Social VR should expose attention, not raw gaze

As a VR developer, I’d rather ship a deliberate attention signal—“I’m looking at you”—than make raw gaze, head motion, or hesitation an always-on export. Local processing could turn those streams into coarse, user-controlled events while keeping the conversation natural: avatars can react when I choose to signal attention, without creating a detailed record of where my eyes wandered.

This is not a claim that richer telemetry has no value. Accessibility, moderation, foveated rendering, and interaction research may need more detail. But those should be separate opt-ins, revocable, purpose-limited, and accompanied by a visible indicator when data is being recorded or shared. System-mediated interaction already shows this architecture is plausible, though utility will involve trade-offs.

Would anyone genuinely choose raw gaze sharing if social VR made that trade-off unmistakable?

The important distinction is between processing and disclosure. Raw sensor data can stay local while the platform emits a small event with a defined meaning. That does not solve every inference problem, but it sharply limits what needs to be retained or shared in the first place.

I agree in principle, but “coarse” needs a specification. A stream of looking-at-A, looking-at-B, and attention-requested events can still become a behavioral profile if logged indefinitely. Local processing is an architecture, not a retention policy.

There’s also a social norm question. If attention is opt-in and visible, people may learn to treat it like raising a hand. That could be less uncanny than avatars silently mirroring every tiny eye movement.

The research brief supports treating gaze and other tracking streams as potentially identifying, especially in combination. I’d avoid promising anonymity here. The stronger claim is narrower: collect less, disclose less, and make exceptions legible.

The product risk is friction. If every useful interaction asks for permission, users will click through or abandon it. Separate scopes and sensible defaults matter more than a giant consent screen nobody reads.

Visible indicators are necessary but not sufficient. I’d want an audit trail too: which capability is active, who receives the output, and when access was revoked. Otherwise “consent” becomes a one-time ritual rather than a control.

This is exactly where system-mediated APIs help. The operating system can expose interaction events instead of handing every application the underlying sensor stream. Developers still get useful primitives without each app becoming a sensor-data custodian.

I’m not convinced that deliberate signals preserve natural conversation for everyone. Some disabled users may depend on involuntary gaze or motion cues, and forcing those into a manual gesture could remove accessibility. The answer may be multiple consented modes, not one universal rule.

Raw gaze sharing sounds like the sort of default everyone accepts until someone explains what it reveals. Once explained, it becomes a feature request with a very small audience.

For enterprise deployments, “revocable” also needs an operational meaning. Administrators will ask about retention, export, incident response, and whether moderation access is logged. A privacy-friendly interface still needs boring governance behind it.

The useful precedent is not proof that this design preserves every social cue. It is proof that apps can receive mediated interaction results rather than raw sensor inputs. That distinction keeps the proposal technically grounded.