A big biometric security company in the UK, Facewatch, is in hot water after their facial recognition system caused a major snafu - the system wrongly identified a 19-year-old girl as a shoplifter.
A big biometric security company in the UK, Facewatch, is in hot water after their facial recognition system caused a major snafu - the system wrongly identified a 19-year-old girl as a shoplifter.
People who blindly support this type of tech and AI being slapped into everything always learn the hard way when a case like this happens.
Sadly there won’t be any learning, the security company will improve the tech and continue as usual.
This shit is here to stay :/
Agreed on all points, but “improve the tech” probably belongs in quotes. If there’s no real consequences, they may just accept some empty promises and continue as before.
Just listened to a podcast with a couple of guys talking about the AI thing going on. One thing they said was really interesting to me. I’ll paraphrase my understanding of what they said:
So basically AI was meandering around trying to find the right road, and in 2020 it found a road that goes a long way in a straight line, enabling the industry to just floor the accelerator.
The direct relationship this model creates between more neurons/weights/parameters on the one hand, and more intelligence on the other, creates an almost arbitrage-easy way to absorb tons of money into profitable structures.
That makes a lot of sense, and I agree that’s where this is going.
One issue, and not a particularly new one, is that virtually everyone is overselling how far that straight line improvement actually takes us.
As someone with LLM expertise, my gut assessment is that almost everyone is wildly overselling the usefulness of the next generation of improvement.
The current error and hallucination rate is probably going to get dramatically better. And certain tasks that no AI can do will have breakthroughs that allow a decent AI to emerge.
But those coming improvements aren’t going to make current AI suck that much less at the tasks it’s currently not well suited to, in the majority of cases.
Source: decades of experience finding clever ways to make previously impossible automation work well enough, and a solid amount of direct LLM experience.
Edit: And that’s a fascinating summary, and a great write up of an important and enlightening aspect of all this. Thanks for sharing it!