Hao Fei Paper Proposes Lifelong AI Companions to Augment Human Intelligence

A paper by Hao Fei proposes AI companions built around a person’s long-term context and daily needs, adding human augmentation as a goal alongside machine intelligence.

In “From Autonomous Agents to Cognitive Symbionts: Toward AI-Augmented Human Intelligence,” submitted to arXiv Nov. 3, Fei presents AI-augmented human intelligence (AHI) as a measure of how much intelligence a system can enable in one person. The concept sits alongside artificial general intelligence (AGI), which concerns the capabilities of the machine itself. The distinction makes the person’s gains part of the design question: whether an AI helps someone think and act, as well as what the software can do alone.

The paper sets out six necessary properties for such a system: lifelong assistance, integration into daily routines, real-time responses, privacy for the individual, minimal disruption and proactive help. Lifelong support would let an assistant retain relevant context over long periods, while embedded, low-friction interactions would fit into existing activities. The system is also meant to respond as needed and anticipate when support may be useful without waiting for a direct prompt.

To enable that role, the proposed companion would use information from external devices and digital traces to model a person’s situation and mental state. Wearable devices and a person’s other digital records are described as possible inputs, giving the assistant context beyond a single conversation. The proposed cycle moves from sensing and preserving context to interpreting the situation, operating alongside the user, taking actions for them and supporting joint reasoning. That progression extends the idea beyond a tool that answers isolated questions; it envisions a system that learns from continued interaction and develops alongside the user.

The design also includes a personal runtime and trust mechanisms, which the paper treats as parts of the system architecture. It targets support for memory and situational awareness, alongside decisions and everyday tasks. Those functions illustrate the paper’s central measure: whether the system improves a person’s ability to act and think.

Fei’s proposal also raises questions about acceptable use. The paper flags covert assistance during exams and social pressure to adopt the technology as risks, and calls for governance to be built into the framework. It also places people using AI augmentation within a wider vision of Physical AI, alongside embodied robots.