The EU Artificial Intelligence Act is set to become the world's first comprehensive legal framework for artificial intelligence. This overall policy was initially proposed by the European Commission in April 2021, endorsed by the European Council last year, and approved by the European Parliament in mid-June. Now, before the policy becomes law, there will be final negotiations and discussions on the details among the three institutions involved, a process known as trilogue negotiations. Once a final agreement is reached, the law could come into effect in the next few years.
Stanford Institute for Human-Centered Artificial Intelligence (Stanford HAI) convened experts in artificial intelligence, law, and policy to discuss the details of the regulation, areas where the three EU institutions may still be negotiating, gaps that need to be addressed, and the extent to which tech companies are likely to comply with the regulation. The discussion was moderated by Marietje Schaake, International Policy Director at Stanford HAI and former Member of the European Parliament. You can watch the full expert discussion here.
Where are the areas of disagreement between the legislative bodies?
Dragoř Tudorache, Member of the European Parliament and rapporteur for the Committee on Civil Liberties, Justice and Home Affairs, highlighted several areas where there may be disagreements during the trilogue.
1. Biometric surveillance using artificial intelligence in public spaces. The Parliament has introduced a ban on law enforcement use, which may face opposition from member state governments.
2. Definition of high-risk artificial intelligence. The Parliament allows for a broader definition of use cases for high-risk AI, while the Council is more inclined to narrow down the definition.
3. Governance. In addition to other matters, the trilogue will need to discuss the level of fragmentation and coordination among countries in implementation and enforcement.
He noted that other areas of disagreement include regulation of generative AI and assessment of high-risk applications after deployment. Alex Engler, a researcher at the Governance Studies department of the Brookings Institution, focusing on the impact of AI and emerging data technologies on society and governance, added that there will be further discussions on technical issues: which types of institutions should be regulated, whether existing institutions should regulate their own expertise (e.g., educational institutions handling educational algorithms), or whether a central organization should be established. He said these technical issues will need to be "addressed first, as they are at the core of the law."
How much will companies comply with the proposed regulations?
Rishi Bommasani, Social Leader at Stanford HAI's Center for Human-Centered Artificial Intelligence and a Ph.D. candidate in computer science, researched how foundational model providers such as OpenAI, DeepMind, and Anthropic comply with the draft legislation.
Areas where they fall short of the standards include failure to provide summaries of copyrighted data used for training models, imbalanced reporting on energy usage, insufficient disclosure of risk mitigation, and lack of evaluation criteria or audit procedures. Foundational model developers taking an open-source approach tend to perform better in terms of resource disclosure requirements, while closed-source or restricted models perform better in terms of deployment-related requirements.
What do tech companies need from policymakers to comply?
Companies need further clarification in three areas, according to Irene Solaiman, Policy Director at Hugging Face, who conducts research on social impact and leads public policy efforts.
1. Transparency: Legislators must define legal requirements for transparency and how to document datasets, models, and processes.
2. Model access: Researchers need access to systems for evaluation and building safeguards, regardless of whether the models are open-source or closed-source.
3. Impact assessment: There are currently no standardized assessments in the research community. There is no good way to robustly evaluate one aspect of bias in language models.
How will the EU's plans impact the United States?
"There is an increasing divergence between the approaches of the United States and the European Union in regulating artificial intelligence," Engler said. The EU has been moving forward with regulations in areas such as data privacy, online platforms, and e-commerce, while the United States lacks similar legislation.
He warned that the passage of the EU legislation will make it more difficult for the United States to pass its own laws - companies do not want to face two sets of different rules in two different markets. "If there are two very different standards for online platforms, corporate interests will fight tooth and nail."
Engler suggested that US lawmakers familiarize themselves with the EU legislation and seek to find common standards to ensure ease of implementation.
Tudorache added that he has noticed an increasing interest from policymakers in Washington, D.C., in regulation and added, "If we stay aligned on standards, if we stay aligned on principles, even if there are differences in legislative details, I think we have a good chance of achieving convergence between the United States and the European Union."
Source: https://techxplore.com/news/2023-07-european-union-ai.html
Alphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Anthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Apple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Google launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
YouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Google grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Mistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.