Responsible Vision
Just because you can.
A camera that counts apples harms nobody. A camera that tracks people is different. Studies have found that some commercial face systems made far more mistakes on darker skinned women than on lighter skinned men. Wrongful arrests have followed from false matches. Computer vision can see people, and that brings a real responsibility.
- Apples are harmless
- Tracking people isn't
- Wrongful arrests have happened
Questions to ask
Before building anything that sees people, stop and ask. Did those people agree? Who could be harmed if it's wrong? How does it perform across groups? Could this be misused, once it exists? Does it need to identify anyone at all? Counting people in a shop needs no identities. And in many places, the law limits this, such as the EU's AI Act.
Privacy by design
You can design for privacy from the start. Process on the device, so that images never leave it. Blur faces and number plates at once. Store counts, rather than footage. Delete data quickly. And test performance across skin tones, ages, and genders, before you ship, not after a complaint. Good engineers can also say no.
Summary
Build what helps, and think about who's on the other side of the lens. You've now gone far beyond most vision courses. Next, let's put these skills to work, on a series of real projects.
Key takeaways- Test across groups
- Design for privacy
- Good engineers can say no