Meta infiltrates its employees’ computers to train the AI
Meta has installed new tracking software on its employees’ computers to compile data and train AI.
Data is the lifeblood of AI, and that fuel is as in demand as it is sadly scarce. For this reason, Meta has decided to take a drastic (and potentially controversial) step to secure the data needed to train its AI models. The parent company of Facebook and Instagram has installed new tracking software on its employees’ computers in the United States, specifically designed to record mouse movements, clicks, and keystrokes.
The installation of this new software is part of a broader initiative whose ultimate goal is to create AI agents capable of performing tasks autonomously, according to an internal memo Meta sent to its employees and seen by Reuters.
To date, AI agents have encountered numerous problems replicating certain interactions between humans and computers (such as using keyboard shortcuts or selecting options from a drop-down menu). And with the ultimate goal of solving these kinds of problems, Meta has decided to install software on its employees’ computers that, like a kind of spy, will be constantly monitoring them.
The software, called the «Model Capability Initiative» and part of the «Agent Transformation Accelerator» (ATA) initiative, will also regularly take screenshots of the content displayed on employees’ computer screens.
Meta wants its employees to help the company refine its AI agents.
«If we want to develop agents that assist people in performing everyday tasks on the computer, our models need real-world examples of how people carry out these kinds of tasks,» says Andrew Bosworth, Meta’s CTO, in the memo the company sent to its staff. «We’re working on a vision where AI agents do the work, and our role is to guide, monitor, and help them improve,» Bosworth adds.
Aware that its new initiative could cross some red lines, Meta has specified that the software installed on its employees’ computers will be limited to recording data about applications and websites strictly related to work. Furthermore, the compiled information will not be used under any circumstances to evaluate the performance of Meta’s workforce.
The company is also encouraging its employees to use AI agents to perform tasks such as programming, even though the use of this technology (still prone to errors) might initially translate into more work.
AI providers have high hopes for AI agents, which are designed to take on complex tasks (developing apps or analyzing vast amounts of data) with only limited oversight from human staff. If the AI agents used by companies like Meta reach the level of sophistication necessary to eventually replace human workers, some companies would be prepared to make severe staff reductions.
Meta, for example, will lay off 10% of its global workforce starting next May and is considering further cuts throughout the year. Amazon has also eliminated 30,000 jobs in recent months (around 10% of its workforce), and the fintech company Bock went even further, laying off nearly half its staff last February.
At Meta, AI is of paramount importance. So much so that Mark Zuckerberg, the company’s CEO, is reportedly working on an AI avatar that would mimic his voice and gestures and whose purpose would be to interact with the multinational’s employees. Furthermore, Zuckerberg is also said to be actively using AI agents to stay informed about the company’s performance and make important decisions.
Source: www.marketingdirecto.com
