In the EU, the European Parliament has adopted its final version of the AI Act. It includes a prohibition on real-time remote biometric identification and a tiered approach for general purpose AI, with stricter obligations for the foundation layer, and labelling and disclosure obligations for the application layer. The Verge examines these obligations in detail.

A Time investigation has revealed that the CEO of OpenAI lobbied for general purpose AI systems such as ChatGPT to not be designated as ‘high risk’, despite publicly advocating for global AI regulation. It also revealed that OpenAI proposed several amendments that appear in the European Parliament version of the Act.

The European Commission’s Digital Europe Programme has launched four testing and experimental facilities ("TEFs") to test the risks and impacts of AI applications before they are launched. The TEFs have been introduced to support the EU’s Data Act which calls for regulatory sandboxes. Each TEF has a specific focus: machine learning software and robotics, health and patient data, food production, and power and mobility in cities.

The European Union and the US have published the EU-US Terminology and Taxonomy for Artificial Intelligence. The document contains definitions of key terms relating to AI and governance.

In the UK, the Department for Science Innovation and Technology has launched a Portfolio of AI Assurance Techniques. The portfolio features a range of case studies of AI assurance techniques. Organisations can submit case studies to the portfolio at ai.assurance@cdei.gov.uk.

The national funding agency UK Research and Innovation has invested £31 million in a new consortium, Responsible AI UK. The consortium aims to create a UK and international research and innovation ecosystem for responsible and trustworthy AI, and is led by the University of Southampton.

Big Brother Watch and AWO’s complaint to the ICO about British supermarkets’ use of Fcaewatch facial recognition technology to combat shoplifting resulted in changes to Facewatch’s policies. Changes include more signage, only sending alerts about repeat offenders, and only sharing information about serious and violent offenders with other stores.

In the US, the Senate majority leader Chuck Schumer has introduced the SAFE Innovation Framework for AI. The Bill aims to promote accountability, introducing explainability obligations; protect employment; and secure against cyber attacks. A Time article critiques this approach and the technical feasibility of AI explainability.

Senators also plan to introduce a bipartisan bill to clarify that generative AI will not benefit from legal immunity under Section 230 of the Communications Decency Act.

The Federal Trade Commission has published guidance on the competition concerns of generative AI. The guidance explains the essential technical building blocks of generative AI and examines the competition concerns which include incumbents leveraging control of inputs and adjacent markets such as cloud computing.

The National Artificial Intelligence Advisory Committee (NAIAC) published its first-year report. Several members of the Committee published their perspective on the report, arguing it missed the opportunity to articulate a people-first, rights-respective AI strategy. The NAIAC also hosted expert briefing sessions with civil society groups.

Mozilla has launched the Open Source Research and Investigations team to investigate platform algorithms using crowdsourced data. The team will initially investigate TikTok’s ‘For You’ page algorithm and platform integrity amid elections outside the U.S. and in non-English speaking countries.

OpenAI has been sued in two class action lawsuits. The first alleges OpenAI violated privacy rights when it scraped data from the internet to train ChatGPT. The second alleges OpenAI infringed copyright law when it scraped data from online book excerpts without the authors’ consent.

In Chile, the Government will introduce the General Instruction on Algorithmic Transparency. This law will mandate public bodies to comply with the algorithmic transparency standard. The Observatory of Public Sector Innovation explores the development of this standard.

The Ada Lovelace Institute has published Expert explainer: Allocating accountability in AI supply chains. This paper aims to help policymakers and regulators understand the challenges and nuances of different supply chains, including systems built in-house, systems relying on another API, systems built for a customer and foundation models.

