Hot Potato: Facebook will soon become the latest platform to use user posts and photos to train AI models, and while users can opt out, Meta appears to have purposely made the process difficult, cumbersome and prone to failure.
Facebook's privacy policy will change on June 26, 2024, at which point public user content, rather than private messages, will be used to train and improve its AI services.
Meta offers a way to object to data scraping, but it's not as simple as checking a box.
If you want to opt out, first visit this help page and click on the section titled “I want to object to or restrict the processing of my personal information by third parties used to build and improve AI in Meta.”
Next, you will enter your country of residence, first name, last name, and email address. The form will ask you if Meta's AI models know about you, including providing prompts you entered that caused your personal information to appear in a response from Meta's AI models, features, or experiences. Meta will also ask for screenshots to prove that your personal information appeared in a response from one of its AI models.
The final section is an additional context box that asks for a brief explanation of your concern and request.
X-user Tantacrul writes that Meta also sends users a notification about the new AI feature. Clicking on this and then the “right to object” link takes them to a similar form with a section on “how this processing will affect you.” Meta also asks users to enter an OTP sent to their email address. The company claims this is for security reasons, but the fact that the submission isn't sent until the code is entered suggests this is another way to discourage people from filling out the form.
2. Let me start by saying that I've worked on growth teams running minimal-friction experiments for over a decade, so I know how to streamline experiences. Rule: Every additional step you add significantly reduces the percentage of people who will follow through.
– Tantacrul (@Tantacrul) May 26, 2024
Meta's notice states that “if the objection is successful, it will apply going forward.”
Tantakurul wrote that the company respected his objections, but that many of his friends and Reddit users who had similar objections received error messages when they tried to submit the forms.
Scraping user content to train AI models has become standard practice recently, and OpenAI partnered with Reddit earlier this month to publish user posts on ChatGPT.
