Meta’s Data Collection for AI Training

A recent internal memo from Meta reveals the company’s plans to gather data from employees to train its artificial intelligence systems.

The training software, referred to as Mehta, will be installed on employees’ computers to monitor their activities at work.

Mehta informed staff that this directive would capture data from U.S.-based employees, including mouse movements, clicks, and keystrokes. This information will contribute to training Meta’s AI, enabling automated agents to handle tasks independently. As reported, this move aims to build more efficient AI systems.

A Meta representative stated that to develop agents that assist users in completing routine tasks, the company requires real-world examples of human behaviors. This data will encompass not just mouse movements and clicks but also actions like navigating drop-down menus.

Recent analyses suggest that Meta utilizes a platform called the Model Capability Initiative, which operates on work-related applications and websites to capture snapshots of employees’ screens.

Mehta described the initiative as an internal tool for collecting data on certain applications intending to enhance AI training.

Concurrently, a spokesperson reassured that employee data would be kept secure. According to the representative, safeguards are in place to protect sensitive information and the data collected will solely be used for training models.

The spokesperson emphasized that this data won’t be for performance evaluations and administrators won’t have access to it.

Questions were raised about the specific “routine tasks” Meta envisions its AI agents undertaking and whether these tasks mimic human activities; however, the company did not provide clarity on these points.

While internal memos remain confidential, Reuters has reviewed several, including one shared within the Meta Superintelligence Lab team.

“This is a platform where every employee can assist in refining the model simply by carrying out their daily responsibilities,” was a noted comment.

Recently, Meta’s chief technology officer, Andrew Bosworth, also communicated to employees that their roles would shift towards overseeing AI agents, suggesting that internal data collection would expand.

“Our vision is one in which agents primarily execute tasks, and we serve to guide, evaluate, and enhance their performance,” Bosworth reportedly stated.