Predictive text is probably the most common AI interaction we have – and it gets more and more sophisticated as technology progresses. Starting with ‘autocorrecting’ typing on numeric keypads using word disambiguation, the latest messaging keyboard function also predicts likely words, emojis or actions you may want to make based on the context of your and your messaging partner(s) recent messages, e.g. it will set up a shortcut to add a diary entry if a time/date is suggested for a meet up, or give a cake emoji suggestion if you are wishing someone a happy birthday. In fact, the system now does not even require that you first enter any words yourself. Relying on a body of knowledge that is based on the context of all user interactions, as well as your own, it can predict to a certain extent likely sentences and phrases that might come up.
Since the increasing sophistication of AI is a trend often touted as leading us to fully sentient/self-aware AI in the future (who might then be considered persons/minds in their own right), I was interested to explore generative art, making use of the models to create works that might reveal the state of these proto-minds.
First I experimented by continually pressing the left hand suggested text until I had a complete text (below) and then sent this to my boyfriend. I especially like how the model has fallen into a loop on the phrase ‘and I hope you have a good day’. That phrase is a cursory sort of nicety that serves more as a signal that a conversation is coming to a close than indication of a genuine feeling – i.e. i hope you have a good day since I don’t anticipate us interacting again for the remainder of it. The fact that that sign-off gets repeated undermines that function, so stripping it of even this meaning. The model is also failing to give us a full sentence, the breaking of syntactic convention makes this seem more like a meditative poem or song lyric, and the repetition of that phrase reinforces that. It’s interesting too that the arrangement of the repeated words appears to create diagonals across the block of text (wanna, day, you, have, good, day), almost like the creation of a pattern.
It’s interesting too thinking about what the ‘I’ in this poem might signify – is it me? Is the model adopting my voice? Or does the ‘I’ refer instead to the predictive model itself?
Later in the day, my boyfriend and I experimented with sending such predictive messages back and forth between our phones, to see how this context might affect the conversation between the predictive models. The text in the green bubbles has originated from my phone, and the white from his.
It’s intriguing that here again, now from my boyfriend’s predictive model, we see a repeated phrase ‘and then I will have to go to pick up the kids tomorrow’. We do not have any kids, so this context must come from this phrase being observed from other messaging users – it’s interesting though that it first appeared in my text ‘let us get the kids’ and that my boyfriend’s one then repeated this. Generally the conversation between the models seemed to mostly orientate around arranging a meeting time and planning for tomorrow. It’s interesting too that they both made some slight use of emoji, though could not be signifying emotion in themselves.
Intriguingly, many of these messages are related to future events/future planning.