AI that completes a task can be evaluated on its own; AI that quietly improves someone’s performance changes the stakes for everyone around them. This paper’s central shift is to measure progress by the intelligence a system enables in a person, not only by what the machine can do independently.
Key Findings
- It frames AI-augmented human intelligence as a distinct goal alongside AGI. This changes the unit of evaluation from the model to the person whose capabilities are being augmented.
- It defines the “cognitive gap” as the difference between what someone could bring to a situation and what they can hold in working memory. The system’s purpose is to help close that gap using context accumulated over time.
- Its proposed symbiosis loop connects sensing, memory, personal modeling, timely participation, assistance, and learning. Trust—especially authorization, privacy, correction, and the ability to exit—surrounds the whole loop rather than being treated as an add-on.
- The paper makes fairness part of the technical framing: discreet help may give one person an advantage in exams, interviews, or negotiations, while also creating pressure for others to adopt similar systems.
Who Should Read It
A good fit for researchers thinking about personal AI, agent design, human–AI interaction, or governance: the paper offers a vocabulary and research agenda for systems designed to live alongside a person. Look elsewhere for a new algorithm, a working implementation, or empirical evidence that these systems improve outcomes—the contribution is a conceptual framework, not a demonstrated product.
