Claude Shannon, the mathematician who built the foundation of information theory, made a decision in 1948 that still shapes how every AI model works: he defined 'information' purely in terms of surprise — how unexpected a signal is — and deliberately stripped out meaning. That was the right move for engineering phone lines. It becomes a strange trap when we forget it's a choice. Shannon's framework, combined with what the pragmatist philosopher William James called 'the cash value of ideas' — meaning that concepts only matter insofar as they produce real differences in experience — reveals something important about AI-generated communication: it can be maximally statistically coherent and simultaneously empty of consequence. A message optimized for low perplexity (the technical measure of how well a language model predicts the next word) is precisely the message that contains the least surprise, which is to say, the least information in Shannon's own sense. The practical implication isn't about AI quality — it's about your reading posture. When consuming or producing AI-assisted communication, the thing worth interrogating is not fluency but unexpectedness: did this actually introduce something that changes how you'll act?

In the last conversation you had — human or AI-assisted — what was said that you could not have predicted, and what did it actually alter?

Drawing from Philosophy of Information / Pragmatism — Claude Shannon (A Mathematical Theory of Communication, 1948) combined with William James (Pragmatism, 1907)