Would You Talk Differently If Someone's Watch Was Taking Notes? (Zach Chmael)
Connections
Zach Chmael’s “Don’t Feed The Algorithm” newsletter examines Apple’s upcoming Apple Watch features — Live Rewind (15-second conversation replay on double-press) and Siri Recap (AI-generated conversation summaries) — from the perspective of the OTHER person in the conversation, not the Watch wearer. The piece applies Helen Nissenbaum’s contextual integrity framework (Privacy in Context) to argue that the relevant question is not “is this technically private?” but “does this match the norms of the situation in which the information was shared?” Connects to Privacy and Data Privacy threads; also relevant to Kyle’s interest in AI memory systems and the ethics of personal knowledge management infrastructure.
Summary
Apple’s Live Rewind shows a text snippet of the previous 15 seconds when you double-press the Digital Crown (audible chime even when silent, visible indicator). Siri Recap generates conversation summaries; raw audio is processed on-device and deleted, the condensed text goes to Private Cloud Compute with “content suggestive of conversation tone” removed. Both are opt-in, English-first, Apple Watch Series 12/Ultra 4 only, no EU launch initially.
The core discomfort: a person who benefits from AI memory aids has obvious reasons to use them; the person ON THE OTHER SIDE of the conversation has no say in what gets retained or how. Chmael notes he’s less comfortable being the source material than he is building AI memory systems — “my enthusiasm for reliable recall becomes more nuanced when I might be the source material.”
Contextual integrity (Nissenbaum): information should be handled according to the norms of the situation in which it’s shared. A casual opinion about a restaurant, an exploratory thought about leaving a job, or a midway-through reconsideration should not be summarized into a “clean conclusion” that forgets the provisionality. The risk is ordinary: a clear summary makes it easy to forget how contingent the original conversation was, and the summary can be forwarded to someone who wasn’t there.
Power dynamics matter: if a manager wants to summarize a conversation with an employee, the employee may reasonably wonder whether objecting will itself become the memorable part of the meeting. A visible control on the watch doesn’t solve that problem.
Proposed habits before turning it on:
- Say what you want to remember — ask specifically, not “are you okay with AI?”
- Make declining uneventful — switch it off without demanding a reason
- Verify a commitment with the person before acting on the summary
- Ask again before sharing — permission to help you remember ≠ permission to send the summary elsewhere
- Keep less when the purpose is finished — Siri Recap’s 7-day auto-delete is a useful default; ask why you’re saving something
The final concern: “I don’t want the summary to win by looking organized.” A nicely formatted paragraph should not be able to defeat someone saying “that isn’t what I meant” — and AI memory shouldn’t make people reluctant to clarify, reconsider, or think out loud.