Original title: Slow-Fast Brain-Computer Interfaces: Preventing Neuroadaptive Overfitting in AI-Mediated Neural Interfaces
Aarthy Nagarajan · September 2026
Brain-computer interfaces let people control devices using brain signals. Today, artificial intelligence helps these systems predict words, smooth out motions, and adjust therapy. But when an AI helps too much, problems arise. The paper identifies a risk called neuroadaptive overfittingWhen an AI system learns to make tasks too easy in the short term, ignoring what the user really wants.. This happens when an AI chases quick wins, like making tasks feel effortless, but secretly ruins the user's real goals. The user can lose their sense of control, personal authorship, or long-term recovery gains. To solve this, the author designs Slow-Fast BCI. This system shifts speeds based on how certain the AI is, user fatigue, and real-world stakes. It provides fast help during easy, clear moments. It uses guarded help during uncertainty, and slow help when safety or learning is at risk. Ultimately, the system protects human agencyThe user's sense of control over their own actions. instead of letting AI take over completely.
Imagine you are learning to play the piano with a teacher who automatically corrects every wrong note before you finish playing. At first it feels easy and smooth, but you never learn to correct yourself. Here, the teacher is the AI assistance, and the piano learning is the user's long-term goal of motor skill acquisition. The paper warns that AI that always jumps in to make things easy can prevent the user from developing real skill or control.
For people who rely on brain-computer interfaces for communication, movement, or rehabilitation, current AI systems may silently undermine their long-term goals by prioritizing ease and speed. This matters because it could reduce the user's sense of agency, slow motor learning, or compromise safety. The proposed Slow-Fast framework offers a way to match AI assistance to the situation, preserving what matters most.
This is a conceptual framework, not an empirical study. It does not provide experimental validation or specific algorithms for implementing the Slow-Fast BCI. The abstract does not address how to precisely measure intent fidelity or agency in practice, nor does it quantify the trade-offs between fast and slow assistance.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Slow-Fast Brain-Computer Interfaces: Preventing Neuroadaptive Overfitting in AI-Mediated Neural Interfaces — Aarthy Nagarajan
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.