Is face recognition good or bad? Look at how it is used
Face recognition can make a voluntary identity check more convenient, but the same technology can enable unwanted tracking. The answer depends on who collects the face data, why they need it and what happens when the system is wrong.
3 misconceptions (table)
The three claims below often turn a question about a particular use into a blanket judgment about the technology.
what it actually is
Face recognition compares facial information to a reference. Whether that comparison is justified is a separate decision from whether it can be performed.
Common assumption
More accurate view
Purpose
Common assumption
Every face scan identifies a named person.
More accurate view
Some systems verify a claimed identity; others search for a match among many people.
Consent
Common assumption
A visible face means permission to identify it.
More accurate view
Being photographed does not, by itself, establish informed consent to recognition.
Accuracy
Common assumption
A match is proof of identity.
More accurate view
A match is a system output that may require independent review, especially when the stakes are high.
Data
Common assumption
Only the original photograph matters.
More accurate view
Stored templates, match records and access logs may also need protection.
Scope
Common assumption
A phone unlock and a public-camera search pose the same risks.
More accurate view
A user-initiated check and identification of passersby differ greatly in scale and control.
Benefit
Common assumption
Convenience alone makes deployment worthwhile.
More accurate view
The benefit must be weighed against errors, exclusion, surveillance and less intrusive options.
boundary conditions
Face recognition is easier to defend when its purpose is narrow, participation is meaningful and a person can challenge an incorrect result.
-
Consent needs to be real
A notice is not enough if someone cannot reasonably decline. A voluntary check is different from scanning everyone who enters a public space.
WorkaroundOffer a practical non-biometric way to complete the same task.
-
Errors need a remedy
Image quality, changing appearance and system performance can affect results. An incorrect rejection can block access; an incorrect match can implicate the wrong person.
WorkaroundProvide human review and a clear route to correct a disputed decision.
-
Data needs a defined lifespan
Keeping facial information indefinitely creates risks beyond the original check. People should know what is retained, who can access it and when it is deleted.
WorkaroundCollect only what the stated purpose requires and set a deletion policy.
-
Creative tools are not identity checks
Faceai is a video-focused creative brand. This page cannot determine whether a particular person has consented or verify anyone's identity.
WorkaroundConfirm permission separately before using a person's likeness in a project.
when NOT to use it
Keep the purpose and the person's choice in view
Do not use face recognition to identify people without a clear, justified purpose, especially where they cannot opt out or an error could seriously affect them. For a creative video involving someone's face, get their permission rather than treating a usable image as consent. Faceai is a place to explore creative work, not a substitute for an identity-verification process.
Explore Faceai- Avoid covert identification or broad monitoring of people who have not agreed to it.
- Avoid making consequential decisions from an automated match alone.
- Avoid using another person's likeness in a creative project without permission.
its own FAQ
Neither label fits every use. A voluntary, limited identity check may be useful, while covert identification or a decision with no way to appeal can cause serious harm. Judge the specific purpose, the person's choices, the risk of error and how the data is handled.
It starts with a specific purpose and a clear explanation to the people affected. Meaningful choice, limited retention, secure handling and a way to dispute mistakes matter more than a general promise that the system is accurate.
A match is the result of comparing images or facial representations under particular conditions. Poor images, lookalikes and system errors can affect that result, so consequential decisions should not rest on an automated match alone.
Not necessarily. Agreeing to appear in a photo does not automatically mean agreeing to have that image searched against an identity database or retained for future checks. The intended use should be explained separately.