Ban Facial Recognition at Festivals

1- For face f, is f a member of F_known? (From national ID cards etc.) Assume this gives an answer [0.0 - 1.0].

2- Is f a member of F_suspect? Where F_suspect is a database of face imagery associated with potentially illegal activity or refugee migration. Not that these faces could be from video surveillance (from a border region, train station, migrant camp) or scraped from war zone images e.g. jihadi propaganda, YouTube video of regime critics, talking politics on HN, video of protests in other countries, etc.

3- Not matching anything, and hence also requiring investigation.

All of these have errors, which generally go up as more people are added to the training database. Apparently some ethnicities (okay, black people) are often trained for poorly. This will likely result in 'over-servicing' of some ethnicities.

If local police get a hit for someone going through festival security and detain them on the basis of face recognition software, what is the legal basis for that? Police would say that they had 'intelligence' but can they explain the basis for it? Is this not essentially maintaining dossiers on people who have not committed any crime? They have no evidence a crime has been committed but instead they are essentially running a police checkpoint and saying "Papers!". These are the mechanisms of the Stasi.

One can argue that police should be able to detain those with outstanding warrants. I have no problem with that, except that when police take photos of citizens it is hard to know what is done with those photos. Ironclad legislation requiring they not leave the device or be stored on anything except ephemeral storage would be a good start, along with civil and criminal penalties for non compliance and direct individual responsibility and penalties for the chain of command.

Without a rigorous push back, America will become like China and the UK, where your face must be shown, and it will be recorded and tracked 24/7. Of course in America it will probably be via a company, and police forces will have a subscription, which lowers their fees the more faces the capture.

I would also note that Nielsen rating are based on an incredibly small number of Nielsen listeners. Their system might be good enough for advertising, but not for dragnet surveillance.

Look, I get that if you are worried about illegal immigration, then using FR at Latin concerts has a rough logic. My point is that targeting ethnic groups for mass surveillance is not good for anyone.

(Note: I see where you said the government should probably not do that, but the rest of your post seemed to support the practice.)

There's no need to be uncivil or rude.

Let's take Example 1.

It would indeed be problematic if we convicted people based solely on the algorithmic conclusion of a machine. But thankfully, we still have jury trials in the U.S.; it requires a unanimous verdict to convict someone; and it requires a level of proof "beyond a reasonable doubt."

It's important to understand that a machine is not a witness. It can't be cross-examined. So whatever result it produces is evidence. How strongly that evidence is believed is up to the jury; and a good defense attorney is going to do whatever it takes to attack the machine (using expert witnesses and so forth) to introduce reasonable doubt if its conclusion is incorrect. The defense attorney is also going to introduce whatever other exculpatory facts and witnesses they can find.

In your example, if a sexual assault occurs, the victim is likely to testify. Without the victim's testimony identifying the defendant, and without any corroborating DNA evidence, the defendant is likely to walk, even with the machine saying it's 90%+ probability match.

In any event, I think what's needed here is some additional law forbidding the sole use of FR data as evidence used to convict a defendant. That, I think, would help assuage most people's concerns -- even though I think it unlikely that a conviction would result this way anyway.

A really important fact that you're not considering is that FR has a strong potential to prevent false identification. Many studies have shown that eyewitness identifications, including line-ups, are extremely unreliable; and yet, we've been using them for decades to convict people. And we've released innocent convicts many, many times, sometimes decades after imprisonment, when it turned out the witness was lying or simply had a faulty memory. And in your example, if you in fact went to the toilet during the crime, the cameras would be able to corroborate that.

OTOH, if a machine strongly disagreed with the eyewitness, I have little doubt that the defendant would be dismissed.

Example 2 is a bit trickier.

You don't need FR technology to have a police state: North Korea has had one for nearly 70 years without one. East Germany had one for over 40 years (see: Stasi). There are cruder methods than using FR technology; spying is a very old game.

So, FR can help enhance control for regimes that are already inclined to preventing freedom by making it much more economical than having a spy network, but it's by no means necessary. Besides, you can trivially foil FR by wearing a mask.

I don't know how to prevent a police state, other than to employ violent resistance, and even that doesn't always work. I don't think anybody knows. But if you want a prevent a police state, I think banning FR isn't really going to help. At least so far, history shows that many police states resulted from democracies or republics ceding control under some sort of "necessity."