Originally an answer on Quora to What are some examples of Machine Learning being used to generate interfaces?
Interfaces serve a mixture of utility functions that’s really hard to encode as an optimization, which is another way of saying designers mostly do intelligent guesswork around lots of experience that’s hard to quantify, but one day, perhaps we will.
There’s two general ways I can think it is plausible though
- Ranking suggested interfaces based on constraints you have already. Products like Beautiful.AI - AI Powered Presentations ask you to dump in content and based on that suggest possible layouts. You have to reformulate as a ranking problem instead of a straight up suggestion because there’s too many false positives.
- With a lot of scale, in theory, you can reinforcement learn something like long-term user engagement by giving an algorithm the ability to turn on and off features. My Facebook app has consistently different tabs at the bottom from that of other people I know, I wonder if they’re RL-ing that. It’s arguable whether the choice of buttons within a set UI paradigm is really “machine learning an interface” though.