop by drop…