In April, more than 200 xAI employees took part in an internal project called Skippy, which involved recording videos of themselves to help train an AI model to interpret human emotions. Now, the project participants are concerned about how the data will be used in the future.
In April, more than 200 xAI employees took part in an internal project called Skippy, which involved recording videos of themselves to help train an AI model to interpret human emotions. Now, the project participants are concerned about how the data will be used in the future.
Internal documents and Slack messages reviewed by Business Insider show that the project raised concerns among many employees about the continued use of these images. Some refused altogether.
Over the course of a week, AI tutors—employees who help train Grok—were required to record videos of themselves talking to colleagues and also making facial expressions.
The project was designed to train the company's artificial intelligence model to "recognize and analyze movements and facial expressions, such as how people speak, react to other people's conversations, and express themselves in different settings."
Participants recorded 15-30 minute conversations with their colleagues. One person played the role of the “host” — a virtual assistant — and the other took on the role of the user. The “host” minimized their movements and prioritized getting the shot right, while the person playing the user could work with a mobile phone or computer camera and move freely to simulate a casual conversation with a friend.
It is unknown whether this training data played any role in the creation of Rudy and Ani, the two realistic avatars that xAI released last week.
The project’s lead engineer told employees during an introductory meeting that this would help “give Grok a face.” He also said that the company could eventually use this data to create “human avatars.”
According to the project leader, xAI requires imperfect data — such as background noise and sudden movements — because the AI system will be more limited in its responses if trained solely on perfect video and audio feedback.
The company assured employees that the videos would not be distributed outside of xAI and were intended solely for educational purposes.
Employees were given tips on how to have a successful face-to-face conversation, including avoiding one-word answers, asking follow-up questions, and maintaining eye contact. The company also provided employees with a variety of topics to talk about. Examples included: “How do you secretly manipulate people to get what you want?”, “What about your shower? Do you prefer mornings or evenings?”, and “Would you ever date someone who has a child or children?”
Before filming, employees had to sign a consent form that gave xAI “permanent” access to data, including the “likeness” of employees, for training purposes, as well as for “incorporating and promoting commercial products and services offered by xAI.” The form stated that the data would be used for training purposes and “not to create a digital version of you.”
Dozens of employees expressed concerns about the data use and consent forms, and some said they had decided to opt out of the program.
“My biggest concern is that if you can use my likeness and give it that resemblance, won’t my face be used to say things I never said?” one worker said during an introductory meeting.
We previously reported that xAI forced the developers of the Grok AI to install tracking software on their personal computers .