The ACM Conference on Fairness, Accountability and Transparency was hosted in Chicago and six papers won Best Paper awards:

• Enhancing AI fairness through impact assessment in the European Union: a legal and computer science perspective
• Queer in AI: A Case Study in Community-Led Participatory AI
• Preventing Discriminatory Decision-Making in Evolving Data Streams
• The Privacy-Bias Tradeoff: Data Minimisation and Racial Disparity Assessments in U.S. Government
• On the Impact of Machine Learning Randomness on Group Fairness
• Counterfactual Prediction Under Outcome Measurement Error

Hertie School professor Joanna Bryson and Meeri Haataja have published The European Parliament’s AI Regulation: Should We Call It Progress? This article examines the changes the European Parliament has made in its’ final version of the AI Act.

The IEEE Spectrum has published The Who, Where, and How of Regulating AI. This article examines the variety of approaches to AI regulation in different jurisdictions, including the EU, UK, China.

Princeton Computer Science professor Arvind Narayanan and researcher Sayash Kapoor have published Generative AI companies must publish transparency reports. This blog argues that transparency reports are essential to enable informed debates about AI harms. It argues that these reports are technically feasible and explores the information that should be included.

Stanford University’s Center for Research on Foundation Models has published Do Foundation Model Providers Comply with the Draft EU AI Act?  In this article, Stanford computer scientists evaluate foundation models against select requirements in the draft AI Act. Hugging Face performs best in the assessment.

The American Civil Liberties Union is hiring an Undergraduate / Graduate Intern focused on Algorithmic Justice for the Legal and Quantitative Research Pod. The pod conducts quantitative analysis, builds data tools and visualises data. Candidates can be fully remote or hybrid.
Applications are rolling with a priority deadline of 09 July.

The Ada Lovelace Institute is hiring researchers to support work on the impacts of AI and data on people and society. The roles are London-based.
Application deadline is 19 July, 09:30 GMT

The UK's Centre for Data Ethics and Innovation is calling for organisations to share use cases on fairness-related problems faced when designing, developing and deploying AI systems. This is for the Fairness Innovation Challenge, which aims to support the development of solutions that address bias and discrimination across the AI lifecycle.

Regulating AI in the UK
Virtual: 18 July, 10:30 - 11:30 GMT
The Ada Lovelace Institute and AWO will present analysis of the UK Government’s AI White Paper and recommendations on how to improve it. This is based on the policy briefing Regulating AI in the UK: three tests for the Government’s plans.

Q: What is the Pôle d’Expertise de la Régulation Numérique (PEReN) and why was it created?
PEReN: PEReN is the centre for expertise for digital platform regulation of the French government. We provide technical assistance to public services on digital regulation in the fields of data processing, data science and algorithmic auditing. PEReN is placed under the joint authority of the Ministry of Economy, the Ministry Culture and the Ministry of Digital Technology, and was created to equip regulators and independent authorities with the technical support that is nowadays necessary to regulate online platforms.

We use two different legal frameworks to conduct our work:

• We assist governmental services with regulatory powers. When we conduct this work we inherit the legal framework from the authority or administration in question, particularly concerning the authorization to collect data or to request assistance of external agents.
• We provide expertise in research work by conducting exploratory or scientific studies. We do this through ongoing collaboration with French research centres.

Previously, we have offered technical solutions for investigations. For example, we have built user-friendly tools for:

• Natural language processing and scraping;
• The audit of delivery platforms;
• The detection of deep fakes.

We have also completed studies on:

• Google’s auto-completion algorithm;
• Multimodal models and their applications;
• Large language model BLOOM and its ability to become a hateful-speech detector for French;
• Differences on Amazon platforms across member states.

Q: Could you tell us more about PEReN’s collaboration with DG CONNECT and ECAT?
PEReN: We are partnering with DG CONNECT and the European Centre for Algorithmic Transparency (ECAT) to work on DSA-related topics. DG CONNECT first reached out to us when it was in the process of creating the ECAT. DG CONNECT wanted to learn from PEReN’s experience as a technical research centre working on digital platform regulation. Both ECAT and PEReN share the same goal of auditing online platforms but we act at different scales, the European and French national level, respectively. We are excited to be working with ECAT, and would be also interested to collaborate with other European centres that are similar to PEReN.