Meta trains its own language model with updates of its users and other public information.
The parent company of Instagram and Facebook, Meta, said early in the week that he would begin to train their artificial intelligence in Europe with its public content.
Meta launched its own meta ai in Europe in March. Originally, artificial intelligence was to be used in June last year, but the commissioning was delayed due to the EU data protection and privacy regulations.
But what does Meta’s announcement really mean?
“It simply means that updates and pictures that have been published on Facebook or Instagram, for example, are used to improve future artificial intelligence answers,” says a security expert Petteri Järvinen.
Artificial intelligence reads the updates and adds them to their own language model. Based on the information it collects, it learns to respond better and better identify images.
-If you have visited Paris and published a picture of the Eiffel Tower, there will be another new picture of the Eiffel Tower on Meta, which will allow the tower to be identified more reliably in the future, Järvinen gives an example.
Meta himself says that the purpose of the information is to teach artificial intelligence to understand European cultures, languages, dialects and history, and the use of humor and sarcasm in different countries.
The use of data can be prohibited
Meta’s users have received messages in which the company says they use their public information, such as publications and comments from people over the age of 18. According to Meta, this contains all the public information that Meta’s services have been shared after the account is created. In addition, the meta also uses users’ interactions with Meta artificial intelligence.
Petteri Järvinen says he interprets this so that only updates distributed to the user’s friends in a private account are not used to train meta.
Meta also offers users the opportunity to resist the use of their own information. In the email received by users, there is a link to a form that you can fill out Meta from using your own information to train artificial intelligence. The company also had a similar form last year, which was hard to find on social media.
The message sent by Meta states that if the user has already stated his or her opponents, he or she does not have to send another request. Järvinen estimates that the previous ban on Meta’s statement is still in force.
Meta has announced that it does not intend to interfere with private messages when training artificial intelligence. But could it be different in the future?
Järvinen says that Meta must apply for permission to use private messages.
– You can never make promises of the future, but reading private messages requires a new permit.
Petteri Järvinen, a security expert, urges you to be accurate with what information is provided to artificial intelligence. Joel Maalmi
How big a risk?
How worried about Facebook or Instagram’s regular user should be?
– You should always be worried about how much meta has already been spy on us, Järvinen says.
Järvinen says that there is always a risk, at least in theory, that the information collected from user updates entered into the language model comes out to another as an artificial intelligence answer.
“However, I do not believe that such recognizable information can be found or revealed there, because the material is terribly,” he says.
– It’s not worth losing your night’s sleep.
However, Järvinen further points out that even users who are not active in Meta’s services may, with the activities of their friends, need to be analyzed by artificial intelligence. For example, on Instagram, people can be marked in pictures. Järvinen says that in the future, artificial intelligence can learn to recognize people who would not necessarily want to do so.
– Many images in different environments and postures make it more reliable than just a profile image that can be old or edited.
Telling intimate things to artificial intelligence
Järvinen points out that training a language model for a private person may be less than the fact that technology companies are collecting a personal profile to target marketing. Information collected in the training of language models when it is not intended to be combined with a particular person.
Järvinen says that people themselves feed a lot of their personal information to artificial intelligence.
– Many people ask for life guidelines or make a medical diagnosis or even their laboratory results and ask for artificial intelligence to interpret them. After all, there is a lot of sensitive information that artificial intelligence in the future can use in its own answers to that person. Such information can be accumulated for many years.
In other words, artificial intelligence learns to know your own user too well.
Järvinen says he considers the use of artificial intelligence to intimate, personal issues as a worrying trend that has been spoken so far.
– Artificial intelligence may seem like a reliable and smart interlocutor. Man opens up to it, and does not realize that this is just a machine that is intended to make a profit for its authors.