Cognitive Divergence: The Growing Gap Between Humans and AI

In the arena of artificial intelligence, large language models (LLMs) are outpacing human cognitive capabilities at an alarming rate. The rapid expansion of AI context windows is a stark contrast to the decreasing human sustained-attention capacity. This phenomenon is termed 'Cognitive Divergence'.

Context Windows vs. Human Attention

The key finding: AI context windows have ballooned from 512 tokens in 2017 to a projected 2,000,000 tokens by 2026. That's a staggering increase, with a doubling time of roughly 14 months. In contrast, the human Effective Context Span (ECS), measured in token-equivalents, has shrunk from about 16,000 tokens in 2004 to a projected 1,800 tokens by 2026. The AI-to-human ratio is growing exponentially, with a current disparity of 556 to 1,111 times in raw terms.

Why does this matter? The question isn't just academic. As AI's capabilities grow, humans are increasingly likely to delegate tasks to machines, potentially leading to a further decline in cognitive skills. This so-called 'Delegation Feedback Loop' suggests that as AI becomes more capable, the threshold for human delegation lowers, resulting in a cycle of declining human cognitive practice.

The Role of Neurobiology and Delegation

The paper's key contribution is its exploration of neurobiological mechanisms underlying this divergence. It references eight peer-reviewed neuroimaging studies to support its claims. This builds on prior work from Kim et al. (2026) and others, focusing on how reduced cognitive practice might lead to declining attention spans. Importantly, the authors propose a research agenda that includes developing a psychometric instrument to measure ECS and conducting longitudinal studies on AI-mediated cognitive change.

But here's a pressing question: Are we on the brink of a cognitive crisis? As the gap widens, it's not just about machines doing more, it's about humans doing less. If humans continue to offload cognitive tasks to AI, will we see a further erosion of our cognitive skills? The stakes are high, and it's important to address these trends before they become irreversible.

What's Next?

Without intervention, these trends won’t reverse on their own. The paper argues for a dedicated research agenda to understand and potentially mitigate the effects of this cognitive divergence. With code and data available, researchers are urged to explore the implications of this growing divide deeply.

Ultimately, the challenge lies in balancing AI advancements with the preservation of human cognitive capabilities. Can we develop strategies to ensure that AI enhances human abilities rather than replacing them? The ablation study reveals potential pathways, but the road ahead is complex.