Research scientist Janelle Shane has announced the forthcoming publication of her new book, You look like a thing and I love you.
Janelle is perhaps best known for her experiments in artificial intelligence gone (slightly, deliberately wrong) in her blog, AI Weirdness. In a way, Janelle's work is a postmodern re-enactment of the ventriloquist and their dummy – with the "dummy" being the AI software spitting out strange, pseudo-realistic responses based on Janelle's input. As such, her obvious heritage as an expert in this field vastly underestimates her own abilities as a storyteller, commentator, and general all-round promoter of the more human aspects of AI: after all, if humans make mistakes, then computers could do so too.
Of course, there's the fun stuff. Janelle's invocation of the infamous GPT-2 neural network to generate cat names gave some rather grandiose ones (Taffeta, Monocle), unwise ones (Scumbag, Dr Fart) and some which are absolutely not advised (Kill All Humans, You're Telling A Lie). Although the AI outputs are clearly weird and hilariously so, the subjects and inputs that Janelle funnels in are esoteric enough to be funny themselves. Feeding debate topics from the UK parliament generated new ones such as "This house would allow parents to develop nuclear weapons".
Janelle's work is important, and needs to be heard beyond those of us having a bit of a laugh about it. Showing the failings of AI indicates a level of care and attention into the topic that goes way beyond what the big vendors and integrators are showing you. We need to see that AI is fallible. We need to see that AI makes mistakes on its way to progress. Because, after all, that's what we do. If you're not superhuman to start with, don't tell the world that you are superhuman from day one.
You look like a thing and I love you, is published by Voracious and available to buy in November.
(This article contains paid placement links.)