#laion-5b

LAION set out to build a dataset for these newer, hungrier models. They built a dataset that is purely constructed by machine processes, by running models and tweaking thresholds: LAION-5B is made by measure.

But what is getting measured? The quality of data? The capacities of CLIP? The success of a model against a benchmark? The benchmark itself?

[...]

Openness in the AI field matters, not just for model biases, but for the structural biases in the ecosystem. An ongoing problem is that curation by statistics amplifies many of those structural biases."

People care TOO MUCH about the IP laws of dumb cartoons like Mickey Mouse than the real abuse of data going on, acting like the only conversation worth having is "is copyright good or bad", but as a med student I have a vested interest in talking about data collection ethics.

You're welcome to address my bias or in less kind words say I'm in the pocket of "big pharma" or that I'm a "copyright maximalist" but I'm doing this purely to explain and educate how the LAION team is dishonest, manipulative, malicious and hides behind the good graces of "open-source" and "non-profit".

To start; how does LAION get hold of your photos? To put it shortly: Common Crawl, a service that has indexed and scraped web pages from 2014 until 2021. But, unlike LAION Common Crawl has a TOS, and states on their website that they do not allow users to violate IP or circumvent copy-protection using their data.

The highlights in orange are important, but for future points.

So how does this affect medical photos? "They shouldn't be on the internet in the first place!" You might say. This is where things get a bit muddy, because in the most popular case being spread the user has signed a consent forum allowing the use of their photos in medical journals, seen here;

Please make note of the first line, "to be used for my care, medical presentations and/or articles".

So how did it get online?

Despite what a lot of people jump to assume, this most likely was not the fault of the doctor – and unfortunately he's not alive anymore to even clarify what went wrong, RIP. There are many journals online on the user's condition – one which is particularly rare and as such requires study and photos for identification, many with attached images that have been scraped too. This user is most certainly not alone.

For background, PubMed is the largest site for sharing medical journals and case studies on the internet. It contains a wealth of information and is crucial to the safety and sanity of every overworked med student writing their 30th pharmacology paper. It also has attached images in some cases. These images are necessary to the journal, case study, research paper, whathaveyou that's being uploaded. They're also not just uploaded willy nilly. There are consent forms like the one seen above, procedures, and patient rights that are to be respected and honored. What I want to emphasize,

Being on a journal ≠ free to use.

Being online ≠ free to use.

If you do not have the patient's signed consent, you are not allowed to use the image at all, even in a transformative manner. It is not yours to use.

So how does LAION respond to this? Lying like shitty assholes, of course. Vice has done a very insightful article on just what LAION has stored within it and showing many harrowing stories of nonconsensual pornography, graphic executions and CSEM on the database, found here.

A very interesting part of the article that I'd like to draw attention to, though, is LAION team's claims about the copyright applied to these images. The claim in blue that all data falls under creative commons (lying about the copyright to every image) directly contradicts the claim in red (divorcing the team from copyright).

The claim in orange is stupid because it claims photos of SSNs and addresses directly linked to your name are not personal data if they dont contain your face. It also is not GDP compliant, as they elevate their own definition of what private data is over what your actual private data is.

But whatever, team LAION is on this!! They got it, they'll definitely be pruning their database to remove all of the offending– aaaand they literally just locked the discord help channel, deleted the entire exchange and accused Vice of writing a "hit piece", as reported on by motherboard here. Classy, LAION!

They don't even remove images from their database unless you explicitly email them, and even then they first condescendingly tell you to download the entire database, find the image and the link tied to it, then remove the image from the internet yourself– somehow. Classy, LAION.

Of course, the medical system isn't completely free from blame here, from the new motherboard article;

Zack Marshall, an associate professor of Community Rehabilitation, and Disability Studies at the University of Calgary, has been researching the spread of patient medical photographs in Google Image Search Results and found that in over 70 percent of case reports, at least one image from the report will be found on Google Images. Most patients do not know that they even end up in medical journals, and most clinicians do not know that their patient photographs could be found online. 

“[Clinicians] wouldn't even know to warn their patients. Most of the consent forms say nothing about this. There are some consent forms that have been developed that will at least mention if you were photographed and published in a peer-reviewed journal online, [the image is] out of everybody's hands,” Marshall said. After hearing about the person whose patient photograph was found in the LAION-5B dataset, Marshall said he is trying to figure out what patients and clinicians can do to protect their images, especially as their images are now being used to train AI without their consent. 

It's a case of new risks that people have not been aware of, and of course people can't keep up with the evolving web of tech bro exploiters chomping at the bit to index every image of CSEM and ISIS beheading they can get their hands on. If artists are still trying to get informed on the topic, expecting doctors who share this information for the benefit of other doctors to be hiding it behind expensive paywalls and elaborate gates just to cut off the tech bros is asinine. But regardless, if you don't want to go on a journal, now you are aware of the possibility and can not consent to it in the future.

LAION however can't be held accountable themselves because, despite facilitating abuse, they're not direct participants in the training of data, they just compiled it and served it on a gold platter. But on the bright side, The Federal Trade Commission (FTC) has begun practicing algorithmic destruction, which is demanding that companies and organizations destroy the algorithms or AI models that it has built using personal information and data collected in a bad faith or illegally. FTC Commissioner Rebecca Slaughter published a Yale Journal of Law and Technology article alongside other FTC lawyers that highlighted algorithmic destruction as an important tool that would target the ability of algorithms “to amplify injustice while simultaneously making injustice less detectable” through training their systems on datasets that already contain bias, including racist and sexist imagery. 

“The premise is simple: when companies collect data illegally, they should not be able to profit from either the data or any algorithm developed using it,” they wrote in the article. “This innovative enforcement approach should send a clear message to companies engaging in illicit data collection in order to train AI models: Not worth it.”

This is likely going to be the fate of any algorithms that take advantage of the illegal data collected in LAION-5B.

So what do we take from all of this?

Please read consent forms thoroughly.

Algorithmic destruction should befall LAION-5B and I wouldn't mind if every member on the team is arrested

That's it, that's the whole thing 😊