Artificial intelligence systems that are developed to work more efficiently can be overlooked by their creators, who only consider the technology’s impact on efficiency. Technology companies create algorithms that optimize efficiency, but the human impact of such algorithms is often not considered until someone is shocked by the algorithm’s true price. A new legal battle involving Meta Platforms is doing just that, with questions being raised about whether automation is beginning to make decisions about the workforce, decisions that should stay well in the hands of humans.
A federal lawsuit filed by 26 former and current employees of the tech company alleges that the firm employed AI-driven software to discriminate against people with disabilities and those who took medical leave in its mass layoffs. Plaintiffs say that productivity metrics and usage of AI tokens were among those factors that Meta considered when cutting thousands of jobs earlier this year, all of which are factors that inherently disadvantage employees who were unable to work because of a medical issue or family duty to care for others.
The case seems to be the first to go after a large U.S. firm for allegedly using AI to make layoffs. Amidst the companies that are trying to apply AI to all aspects of their business, there is a growing concern about whether AI can be used to make fair and unambiguous decisions with respect to human employment. Even the most complex and advanced AI products could be trained to mirror or reinforce pre-existing biases in real human contexts, as suggested in the Meta lawsuit.
Plaintiffs were told in May that their jobs were to be eliminated as of July 22. Instead, they are looking for a preliminary court order to stop Meta from making the cuts while they try to seek a settlement through private arbitration. The workers claim that Meta’s employment agreements mandate that workers resolve disputes on their own before an arbiter, except when they’re seeking temporary relief from looming harm.
On Tuesday, a spokesperson for Meta responded to the lawsuit, vigorously challenging the claims. The spokesperson said that while AI may have been assisting in making workforce management and organizational decisions, it was a human being who ultimately did the decision-making. In this defence, Meta will say that people were ultimately responsible for determining who to fire, even if the AI tools gave information and advice.
The case poses difficult issues of corporate responsibility in the context of algorithms in the workplace. While the decisions to lay off may ultimately be made by a human manager, plaintiffs can still show that the AI system was designed or used in a manner that created discriminatory results. Federal employment discrimination laws allow for companies to be liable for policies or practices that have a disparate impact on protected groups, whether intentional.
The metrics at the heart of the case provide a window into the technologies and metrics used to assess employee performance and productivity. In the tech industry, there is a new measurement for productivity — the amount of AI tokens that employees use with AI systems. If someone took medical leave or needed to tell their employer that they had a disability that required them to take some time off, they would be less likely to use tokens than a colleague without such a disability who worked full time without interruption, and they may be misjudged by an algorithmic evaluation system.
It is not the first time that Meta has come under fire for its workforce cuts. In the last several years, the company has shed around 20,000 jobs as part of a wider cost-cutting initiative that CEO Mark Zuckerberg called a “response to the evolving market conditions. Some departments and groups of workers have been hit harder than others by these cuts, and the cutbacks have been hard on employee advocates and civil rights groups.
The plaintiffs are litigating their case in a multi-prong fashion, typical of today’s employment litigation. The move for a preliminary injunction to block the layoffs and then to arbitrate is designed to lock them in for the time being while they attempt to make that case to the arbitrator for a long-term solution. This double approach takes into account that although they may ultimately win in arbitration, being forced out of their jobs in the interim may do irreparable damage.
The case is closely followed by experts in employment law, as its decisions might shape the future for other companies looking to cut staff in today’s increasingly AI-driven business landscape. Should the court rule that Meta’s algorithmic tools in layoff selection may be discriminatory, it’s possible that companies will reassess their approach to using AI for HR decisions. The case may also motivate other employees who think they’ve been unfairly singled out by automated systems to step forward with similar claims.
The case also exemplifies a larger social issue regarding the use of AI in making life-altering decisions for average citizens. The use of Artificial Intelligence (AI) in decision-making processes is becoming more common than ever, and it has the ability to significantly impact people’s lives in various ways, such as in hiring decisions, credit scoring, and criminal justice risk assessment. The Meta case is a good example of how these systems can fail to capture the messy nature of human experience: illness, family responsibilities and personal obstacles will inevitably disrupt even the most diligent employee’s work rhythms.
It’s actually a noteworthy case because of the timing. The Meta lawsuit could serve as a practical application of where the current anti-discrimination laws stand with regards to algorithmic decision making as the Biden administration and other state legislatures propose new rules on the use of AI in hiring. The laws—such as the Civil Rights Act and the Americans with Disabilities Act, which were enacted decades before the advent of AI in the workplace—should be applied to algorithmic discrimination as they are to human discrimination, the plaintiffs argue.
