Artificial Intelligence Act

Publication

The report presents the findings and recommendations of the first part of the Open Loop’s policy prototyping program on the European Artificial Intelligence Act (AIA), which was rolled out in Europe from June 2022 to July 2022, in partnership with Estonia’s Ministries of Economic Affairs and Communications and Justice, and Malta’s Digital Innovation Authority (MDIA).

We enlisted 53 AI companies to participate in the Open Loop Forum (OLF), a dedicated online platform where they met to discuss topics and complete several research-related tasks.

The overall picture is that most of the AIA provisions addressed in this program are clear, feasible and may contribute to the overall goal of creating trustworthy AI. However, there are several areas in the AI Act where there is room for improvement, and some provisions that might even undermine another goal of the legislator: the uptake of AI in Europe.

Publication

In this report, we explore the efficacy of the taxonomy of AI actors in the EU Artificial Intelligence Act (AIA) (e.g., provider, user, and importer), proposing an alternative for the taxonomy of AI actors currently included in the proposal. This research is part of the Open Loop Program of the EU AIA.

The question we pose is whether the taxonomy of AI actors in the AIA is effective and, if not, what an alternative taxonomy would look like. Our hypothesis is that the current taxonomy of AI actors in the AIA does not accurately reflect the AI market, and this may lead to issues in assigning responsibilities for market actors and apportioning liability.

To test our hypothesis and address our research questions, we surveyed AI companies in our Open Loop Forum (OLF), conducted expert interviews, and performed desk research.

Based on the information gathered, we conclude that the existing taxonomy does not accurately reflect the actors in the AI ecosystem. In particular, roles such as the subject, third-party service providers, and data providers seem to be missing from the AIA’s text.

Publication

This report, which is part of the Open Loop Program on the EU Artificial Intelligence Act (AIA), explores the AI regulatory sandbox provision described in article 53 of the AIA.1

More specifically, we exploredthe goals of the AI regulatory sandbox and the conditions necessary to achieve them. In particular, we sought answers to the following research questions (RQs):

• RQ1: What are the objectives of the EU AI regulatory sandbox?
• RQ2: What conditions are necessary to achieve the objectives of the EU AI regulatory sandbox?
• RQ3: Does article 53 of the AIA enable the necessary conditions for a successful EU AI regulatory sandbox?
• RQ4: Are there alternative governance mechanisms to achieve the objectives of the EU AI regulatory sandbox?

To answer the RQs, we collected data from three different sources: desk research, interviews with experts, and a Sandbox Policy Design Jam. This mixed-method approach allowed us to triangulate the data and address the four RQs from various perspectives.

Publication

As part of the Open Loop program on the Artificial Intelligence Act, we tested one of the requirements of Article 52(a) (on transparency obligations for AI systems interacting with individuals) of the proposed regulation to assess when and how individuals should be informed when they are interacting with an AI system.

We conducted an online survey with a sample of 469 participants from five European countries (Spain, France, Germany, United Kingdom and Sweden). Via this online survey, participants were exposed to videos of two different AI-powered systems, a chatbot and a news app. Different styles of AI notifications were presented to the participants:

• no notification
• content-integrated notification and
• notification banner

The survey showed that 30% of respondents did not notice the banner notification, and 49% failed to notice and comprehend the content-integrated notification. People with prior experience with AI systemswere more likely to perceive and understand notifications. Participants’ understanding of a notification did not significantly affect their sense of control and trust in the tested AI applications

Publication

This report, which is part of the Open Loop Program on the EU Artificial Intelligence Act (AIA), presents the findings of a policy prototyping exercise on risk management and transparency in the AIA.

The objective of this Deep Dive was to assess the clarity and feasibility of selected requirements from the perspective of participating AI companies.

Valuable insights gathered from participating companies have highlighted areas that require improvement, which are discussed in two parts:

1. i) transparency and human oversight and
2. ii) risk management requirements in the AIA.