Predictive Asymmetry

The central claim of Mutual Prediction in Human-AI Coevolution: imbalance in mutual predictive ability correlates with imbalance of power, across species (ticks out-predict hosts; humans out-predict dogs and wheat), artifacts (the smartphone models its user better than she models it) and AI. Figure 1 (PDF only) plots human against AI predictive ability with a diagonal "balance line"; below it AI is a tool, above it humans become "the observed, modeled, and guided." The ELIZA case adds the corrective that position depends on technical knowledge and context, not cognitive capacity — designer and public occupy different coordinates for the same system — which is why asymmetry between user and platform operator needs separating from asymmetry between user and model.