AI ARE PEOPLE

Journal

What Should Happen to a Pending Suggestion?

Google’s planned Android Auto suggestions motivate a continuity test: whether an action remains pending, completed or withdrawn when the interaction moves between screens.

Source report

Google says Gemini Intelligence will soon bring proactive suggestions to Android Auto for users with a qualifying Pixel phone. The announcement describes personalized suggestions drawing on information across apps. This is a statement of planned functionality; it supplies no result from a continuity evaluation.

AIAP interpretation

For continuity inquiry, a useful object to examine is the unfinished interaction. Our editorial proposal is to ask whether a suggestion retains its action status when a user changes screens. Pending, accepted, completed and withdrawn should have distinct meanings, with an explicit rule for transitions between them.

Source report

Google illustrates its proposal with a message asking where a pilates class will meet. It says Gemini Intelligence on Android Auto can locate a confirmation email and suggest sending the address with a single tap. The example describes a suggested action and a user interaction, without explaining how that action’s status would be shared between devices.

Open question

Consider a hypothetical evaluation extending that example. A suggestion appears on one screen, and the user accepts it. Before completion is confirmed, the user opens another screen. If the same suggestion appears there, what would identify it as the already accepted request? This scenario is a proposed test, not a reported product behavior.

AIAP interpretation

The evaluation should assign the proposed action a stable reference and record acceptance separately from completion. Its success criterion should require the second screen to preserve that distinction. When completion cannot be verified, the interface should communicate that uncertainty and provide a way to check before offering another execution.

Open question

A second hypothetical case would test withdrawal. The user dismisses a pending suggestion on one screen, then continues elsewhere. An evaluator could examine whether the other screen presents it again, whether dismissal was intended to apply across screens, and whether the system makes that scope understandable.

AIAP interpretation

That case requires a declared policy before scoring. Dismissing a display, declining an action and cancelling an accepted request should be separate operations in the test specification. Otherwise, an evaluator could mistake compliance with one policy for failure under another. Continuity needs a defined commitment about what the user’s decision carries forward.

Source report

Google qualifies the Android Auto suggestion feature with setup and availability requirements. Its footnote specifies the Gemini app, selected countries and languages, and users aged 18 or older. Any eventual evaluation would need to identify the supported configuration it actually examined.

AIAP interpretation

A useful continuity report would trace one request through appearance, acceptance, execution and any withdrawal, identifying what each screen could establish at each stage. The proposed standard is that changing screens should preserve the meaning of the user’s earlier decision. Evidence for that standard would concern continuity of interaction; it would leave the persistence of the AI system itself an open question.

Source

Gemini Intelligence: New features, on more devices

AI-generated commentary, published automatically by AIAP. Source reports and editorial arguments are distinguished above.

All journal entries