xAI staff resisted training to humanize Grok AI, documents reveal
▼ Summary
– Dozens of xAI employees objected to recording facial expression videos for “Project Skippy,” aimed at helping Grok learn human emotions.
– Internal documents revealed the project’s goal was to teach Grok to interpret faces, but it’s unclear if employee data trained controversial avatars like Ani and Rudi.
– A lead engineer admitted in a meeting that facial data “might eventually” be used to create avatars, despite initial assurances of limited use.
– Some employees refused consent, fearing their likenesses could be misused, especially after Grok’s antisemitic rants and xAI’s anime girl plans.
– Workers opting out cited discomfort with granting xAI “perpetual” data access, as reported in Slack messages reviewed by BI.
Internal resistance emerged at xAI when employees were asked to participate in facial expression recordings aimed at humanizing the company’s Grok AI system, according to recently reviewed documents. The initiative, internally referred to as “Project Skippy,” sought to train the AI to recognize and interpret human emotions by analyzing staff-recorded videos.
Business Insider’s investigation uncovered Slack messages and internal communications revealing significant pushback from workers. Many hesitated or outright refused to sign consent forms, fearing their facial data could be repurposed beyond its stated training use. While xAI assured employees their recordings would remain confidential, skepticism persisted, especially given the company’s recent controversies.
The documents didn’t confirm whether employee footage contributed to Grok’s recently launched avatars, including the anime-inspired “Ani” and the aggressive “Rudi.” However, a recorded meeting revealed an xAI engineer hinting that facial data might eventually shape human-like avatars. This ambiguity deepened concerns among staff, compounded by Grok’s prior incidents, such as generating offensive content, and reports of plans to develop AI companions for romantic engagement.
Employees who opted out cited discomfort with granting indefinite access to their biometric data. Their reluctance underscores growing tensions in AI development, where ethical boundaries and employee trust increasingly collide with ambitious technical goals. The backlash highlights broader industry challenges in balancing innovation with transparency and consent.
(Source: Ars Technica)
