Markov to autocomplete
The statistical insight is a century old: the probability of the next letter in a text depends on the letter before it, demonstrated in 1913 on a Russian novel. A century later, autocomplete โ and the sharp point is not that the prediction is good. The suggestion does not need to predict what you wanted to type, because it changes your behaviour to make the prediction fit. You accept it, and now you have typed it. Prediction that reshapes what it predicts: the mechanism underneath the invisible influence, and the collapse of information's almost predictable band from the predictable side.
Keywords
chain dependenceself-fulfilling predictionbehavioural feedbackReferences
Prado Casanova, Ch. 4 (~pp. 121โ123).