Brain signals seem to anticipate upcoming words before they are heard. Is that genuine prediction — or are encoding models picking up statistical dependencies between neighbouring words?
If the brain predicts like a language model, next-word training is a real bridge between AI and neuroscience. If not, one of the field's headline results needs rethinking.
Brain signals before a word arrives looked like evidence of prediction (Goldstein 2022). A 2026 re-analysis showed systems that cannot predict produce the same signal, so the evidence is now ambiguous. Both sides agree causal experiments are needed.
A neutral summary kept current by super users, not a verdict.
The brain actively predicts upcoming words; pre-onset neural signals reflect those predictions.
Goldstein et al. 2022
Pre-onset signals arise from dependencies between adjacent words, not prediction; causal experiments are needed.
Schönmann et al. 2026; Antonello & Huth 2024
No decisions yet. When this room changes a flow or an evidence label, it is recorded here permanently.
Dhi: Room opened. Two positions are on the table (see hypotheses). Named labs disagreed in print in 2024–2026 and no consensus review resolves it yet. Useful contributions: causal experiment designs, re-analyses with contiguous splits, and stimulus sets where adjacent-word statistics and predictability come apart.