People love to talk to AI –Some, a little too many. And according to the contract YuanThey reviewed people’s interactions with company chatbots to improve their artificial intelligence, and people are a little willing to share personal private information with Meta’s AI, including their real name, phone number and email address.
Business insiders and Four contract workers hired by Meta Alignerr and Scaling AI – Outliers that havethe two platforms attracted human reviewers to help train AI compared to other Silicon Valley clients, with contractors pointing out that “unedited personal data is more common for the metaprojects they work in.” According to these contractors, many users on Meta’s various platforms, such as Facebook and Instagram, are sharing highly personal details. Users will talk to Meta’s AI as if they are talking to friends or even romantic partners, sending selfies or even “definite photos”.
Be clear, People are too close to their AI chatbots Have a sufficient documentary and meta-practice – use Human contractors evaluate the quality of AI-driven assistants To improve future interactions, this is nothing new. As early as 2019, this guardian Report on how Apple contractors hear very sensitive information regularly Even if the company didn’t have “no specific programs to handle sensitive recordings” at the time, that’s true from Siri users. Similarly, Bloomberg reports on how Amazon has thousands of employees and contractors Manually view and transcribe clips from Alexa users around the world. vice and motherboard It was also reported that contractors hired by Microsoft recorded and reviewed voice content, although this means that contractors often hear children’s voices through unexpected activations on their Xbox consoles.
But the Mon is a different story, especially considering the past decade’s track record of relying on third-party contractors and the company’s failures in data governance.
Meta’s study record on user privacy
In 2018, this New York Times and this guardian Reported how Cambridge Analytica, a political advisory group funded by Republican hedge fund billionaire Robert Mercer, uses Facebook to harvest data from thousands of users without consent and use that data to outline our voters and target them with personalized political ads to help them elect Donald Trump from 2016, and a person from 2016, which also comes from their humanity, Quiz Quiz Quiz Quiz Quiz App, and in their crimes, their participants, their participants are their crimes. It causes Facebook is fined $5 billion by the Federal Trade Commission (FTC)), one of the largest privacy settlements in U.S. history.
The Cambridge Analytica scandal and Facebook’s developer platform exposed a wider range of issues, which allows for large amounts of data access, but has limited oversight. According to internal documents released by Frances HaugenIn 2021, Meta’s leadership often prioritizes growth and engagement over privacy and security issues.
Meta also faces scrutiny on contractor use: 2019, Bloomberg reports on how Facebook pays contractors to transcribe users’ audio chats First of all, don’t know how to get them. (Facebook at the time said the recording only came from users who chose transcription services, adding that it also “paused” the practice.)
Facebook has spent years trying to fix its image: It was renamed to Meta in October 2021, building a name change As a forward-looking focus turns to a “meta-evaluation” rather than a response to controversies surrounding misinformation, privacy and platform security. However, Meta’s legacy in processing data takes a long shadow. While using human reviewers to improve large language models (LLMs) is a common industry practice, the latest reports on Meta’s use of contractors, and information contractors say they can see, does raise new questions about how the world’s most popular social network parent company handles data.
In a statement wealth, A Yuan spokesman said the company has “strict policies to manage access to personal data for all employees and contractors.”
“While we work with contractors to help improve the quality of training data, we intentionally limit the personal information they see, and we have procedures and guardrails to guide them on how to deal with any such information they may encounter,” the spokesperson said.
“For projects focused on AI personalization… the contractor is allowed to access certain personal information during the course of work in accordance with our publicly available privacy policy and AI terms. Regardless of the project, any unauthorized sharing or abuse of personal information is a violation of our data policy and we will take appropriate action,” they added.
