February 20, 2024
Get Ready for the EU's AI Act
The EU is about to start regulating the use of AI in employment decisions. The requirements make sense, but I don't know if they will make a difference.
While the new law is not final, it looks like employers will have to give notice that they use AI in employment decisions, provide competent human oversight, and monitor outcomes. If problems arise, the employer must stop using the system.
This increases the risk of using AI in employment decisions. But it probably doesn't change things too much. Giving notice is not that hard and most larger organizations are already using it. So, it's not like candidates are going to have a choice. Even a human only hiring track is required, who would demand that and make themselves appear to be a troublemaker. As someone with troublemaking inclinations, this risk is real.
Most organizations also have human oversight and human participation in the hiring and decision making processes. This may or may not help, since the primary problem with AI systems is that they are biased because of past decisions and actions by humans. And you're completely bonkers if you are not monitoring the outcomes because the risk of a discrimination claim is also very real. It's not just data; it's evidence.
While the law only applies to systems intended to be used in the EU, that covers a lot of global organizations. And similar regulation is probably coming to AI systems in the US and other places.
Here is a great discussion about what we know so far about the requirements of the EU Artificial Intelligence Act.
On February 2, 2024, European Union policymakers reached an agreement on the final text of the Artificial Intelligence Act (AI Act), taking us another step closer to the world’s first comprehensive AI legislation, and, fortunately, giving employers much-needed guidance on their forthcoming notice, human oversight, and other obligations.
Quick Hits
Following a provisional agreement in December 2023, and last week’s leak of the latest and purported final draft, EU policymakers have now officially agreed on the final text of the AI Act, which appears to be more prescriptive for employers than the original draft. Following procedural formalities, including a formal vote and publication in the Official Journal of the European Union, and barring any surprises, the AI Act will enter into force twenty days later and become effective twenty-four months after that (with some exceptions). Although that may seem like a long time, getting into compliance will be a heavy lift for employers. And the lift is likely to be even heavier, given that EU member states may implement their own regulations in the interim with shorter compliance windows. In fact, Portugal has already codified notice obligations for employers using AI in the workplace.
Employer Obligations Under the AI Act
Under the AI Act’s four-tiered, “risk-based approach” to regulating AI, the higher a tool’s risk, the stricter the rules governing it. AI tools intended to be used in the workplace are generally considered “high risk,” and therefore subject to significant regulations. Fortunately, most of the obligations fall on those creating AI tools, but there are also requirements for employers.
Generally, under the AI Act, employers must:
Other requirements may apply depending upon the circumstances, such as the employer’s industry, what the AI system does, and whether the employer has control over the data inputted into the AI system.
Extraterritoriality and Heavy Fines
The AI Act is relevant for any employer using AI systems with output “intended to be used” in the European Union. That means even employers without a physical presence in the European Union may have compliance obligations under the AI Act if, for example, they make job postings available to EU candidates or if they have independent contractors or contingent workers in the European Union.
Also, much like the European Union’s General Data Protection Regulation (GDPR), there are significant penalties for noncompliance. The AI Act’s penalties range from €7.5 million (approximately USD $8 million) or 1.5 percent of a company’s total worldwide annual turnover (whichever is higher) to €35 million (approximately USD $38 million) or 7 percent of a company’s total worldwide annual turnover (whichever is higher).
A Model for Other Countries
The AI Act is, as it was intended to be, a comprehensive framework for AI, regulating AI in everything from policing to commercial products to education and employment, and everything in between. The legal community and policy scholars agree that it is likely to be to AI what the GDPR was to data privacy.
Employers using AI tools, and particularly those with cross-border operations, may wish to follow these developments and evaluate proactive compliance efforts.
CompAnalyst® Pay Equity Suite can help you achieve and sustain pay equity