A hot potato: Facebook will soon become the latest platform to utilize posts and photos from its users to train its AI models. People can opt out, but Meta appears to have made the process intentionally difficult and cumbersome. It's also prone to failure.
Facebook's privacy policy will change on June 26, 2024, at which point public user content – not private messages – will be used to train and improve its AI services.
Meta provides a way to object to the data scraping, but it isn't as simple as checking a box.
If you do want to opt out, first go to this help page and click on the section that states, "I want to object to or restrict the processing of my personal information from third parties used for building and improving AI at Meta."
It's then a matter of filling in your country of residence, name, surname, and email address. The form asks whether AI at Meta's models have knowledge of you. This involves providing any prompts you entered that resulted in your personal information appearing in a response from an AI at Meta model, feature or experience. Meta also asks for a screenshot proving your personal information appeared in a response from one of its AI models.
The last section is an additional context box that asks for a brief explanation of your concern and what you're requesting.
X user Tantacrul writes that Meta is also sending out notifications to users about the new AI features. You can click on this and then a "right to object" link to take you to a similar form that includes a section about "how this processing impacts you." Meta also requires users enter an OTP sent to their email address. The company claims this is for security reasons, though the fact the submission won't be sent until the code is entered suggests it's another way of trying to put people off filling in the form.
2. I should start by mentioning that I've worked in growth teams who conduct experiments to minimise friction for over a decade and I know how to streamline an experience. Rule: every additional step you add dramatically decrease the % of people who'll make it through to the end.
– Tantacrul (@Tantacrul) May 26, 2024
Meta's notice states that "If your objection is honored, it will be applied going forward."
Tantacrul writes that the company did honor his objection, but many of his friends and Reddit users who also objected received an error message when trying to submit the form.
Scraping user content to train AI models is becoming a standard practice these days. OpenAI partnered with Reddit earlier this month to put users' posts in ChatGPT.