Meta reached a settlement that could cost it about $17 billion, putting an end to a significant part of the litigation concerning the effects of Facebook and Instagram on minors. The agreement requires changes to the platforms, including usage limits, nighttime blocking, notification restrictions, and the option for a non-personalized feed. The author analyzes this case as an example of one of the effects of algorithms on social media: the capture of attention. The algorithms learn from users' behavior, but knowing attention is not knowing will. They may read behaviors as preferences and give more of the same.
The problem extends far beyond social media. Algorithms already support clinical diagnoses, assess credit risk, select job candidates, adjust prices, detect fraud, and establish priorities in security systems. It is in the translation of legitimate objectives into data, criteria, and metrics that legal problems can arise. A recruitment algorithm can reproduce past discriminations, a credit model can disadvantage certain groups, and clinical systems can be less reliable for populations poorly represented in the data.
The author argues that intention does not solve the problem: an algorithm can fulfill exactly the purpose for which it was created and produce legally problematic effects that no one intended. As algorithms enter the ordinary operations of companies across all sectors, the possibilities for errors, discriminations, or other unforeseen effects on rights, property, health, opportunities, or safety increase. The use of algorithms is also a risk management problem, requiring companies to explain anticipated risks, data and criteria used, tests performed, and mitigation mechanisms.
The AI Act seeks to anticipate some of these problems, but the comparison with the United States is misleading. In the United States, regulation also occurs through regulators, courts, civil liability, and settlements that impose concrete changes to products. The author argues that the next major wave of mass litigation will revolve around algorithms, and that in Portugal, where class actions favor collective litigation, a design error can quickly turn into a lawsuit against a company on behalf of thousands of people. He concludes that many of the decisions that will tomorrow have to be explained to a regulator or court are today being made by those who design the algorithms.
