Sarah Cameron Pinsent Masons’ Luke Scanlon and Luke Scanlon made the comments after the so-called tripartite negotiations over the AI law began on Wednesday. The consultations involved representatives from three EU bodies – the European Parliament, the Council of Ministers and the European Commission – each with different views on the scope and content of the bill.
A proposal for a new EU AI law was formulated by the European Commission in April 2021. The European Commission has proposed regulating AI according to the level of risk the system is believed to pose to people. The plan prohibits AI that poses an “unacceptable risk” to people, while ensuring the quality, record-keeping, transparency and user-information of the datasets used, human oversight, robustness, accuracy, cyber security. “Low-risk” AI systems will be subject to limited transparency obligations.
Congress and the Board have since scrutinized the plan over the past two years and adopted their own negotiating positions prior to tripartite talks. The aim is to reach a final text that can be adopted into EU law.
The council adopted its position late last year, and on Wednesday parliament approved a proposal put forward by the Internal Market Committee and the Human Rights Commission last month, paving the way for tripartite talks to begin.
Prime Ministers Cameron and Scanlon said their respective positions on the Council and Parliament were very different and largely reflected late amendments proposed by MEPs earlier this year to reflect their views on rapidly advancing technological developments. said that
In its proposal, Congress also regulates not only AI systems, but “foundation models.” The term is defined by MEPs as AI models that are trained on large-scale extensive data and designed for versatility in their outputs. Adaptable to a wide range of specific tasks. Providers of “generative underlying models” used in AI systems specifically for content generation are subject to certain obligations. This includes the need to provide transparency when content is created by AI systems rather than humans. Provide the public with a sufficiently detailed overview of your use of copyrighted training data.
The basic model concept has not been addressed in the Council’s negotiating position.
“Underscoring the challenges of the Act’s prescriptive risk-based approach, the Foundation model has not been assigned to the high-risk category,” Cameron said. Instead, the new draft imposes specific obligations, such as transparency requirements on training data and designing models to avoid generating illegal content. ”
“The risks posed by generative AI have also raised significant concerns among AI advocates. The real concern about and generative AI could be to address risks and allow for rapid adjustments to avoid a loss of public confidence in AI as a whole, Rishi Sunak will host an AI summit later this year. Looking at the plans to host it, it’s clear that the UK wants to be at the center of this discussion,” she said.
Mr Scanlon said: “The EU’s position on formal recognition, conformity assessment and now on transparency regarding the data sources used for the underlying models is gaining traction in the international debate. It is not clear whether the approach will be taken: large-scale language models and generative AI “raw materials” are readily available in almost all jurisdictions, so the development or use of AI in the EU is / Or international inconsistencies regarding the licensing process could have serious unintended consequences. About our efforts to promote responsible and safe use of AI and economic development. ”
“When it comes to the EU legislative process, it is difficult how quickly the approach put forward by parliament can be reconciled with that of the Council, especially with respect to the foundation model,” he said.
