Should AI customer support always disclose that it is AI?
For: disclosure protects trust and gives people a fair chance to ask for a human.
Against: a capable system may resolve simple problems faster, and a label can create distrust before performance is judged.
Switch test: would the rule change if the AI solved the issue faster but could not handle an appeal?
Should personal data be stored on-device by default?
For: local processing reduces the number of companies that can copy, sell, or lose sensitive information.
Against: cloud processing can make services cheaper, more accessible, and more useful across devices.
Switch test: would a verifiable deletion guarantee justify sending the data to a server?
Should workplace monitoring record keystrokes?
For: activity records can help investigate misuse and show whether a remote process is functioning.
Against: keystrokes measure motion, not judgment, and constant surveillance can make careful work look unproductive.
Switch test: would the rule change if workers could inspect, correct, and challenge the record?
Should platforms optimize for wellbeing instead of engagement?
For: a product should not quietly reward outrage, compulsion, or the most polarizing version of a story.
Against: wellbeing is hard to measure and gives a platform too much power to decide what users should see.
Switch test: what transparent metric would count as evidence of healthier use?
Should creators label synthetic or heavily edited media?
For: viewers deserve to know when a face, voice, or event has been materially constructed.
Against: synthetic media includes harmless art as well as deception, and a broad label can become a stigma.
Switch test: should disclosure depend on whether the edit changes a factual claim?
Should software companies make devices easier to repair?
For: repair access can reduce waste, extend ownership, and keep a broken component from becoming a discarded product.
Against: poorly performed repairs can create safety and reliability problems, and design changes can raise upfront costs.
Switch test: would a certified repair network satisfy the duty without opening every component?
Should recommendation systems explain why they show a post?
For: a reason gives people a way to understand, contest, or change the feed that shapes their attention.
Against: simple explanations can reveal little, invite gaming, or falsely imply that a complex ranking has one cause.
Switch test: would an independent audit be more useful than a sentence shown beside each post?
Should AI hiring scores always be appealable by a human?
For: an appeal protects candidates when a proxy, missing context, or data error turns a score into a hidden rejection.
Against: mandatory review can slow hiring and make a structured process inconsistent again.
Switch test: what evidence should trigger a human review before a score becomes a decision?