- Meta is recording employee clicks, keystrokes, and screen activity to train AI agents on real work behavior
- The program is part of a broader push to build AI systems that can perform everyday tasks with minimal human input
- The move comes just ahead of reports of layoffs at the company
Meta has begun collecting everything its employees do as they go about their normal work to train its AI models, as first reported by Reuters. The Model Capability Initiative records mouse movements and clicks, keyboard keystrokes, and even occasional screenshots from computers used by Meta employees in the U.S. The company wants to observe how people actually use software, then feed that behavior into AI models so they can learn to do the same things.
Meta essentially wants to make its systems more reliable for the small actions that define a workday. That means everything from navigating a menu and moving between windows to parsing different website formats. These aren’t easily solved with text data alone.
“This is where all Meta employees can help our models get better simply by doing their daily work,” the internal memo states.
Article continues below
Training your own successor
AI systems are moving from generating content to performing actions. They are being trained to complete tasks that have always required a person at a keyboard. That requires more examples than just a list of steps to complete a task. They need to see how work unfolds. Meta’s approach is to capture those steps directly, turning everyday activity into training material.
Workplace monitoring has long existed, but Meta’s approach is more detailed and more specific in its purpose. The system records the fine-grained interactions that are usually overlooked, building a detailed picture of how tasks are completed in practice. According to the company, the data is not intended for performance evaluation, with safeguards in place to protect sensitive information.
The tracking program sits within a broader push at Meta to develop AI agents capable of handling everyday tasks. This Agent Transformation Accelerator focuses on building AI models for routine work across different tools and platforms.
The timing of the rollout is difficult to separate from other changes at the company. Meta is preparing to lay off around 10% of its global workforce, with more to follow. additional cuts expected later in the year.
All-seeing AI eye
Beyond how Meta plans to use the data, the level of detail the program is collecting is unusually comprehensive. Logging every keystroke and mouse movement is more familiar to factories and warehouses than corporate offices. It’s a new level of visibility, and possibly an uncomfortably intrusive one for many.
The fact that it’s happening in the U.S. is not surprising. Companies here are generally required only to inform employees of the surveillance, whereas European labor and data privacy rules put much stricter limits on this kind of oversight.
For Meta, the fact that need to be trained on examples of everyday tasks makes this monitor program the obvious move. Employees may feel less comfortable about having no choice but to expose every moment of their workday to observation and having that data used to potentially replace them and all of their coworkers.
If Meta’s program works like the company hopes, it’s unlikely to remain unique to the company. The demand for real-world behavioral data will increase as AI capable of carrying out those tasks becomes more common.
Meta wants to make AI models that can completely mimic what human employees do at work. Whether that leads to more efficient tools or just a more uncertain, and potentially depressed workplace depends on how those AI models are deployed, but there’s no question that they will be watching every click soon.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
The best business laptops for all budgets
https://cdn.mos.cms.futurecdn.net/wVydSQNQJReMzxQNudy2w7-2000-80.jpg
Source link
ESchwartzwrites@gmail.com (Eric Hal Schwartz)
