Hacker News Reader

The difference is scale. With ML Face recognition we can identify people in real time, without some safeguards we very rapidly reach a point where we can track anyone anywhere. Imagine if every camera in the city of London could identify you, forget metadata we will be able to actively track individuals taking public transit across a city and identify every individual who the interact with.

Yes I agree, I just think that you can build a system to keep a token person in the loop that will drive the cost down regardless. It's similar to how regulators tried to ban "computerized trading" and traders literally built a machine hand that presses the enter key.

Making the cost artificially high by banning a particular type of technology is not addressing the real problem of civil liberties.

"Making the cost artificially high by banning a particular type of technology is not addressing the real problem of civil liberties."

I think it helps.

Wasn't that the point?

You can't seriously expect to build a system like big brother and be so naive as to think this wouldn't be result sooner or later.

> I don't understand how you can ban a machine from doing something that a human can do.

We do it all the time.

- Phone autodialers are legal, but particular uses are fairly heavily regulated (in theory, at least).

- Heavy machinery use is regulated in a variety of ways.

- LEOs in the US recently were told GPS bugs (which "just" do what humans can do - taking notes on where a human goes) require warrants.

- Explosives dig much faster than humans. These are heavily regulated.

- Most radios sold in the US will refuse to tune certain bands for legal reasons, even though it is trivial to modify some of them to do so.

Etc.

As far as your slippery slope argument, it simply doesn't matter. The point is to reduce the harm done by a given tech, not achieve some sort of abstract purity of thought.

None of these steps involves looking up a face in a seventeen million page binder of faces with names and associating names, government ids, traffic violations, credit scores, etc with the person.

I don't think people are worried about computer vision identifying Human vs Not Human as much as they are about the (not humanly possible) pinpointing and following of every move.

> I don't understand how you can ban a machine from doing something that a human can do.

Quantity has a quality all its own. Something that is acceptable at a small scale may have unacceptable consequences when done at much larger scales.

Same reason you might want place more restrictions on machine guns than on muskets.

The reason one would ban a machine from doing what a human can do is that it's stupendously faster, and enables all sorts of dystopian effects that manual face-tagging doesn't.

The difference is that machine-driven facial recognition can be collated with other data to create unwanted results for society.

Rando Startup can start scanning faces in public. Then it starts matching those faces with actual identities. Then it adds in the cell phone location data to determine which ones visit synagogues regularly. So it adds a little notation to the data table about each of these people. Perhaps a yellow star will do nicely.

If you know anything about recent European history, you will understand why this is a bad thing.

The point is to prevent a reality where some people can track the whereabouts of all (or nearly all) people living in a city in an efficient way. You could of course try automating this using humans, but the cost will indeed be much higher, the efficiency much lower, and even this process itself might be banned if it were to lead to the same result.

That's actually a good extreme showing the difference. Tracking literally everyone all the time would require a huge number of police if done manually. And with that many police, you'd need police to track the police because they are people too. It gets ridiculous fast. With tech, it actually becomes feasible.

Part of the problem with facial recognition is it is potentially inaccurate, especially if you are not a white male.

So you run the risk of having large numbers of innocent people being incorrectly tagged as persona non grata - be that with the police, in-store detectives, hotels, potential employers etc.

Having a computer say "this person is bad/unwelcome/a shoplifter/bad credit/a sex offender/etc" is powerful, and difficult for laymen to counteract. Computer says no - sorry, it's policy, nothing I can do.

And you can do this at huge scale for pennies and in the blink of an eye with a computer. You can't do that with a human manually/semi-manually doing it.

white male white male white male

> Asian and African American people were up to 100 times more likely to be misidentified than white men, depending on the particular algorithm and type of search. Native Americans had the highest false-positive rate of all ethnicities, according to the study, which found that systems varied widely in their accuracy.

https://www.washingtonpost.com/technology/2019/12/19/federal...

It doesn't matter. "White male" is a slur now.

Well, making the cost artificially high can be a point in itself. Sometimes the concern about a technology is the ubiquity and scaling.

But aside from that, to address your questions:

1. If there's a human in the loop, the human can be accountable. Explainability and accountability of algorithmic decision-making is an emerging concern, especially to the extent that algorithms may encode bias in their training data.

2. Real-time systems. Being real-time qualitatively changes the impact this can have on your day-to-day life. Taking the human out of the loop makes many applications feasible that otherwise wouldn't be. Like doors.

Other commenters have made the point that scale / quantity matters. Another example: Formula 1 has limits on computational performance. Humans can design aerodynamics. Machines can do it better, faster, more exhaustively. The latter is quantitatively limited.

Scale does matter. Consider how your gut-feel attitude toward a thief that have stolen $1 vs a thief that have stolen millions.