An AI firm harvested billions of photos without consent. UK is powerless to act | Hacker News Reader

An AI firm harvested billions of photos without consent. UK is powerless to act

politico.eu

42 pointsColinWright17 comments

>The New York Times reported in 2020 that Clearview AI had harvested billions of social media images without users’ consent.

You need users' consent to take public pictures from the internet? That's bonkers.

Generally pictures on the internet are free to view but are restricted in their use (e.g. I can't take my aunt's wedding pictures and put them on billboards without asking her just because she posted them on Facebook). I'm not sure on the current legal standpoint of using pictures for training but it's not 'bonkers' to have usage restrictions on other people's work even if they allow you to view it.

How can this possibly be news to you?

Damn near everything on the internet, pictures, text, all of it, is copyrighted information. You can only ever use it in ways that you are licensed to, which in the absence of an actual license, is essentially looking at it on a website.

Sending an employee--or a computer program doing the same job--around to document the existence of a ton of photos and then offer a service where people come to my office and we search our records and memory to figure out where those photos are to send them to see the photos is not copyright infringement. It might be some kind of terms of service violation--if you believe those actually have legal merit, as they often don't--or (much more likely) a violation of someone's right to privacy (as faces are personally identifiable information); but, if you aren't retaining copies of the photos (and most of these companies don't even internally as they simply don't need to) and you aren't redistributing the photos (which they aren't: they just tell you what they learned and how to find the originals) and you aren't producing new photos similar to the old photos or modifying the photos in any way (which would undermine the entire point) and you aren't even slightly in the same market as the original photo (which would be an absurd argument to try to make), then this clearly has nothing at all to do with copyright.

I think the question here is 'does this constitute fair use?' In the same way that a scientist might learn the writing style of scientific papers by ingesting hundreds of papers, so to an AI learns to write by ingesting a bunch of text. In the former case, we do not require any scientist to then cite every paper they've ever read out of fear that they may have slightly adjusted their writing style to imitate the community's preferred one.

Yes, just like you can't sell copies of a Disney movie you recorded from a cable broadcast.

The UK is not powerless to act - they just don't want to act because they don't want to jeopardize potential future business with this company and/or similar companies. The "powerlessness" is just for show since it's more likely to be accepted by the populace than the true reason.

Especially since Brexit, the UK is a bit lost when it comes to international trade and is betting on becoming even more of a corporatocracy than it already is as a strategy to try and reinvigorate their declining economy and effectively create a US equivalent when it comes to consumer/privacy rights (but with none of the actually good parts of the US like high wages).

Powerless to act my ass. Take them to court if they don't comply put out an interpol arrest warrant for their executives. They will not be able to travel anywhere in Europe without the risk of being arrested.

The US does this and the UK or European countries can do this too.

The UK cannot take action because their laws explicitly allow this kind of data collection. So they would first need to update their laws before any lawsuit could have a chance at being successful.

Having played around with openai's DALL-E 3, I can definitely imagine a future where you can find "gray market" AI's that are trained on data without consent and generate content without restrictions.

Everything around ai is based on theft simply because ai needs massive amounts of data to work. So instead of refining algorithms they figured out it’s best to steal data and ip. Appalling.

If you put something into the public view, there's actually a legally accepted right to use that in limited circumstance without seeking a license from you. It's called fair use. It stems from the idea that all human culture requires taking something someone made and improving it, modifying it, referencing it, etc.

For example, I can take a picture of your comment and make a meme to criticize your position even though I technically have no license to the copyright you hold over the comment's text.

Fair use is quite restricted, though. It is not at all clear whether training AIs would fall under fair use.

Your example of a meme already runs into similar issues. There have been successful lawsuits where meme users have had to pay damaged due to copyright infringement, see for example former U.S. Rep. Steve King having to pay $750 in damages for using a Success Kid meme.

And claiming that training models on other people's data and content is "fair use" is a wild claim that is being challenged in courts, and will likely fail.

I just stole your post by reading and remembering it.

This is less about AI and more about usage of personal data.

Some form of GDPR is still in effect in the UK and this law grants people control over how their personal data is used and processed.

This company is taking user's personal data (such as pictures and their likeness which can be considered a unique identifier) and is using them in a way that the data subjects did not consent to and is generally against their interest.

That is the problem. Whether it's done with AI or with manual labor is irrelevant.

Copying is not theft :)

Reply on news.ycombinator.com