AI-powered facial recognition proves machines won’t always replace humans

Credit: Ryan Jorgensen - Jorgo /

AI is crawling into almost every part of our lives. We are trying to harness it to augment our lives and capabilities while running the risk that AI will supplant our capabilities and ruin our lives.

A new example of the power of AI has been announced. AI-powered neural networks at Shanghai Jiao Tong University have identified criminals with an accuracy of 89.5%.

By looking at their faces.

Actually, 1,856 of them.

This is, of course, extremely clever. And scary. Naturally, it also raises a few questions.

Leaving aside the question of whether the technology is transferable to other regions and therefore facial formats, you have to wonder about the 10.5% of people who were identified inaccurately. In an age where proof is becoming more and more difficult to nail down, 90% accuracy is nowhere near acceptable.

Remember the hype around OCR? [Optical character recognition, right? – Ed.]

When it first hit the radar, accuracy was around 90%. We all thought this is great and looked forward to a future without keyboards. Re-typing stuff would become as obsolete as hand writing. But 90% accuracy with anything means that there is still a significant margin of error. It means that one in 10 words are wrong. Or one in 10 criminals are not criminals.

And those first announcements about OCR were 20 years ago. Even now, accuracy is not 100%.

While the experiment in Shanghai might be extraordinary in many respects, and will complement a lot of work that is going on with profiling, it will be a while before the police line up does not require people to be present.

Without doubt we are entering the age where machine learning and AI are a critical part of progress.

Experiments like the AI-powered facial recognition one in Shanghai demonstrate not so much that we will be able to leave things to machines and saunter off to “be creative but that machines might just, after all, really be about augmenting our capabilities, not supplanting them.

Which is a good thing.

Be the first to comment

What do you think?

This site uses Akismet to reduce spam. Learn how your comment data is processed.