Meta needed employee data for AI edge, but faced a very 'human problem'

Meta's ambitious venture to bolster its artificial intelligence capabilities by leveraging employee data presented a stark human dilemma within the company. To overcome AI limitations, Meta's elite AI unit, TBD, conceptualized the Model Capability Initiative (MCI) to gather insights on employee computer usage patterns, such as keystrokes and screen activity. However, this ambitious project quickly encountered resistance from Meta's workforce.

Employees exhibited strong opposition to MCI, citing concerns over data privacy and potential misuse. Sensitive information, including passwords, immigration documents, and credit card details, were deemed vulnerable to capture through the program. Fears of job automation further fueled discontent among workers. As the backlash intensified, internal communications were flooded with anti-MCI memes, and the company's recording icon became a target of mockery as the "Eye of Sauron".

In response to the mounting opposition, Meta initially allowed employees to opt-out for 30 minutes to personally manage their affairs. Nevertheless, activists decried this measure as insufficient and demanded the immediate cancellation of MCI. The tide turned with a significant data leak towards the end of June, exposing that the collected data was accessible across Meta and even contained private conversations.

This revelation rapidly spread throughout the company, confirming employees' apprehensions about data control. Consequently, Meta swiftly shut down MCI and announced its suspension, with plans to revisit the initiative on an opt-in basis should it be reconsidered. Despite these actions, uncertainty lingered regarding the extent to which the gathered data was utilized in training Meta's AI models.

Reports suggested that the data had not been employed, while Meta's Chief Technology Officer, Andrew Bosworth, indicated the data's unanticipated value.