You’re admitting the technology is in fact flawed if you think it needed to be implemented with supervision.

You're absolutely right. The technology isn't perfect if it needs to be implemented with supervision, but it can be good enough to have a role in everyday society.

Great examples are self checkout lanes, where there's always an employee watching, and speed cameras, which always have an officer reviewing and signing off on tickets.

An uno reverse is, every set of traffic lights needs a traffic controller to stop drivers running red lights.

Traffic lights are meant to direct traffic. Yet you don't expect them to prevent folks from running red lights. Folks don't expect them to, because that's not their role in their implementation - they are meant to be used alongside folks who will enforce traffic laws, and, maybe in fact, traffic controllers. This is arguably an example of an implementation done right.

This technology is meant to flag car damage. If there was a correct implementation, I would be able to say "folks don't expect them to be perfect, because that's not their role in their implementation - they are meant to be used alongside employees trained to verify damage exists, who can correct the algorithm if needed", but the implementation in this case is sadly bad.

At the end of the day, you will never have a "perfect" computer vision algorithm. But you can have many "good enough" ones, depending on how they're implemented.