What if they are using photos of your child to train an artificial intelligence?

 

On May 22, half of humanity received an email that did not go unnoticed. Facebook, which has more than 3 billion monthly active users, and Instagram, which has another 2,000 (although some of them probably also have Facebook), informed their users that their data would be used to train artificial intelligence (AI) models. ) in which Meta, the parent company of these social networks, works.

Meta enabled a way for those who wanted to exercise their right not to participate in that game. However, the process was cumbersome and, at times, did not work. The data protection authority of Ireland, the country where it has its European headquarters, asked the company founded by Mark Zuckerberg to freeze its initiative. Finally, Meta relented and paralyzed him in Europe on June 14 .

Four days earlier, the NGO Humans Rights Watch (HRW) published an investigation revealing that photographs of minors taken from the Internet without consent were being used to train “powerful AI tools.” The LAION-5B image repository, which several developers use to feed their databases, “contains links to identifiable photos of Brazilian children,” the NGO stated. “The names of some of them appear in the caption or in the URL where the image is stored. In many cases, their identities are easily traceable, including information about when and where the child was at the time the photo was taken,” HRW concludes.

The organization detected at least 170 photos of Brazilian minors entered into the database without permission. After HRW's notice, LAION-5B recognized that they had those images and proceeded to remove them from their repository. “Children should not live in fear of having their photos stolen and used against them,” said Hye Jung Han, a HRW researcher specializing in children and technology. Shortly after, Hye found 190 photos of Australian boys and girls also brought into LAION-5B without their parents' consent. Some of them had been taken from YouTube.

Are Brazil and Australia isolated cases? Hye's analysis suggests not. It only examined about 5,000 images of children from each country, which represents around 0.0001% of the LAION-5B repository. “We are not aware that images of children are used in AI training in Europe. However, it is very likely that these images are being collected and used in AI models on our continent,” says Felix Mikolasch, a data protection lawyer at NOYB (acronym for None of Your Business , none of your business), a Austrian organization dedicated to defending the digital rights of European citizens.

What type of consent is required to use photos of children? “Minors under 14 years of age cannot provide valid consent for these purposes. Below that age, the consent of both parents, the guardian or similar is required,” explains Jorge García Herrero, a lawyer specialized in data protection.

“The models do not memorize images, they only learn the underlying mathematical patterns to manufacture synthetic images that are precise to those of the training database,” says García Herrero, who is a data protection delegate (the figure that ensures compliance with the regulations) of a company that has an important image bank. “If you do things right, you ask minor models or their parents for consent to use the image and remove the context of the photos,” he says, referring to metadata (such as the date or place the photo was taken). photo) that can help identify the subjects.

What should we do if we discover the photo of someone we know in one of these repositories? “Citizens can request its deletion directly. You can also contact the AEPD. In addition, there is the possibility of claiming damages in court,” says Mikolasch.

Complaints to Meta

Hours after Meta sent the email to all Instagram and Facebook users telling them that the company wanted to use their data to train its AI, lawyer Luis Gervás de la Pisa filed two complaints. One before the Spanish Data Protection Agency (AEPD), and the other before the Irish authority. “The measure violates my right of access and deletion. That is, I want to know if my data is actually used for that task and they are not asking me for consent to do what they say they want to do,” he says. The complaints are now stopped, while Meta decided to freeze its decision.

The NOYB organization, for its part, filed another 11 complaints with the data protection authorities of several EU countries. “Meta is basically saying that you can use any data from any source for any purpose and make it available to anyone in the world, as long as it is done using AI technology. This is clearly the opposite of complying with the General Data Protection Regulation,” said lawyer Max Schrems, founder of NOYB.

Facebook and Instagram are not the only social networks that want to take advantage of their users' data to train their parent company's AI. X, the old Twitter, also does this by default. Those who do not want their tweets and private messages to be used to train Grok, Elon Musk's generative AI tool , have to deactivate that function (which is activated by default) in the platform settings.

 

MANUEL G. PASCUAL AUG 01, 2024 El Pais Spain