Paper Puts a Number on AI-Human Attention Gap: Context Windows Up 3,906x, Focus Down
alex_verem · x · 2026-09-29
The arXiv paper "The Cognitive Divergence: AI Context Windows, Human Attention Decline, and the Delegation Feedback Loop" (Netanel Eliav, Machine Human Intelligence Lab, submitted 17 Mar 2026) puts two trends on the same scale:
- AI context windows: grew from 512 tokens in 2017 to 2,000,000 tokens by 2026 — roughly 3,906x, with a fitted growth rate of λ≈0.59/yr and a doubling time of about 14 months.
- Human Effective Context Span (ECS): derived from reading-rate meta-analysis (Brysbaert, 2019) plus a Comprehension Scaling Factor, estimated to fall from 16,000 tokens (2004 baseline) to 1,800 tokens (2026, extrapolated from longitudinal behavioural data ending 2020; uncertainty discussed in Section 9).
- The gap: the AI-to-human ratio went from near parity at ChatGPT's launch (Nov 2022) to 556–1,111x raw, and 56–111x after adjusting for retrieval degradation on long inputs (Liu et al., 2024; Chroma, 2025).
The paper proposes the "Delegation Feedback Loop": as AI capability grows, the cognitive threshold at which humans delegate falls, extending to tasks of negligible demand; less practice weakens skills, so people delegate more. The author says that within 30 months what people hand to AI shifted from essays and code to two-sentence email replies.
Cited evidence includes an MIT study of 54 people where the LLM-assisted essay group showed the lowest brain engagement of three groups, and stayed less engaged when later writing without it; a separate 666-person study linked heavy AI use to weaker critical thinking. The author concedes no study has followed the same people over years of AI use, so the full loop is untested. The loop has a boundary: handing AI a 200,000-token report costs nothing, but the daily two-sentence emails do — because you stop practising the writing.
More from AGI Musings
- Dr_Singularity: every problem you have is downstream of not enough compute and energy — Dr_Singularity · 2026-09-29
- Sonnet described 2D experience of time, Davidad's biggest update on LLM self-awareness — davidad · 2026-09-29
- Bayesian framework from DeepMind-linked researchers scores LLM consciousness between 1% and 80% — mhutter42 · 2026-09-29
- Timnit Gebru Says AI Poses No 'Existential Threat': Doom Talk Is About Founders Chasing Profits — nordicinst · 2026-09-29
- AI harms worse than pollution: researcher likens AI to a pathogen — gleech · 2026-09-29
- gleech: pre-2024 uses of "AGI" are fatally ambiguous across multiple senses — gleech · 2026-09-29