Europe is preparing a dedicated research team to help regulators examine how major online platforms use algorithms. The European Centre for Algorithmic Transparency, or ECAT, is part of the European Commission’s Joint Research Centre and has a specific role supporting oversight under the Digital Services Act.
Research designed to support oversight
ECAT’s job is to investigate algorithmic systems and help the Commission assess whether platforms are meeting their obligations. The research is tied to enforcement: evidence gathered by the team could help regulators evaluate possible breaches of the new digital rules.
That includes checking claims platforms make about the steps they have taken to reduce risks. For example, researchers may examine whether a recommender system continues to serve discriminatory content even after a platform says it has worked to reduce bias. The aim is to give regulators evidence they can use to assess those claims.
The Centre sits within the Joint Research Centre, an established EU science facility that supports policymaking across areas including climate change, crisis management, taxation and health sciences. ECAT is expected to draw on expertise across that wider research organization while focusing its work on digital regulation.
What large platforms may have to show
The Digital Services Act and its sister regulation, the Digital Markets Act, were adopted last year. The article reports that the rules were due to come into force in the coming months, while the DSA’s full set of provisions was not expected to be enforced until early 2024.
A subset of very large online platforms and very large online search engines would face special oversight sooner. The Commission said it would designate the platforms “very soon.” The article noted a DSA criterion of 45 million+ regional users, but said the exact list had not yet been confirmed.
Once designated, platforms would have four months to comply with their obligations, including submitting their first risk assessment reports. They would need to assess systemic risks connected to their algorithms, take steps to mitigate those risks and allow regulators to scrutinize what they say they have done.
The risks described in the article include the spread of disinformation and illegal content, and possible effects on freedom of expression and users’ fundamental rights. Privacy and child safety are among the issues regulators may consider. The DSA also places limits on profiling-driven feeds and the use of personal data for targeted advertising.
Testing systems through practical research
At its launch, ECAT staff described projects already under way. These include examining racial bias in search results, studying how voice assistants can respond appropriately to children’s vulnerability, and testing social media recommendations with profiles that have different patterns of likes.
Other areas of early research include facial expression recognition, algorithmic ranking and pricing. The team has also built a data analysis tool to help the Commission review the risk assessment reports that designated platforms must submit. Such reports could contain large amounts of material, making it harder to identify information relevant to regulators’ questions.
ECAT’s work is not limited to near-term enforcement. Researchers also plan to study the longer-term effects of people’s interactions with algorithmic technologies, including topics connected to gender-based violence, child safety and mental health. Understanding those effects can involve more than examining computer systems alone.
For that reason, the Centre plans to bring together computer and data scientists, social and cognitive scientists, and other researchers. Its stated approach is to apply different kinds of expertise to understanding how algorithms operate and affect people.
A research network beyond the Commission
ECAT also hopes to make its research accessible to the public and work with the wider European research community. Its staff described an ambition to connect research institutions, civil society groups and others so more people in the region can study algorithmic effects.
One early partnership is with France’s PEReN, a research organization that supports national policymaking and regulatory enforcement. At the launch, PEReN described testing how quickly TikTok recommendations respond when a user’s interests change, using a profile that switched from watching cat videos to videos of trucks.
The broader test for ECAT will be whether its research can help regulators move from platform assurances to evidence they can examine. The Centre is being set up as part of a system in which large platforms must assess risks and explain their actions, while researchers help the Commission scrutinize those explanations.
The article also reports that the planned team would include 30 to 40 people, with some 14 hires made so far and over 500 applications received following job ads last year. Funding was to come from the Joint Research Centre’s existing budget, with a 1% supervisory fee on very large platforms and search engines intended to cover staff costs as that mechanism is established.