把照片或声音交给AI之前,要想到什么? / What Should You Consider Before Giving AI a Photo or Voice Recording?

Judgment in the Age of AI · Article 14

A photograph or recording is more than “content”. A photo may reveal faces, children, address clues, number plates, workplaces and location; a voice clip may contain identity, emotion, background conversation and material capable of imitating a speaking style. Text can be rewritten, but biometric characteristics are difficult to replace.

Inventory what the media contains

  • Subjects: Who appears or can be heard, and have they agreed?
  • Environment: Does the background reveal an address, school, document, screen or identity card?
  • Metadata: Does the file retain time, device and location information?
  • Later use: Will the service retain, train on, share or allow deletion of the file?

Do not make another person’s decision with your risk tolerance

Your willingness to upload a group photo does not establish everyone else’s consent. Apply a higher standard to children, clients, patients and unaware bystanders. Crop, blur, alter or substitute material whenever the full original is unnecessary.

Deleting a file may not remove every copy

Check the service’s privacy notice, history settings and deletion process before upload. If the task concerns composition, transcription format or voice style, begin with a low-resolution, de-identified excerpt or fictional sample.

The convenience of uploading media is temporary; exposure of a face or voice may be lasting.

References


Discover more from Geoffrey Chen

Subscribe to get the latest posts sent to your email.