(This is the featured meat of this week’s newsletter, which just went out. If you want to subscribe you can find it here.)
Thinking about artificial intelligences a lot lately. Our thinking about machine thinking feels like it’s matured significantly over the past six months or year, entered a new phase. Most of the conversation seemed stuck for a while on fears like Bostrom’s paperclip maximizer – the idea that an AI programmed even to a trivial task could be dangerous because it thinks machinistically. That, programmed to create paperclips, an AI would consider humans detrimental to its programming because they might unplug it, and also humans are made up of a lot of atoms that could be better served assembled into more paperclips. The logical issues with overblown fears like the paperclip maximizer are astounding – first and foremost, it pretends we can think non-biologically about how an AI will approach biology, or really any task. The argument invalidates itself by wrapping around the idea that we can in no way conceive of how AI will “think.” But, apparently, we can know enough to be afraid of it. In fact, we can guess enough about its possible thought processes to consider it an existential threat, according to Bostrom!
The idea that non-biological intelligences will “think” in ways totally alien to us is not new, but I’ve seen it explored a lot more lately, and with a lot more depth. It’s entering more conversations about AI in general and our interactions with it in particular. Take AlphaGo, for instance – Google’s deep learning program devoted to playing the board game Go. AlphaGo’s faced four matches with the second best Go player in the world, winning the first three (the fifth has yet to be played). The ability to watch and analyze in detail human-AI interaction can produce some astounding insights, and in this case seems to point to an entirely new way to approach Go gameplay.
Delving into the idea of nonbiological artificial intelligence in a different way is Injection, a comic written by Warren Ellis with stunningly beautiful art by Declan Shalvey and Jordie Bellaire. Injection’s relatively new – issue 8 comes out this week – but quickly becoming a favorite comic of mine thanks both to the art and the wonderful ways in which Ellis intertwines things like technology, philosophy and folklore. It involves a crossdiscipline team of varied experts including Brigid Roth, a hacker and programming phenomenon who, well, had a bit of fun with the Turing Test…
Injection goes much, much deeper into the issue of machine intelligence, in fascinating ways. And I’m excited to see where Ellis takes it, especially given the richer environment. We might be ready for this conversation. We might. Injection approaches it in a great way, though – through a lens not necessarily but not unlike horror. And as Eugene Thacker (who I talked about in the last issue of this newsletter) states, horror is a way to think about the unthinkable, a way to process past what might be the limit of human thought. Using horror to approach nonbiological intelligence – a form and function of “thought” that we cannot comprehend – is nothing short of perfect, a sort of speculative machine intelligence metacognition.