Artificial intelligence in SF tends to be “general AI” – machines that are basically humanlike in their abilities, and in possessing consciousness, but that are either slightly less or slightly greater than humans (sometimes both at once). AI in SF, in fact, tends to be about AI as an allegory of humanity more than it is about AI-as-it-actually-might-be; robots and AIs stand in for underclasses or aggressive foreign Others.
This is a fascinating lecture by a professor at Oxford, who has a computer science background but is in the philosophy department, about the various philosophical challenges and implications of actual AI (all of which, so far, is “narrow” AI, confined to a specific domain, rather than AGI – artificial general intelligence).
The biggest issue is that we don’t really know how to make AGI, and if we do succeed in making it we won’t understand exactly how it works or be able to predict what it will do in new situations it wasn’t designed for – even if we managed to build it correctly to our original design, which, as anybody who actually practices in the software world will tell you, tends not to happen.
There’s also the problem that we don’t really understand how our own ethical system works, and even our best approximations aren’t susceptible to being reduced to code. Perhaps we need to make sure that robots recognise “human” as a very basic category with special value and importance (which immediately put me in mind of an Asimov story, “That Thou Art Mindful of Him”, in which the robots decide that everything they’ve learned about humanity convinces them that they’re a part of it and should have the same rights).
One of the questions in the Q&A session is about how SF might or might not prepare us to encounter the kind of new situations that AI will bring. Dr Sandberg’s answer is that specific stories generally aren’t that useful (because they’re not about how AI really works), but the general mindset of SF – thinking about how to deal with new things and how to interact with the Other – is helpful.
(1.5 hours, including half an hour of excellent Q&A.)
imd
imd
imd