Meta has discovered a new source of training data for its AI models: its own employees. The company plans to use data culled from its staff’s mouse movements and keystrokes in its quest to build more capable and efficient artificial intelligence.
The article, first reported by Reuters, shows how long it can take for technology companies to find new sources of training data. Training data is the lifeblood of AI models, helping programs learn how to perform tasks more effectively and respond to user queries.
When contacted by TechCrunch for comment, a Meta spokesperson issued the following statement:
“If you’re building an agent that helps people use their computers to complete everyday tasks, your model needs real-world examples of people actually using the agent, such as mouse movements, button clicks, and drop-down menu interactions. To help with this, we’re launching internal tools to capture this type of input in certain applications to help train the model. Safeguards are in place to protect sensitive content, and the data is not used for any other purpose.”
This trend appears to reveal the AI industry’s troubling privacy implications, as yesterday’s internal corporate communications increasingly become the fodder for new corporate supply chains. Last week, it was reported that a legacy startup could be having its corporate communications (from Slack archives, Jira tickets, and other internal messaging platforms) harvested and turned into AI fuel.
