小程序
传感搜
传感圈

EU AI Act should ‘exclude general purpose artificial intelligence’ – industry groups

2022-10-05
关注

Ten European software industry associations have called on the EU to scrap plans to include the regulation of general purpose AI including natural language processing and chatbots in its new AI Act, describing it as a “fundamental departure from its original objective” and saying that it could stifle innovation and hit the open source community.

Artificial intelligence regulations are being developed around the world and the EU plans to include general purpose AI (Photo: DedMityay/Shutterstock)
Artificial intelligence regulations are being developed around the world and the EU plans to include general purpose AI. (Photo by DedMityay/Shutterstock)

The European Union AI Act aims to establish a framework to regulate the use of artificial intelligence, taking a “risk-based” approach to its use and establish a worldwide standard. The Act includes core provisions including tighter regulations in high-risk areas such as healthcare and transparency requirements, focusing on specific-purpose narrow AI.

However, the group of industry associations, led by BSA, the software alliance, has published a joint statement urging EU institutions to reject recent additions to the Act that include regulation of general purpose AI and instead “maintain a risk-based approach”.

The objective of general purpose AI is to create machines that can reason and think like a human. It has no set or clear path or pattern to follow and includes chatbots using natural language processing and driverless vehicle systems.

Companies Intelligence

View All

Reports

View All

Data Insights

View All

Fundamental departure from objective

In May the French presidency of the Council of the EU announced plans to include this in the scope of the Act, requiring providers to comply with risk management, data governance, transparency and security articles of the Act, where previously it was excluded.

“This inclusion would be a fundamental departure from the original structure of the AI Act, by including non-high-risk AI in the scope of the regulation, against all objectives set out by the European Commission,” the group behind the statement wrote.

The ten associations from across the EU call on policymakers to ensure the AI Act maintains a risk-based approach to regulation, excluding general purpose AI.

“The businesses we represent have been supportive of the objectives of the AI Act, in particular the goal to create a balanced and proportionate regulatory approach,” the statement says.

Content from our partners

How to integrate security into IT operations

How to integrate security into IT operations

How Kodak evolved to tackle seismic changes in the print industry and embrace digital revolution

How Kodak evolved to tackle seismic changes in the print industry and embrace digital revolution

Why all businesses must democratise data analytics

Why all businesses must democratise data analytics

The associations claim that the inclusion of general purpose AI would overturn this risk-based approach and “severely impact open-source development in Europe” as well as “undermine AI uptake, innovation, and digital transformation”.

Data, insights and analysis delivered to you View all newsletters By The Tech Monitor team Sign up to our newsletters

“For these reasons, we strongly urge EU institutions to reject the recent proposals on general purpose AI and ensure that the scope of the AI Act maintains its risk-based approach and a balanced allocation of responsibilities for the AI value chain, for a framework that protects fundamental rights, while addressing the challenges posed by high-risk AI use-cases and supports innovation.”

By including general AI, the group say the Act would subject low-risk AI systems to heavy scrutiny, simply because there is no clear definition of how they will be used.

“As a result, the AI Act would no longer regulate specific high-risk scenarios, but a whole technology regardless of its risk classification,” the group wrote in a paper on the issue.

It cites the example of a developer that has designed a general purpose AI to read documents ranging from university transcripts to government forms and extract relevant information for a specific use. The system is designed to be further customised and the developer would have to comply with a pre-emptive risk assessment of a diverse and different range of sectors and future use cases.

That same developer would also have to continuously monitor the operation of the tool, creating possible GDPR conflicts as it could also be used to extract personal data.

Unforeseen outcomes of changes

The changes to the Act would “run counter to the commission’s own assessment—and that of stakeholders who provided comments on the AI Act—on which option for legislation would better serve the EU for supporting the development of trustworthy AI,” the statement continues.

What’s more, the definition of AI in the amended legislation includes some software products not traditionally considered AI, according to the group, including APIs and tools used to develop AI systems.

“One thing the AI Act would benefit from is more clarity on the allocation of responsibilities between AI developers and deployers, ensuring that compliance obligations are assigned to the entities best placed to mitigate challenges and concerns and encouraging coordination along the AI value chain,” Matteo Quattrocchi, BSA policy director, told Tech Monitor.

However, not everyone agrees with the group’s assessment. Adam Leon Smith from Dragonfly, a UK industry representative to the EU AI standards group, said general purpose systems need to be included in the legislation due to potential unintended consequences.

“Such components can be the cause of the harm the AI Act is intended to prevent,” Leon Smith said. “In fact, the risk is arguably greater if the AI component is being used in a way that the developers did not envisage. 

“The latest draft proposals include a get-out clause for developers, they simply need to include a usage limitation in the instructions for use.  Frankly, it would be unusual to download open source software without legal documentation already being included – so this is an extra line or two in there – not a huge burden upon developers.”

Read more: UK government sets out AI regulation plans

Topics in this article: AI, EU

参考译文
欧盟人工智能法案应该“排除通用人工智能”——产业团体
10个欧洲软件行业协会呼吁欧盟放弃将自然语言处理和聊天机器人等通用人工智能监管纳入其新AI法案的计划,称这“从根本上背离了其最初的目标”,并表示这可能扼杀创新,打击开源社区。《欧盟人工智能法案》旨在建立一个监管人工智能使用的框架,采取“基于风险”的方法来使用人工智能,并建立一个全球标准。该法案包括核心条款,包括在医疗保健等高风险领域加强监管和透明度要求,重点关注特定目的的狭义人工智能。然而,由软件联盟BSA领导的行业协会团体发表了一份联合声明,敦促欧盟机构拒绝该法案最近增加的内容,其中包括对通用人工智能的监管,而是“保持基于风险的方法”。通用人工智能的目标是创造能够像人类一样推理和思考的机器。它没有固定或明确的路径或模式可循,包括使用自然语言处理的聊天机器人和无人驾驶汽车系统。今年5月,欧盟理事会轮值主席国法国宣布计划将这一点纳入法案的范围,要求提供商遵守法案的风险管理、数据治理、透明度和安全条款,而此前这些条款被排除在外。该声明背后的团体写道:“这一纳入将从根本上背离人工智能法案的原始结构,将非高风险人工智能纳入监管范围,违背了欧盟委员会制定的所有目标。”来自欧盟各地的10个协会呼吁政策制定者确保《人工智能法案》保持基于风险的监管方式,将通用人工智能排除在外。该声明表示:“我们所代表的企业一直支持《人工智能法》的目标,特别是创建一个平衡和适当的监管方法的目标。”这些协会声称,通用人工智能的加入将颠覆这种基于风险的方法,并“严重影响欧洲的开源开发”,以及“破坏人工智能的吸收、创新和数字转型”。基于这些原因,我们强烈敦促欧盟机构拒绝最近关于通用人工智能的提案,确保《人工智能法》的范围保持基于风险的方式,并平衡分配人工智能价值链的责任,建立一个保护基本权利的框架,同时应对高风险人工智能用例带来的挑战,并支持创新。该组织表示,通过将通用人工智能包括在内,该法案将使低风险的人工智能系统受到严格审查,原因很简单,因为没有明确的定义,它们将如何使用。”因此,《人工智能法案》将不再监管特定的高风险场景,而是监管整个技术,而不管其风险分类如何,”该组织在一份有关该问题的论文中写道。它引用了一个开发者的例子,该开发者设计了一个通用AI来读取从大学成绩单到政府表格的各种文件,并为特定用途提取相关信息。该系统被设计为进一步定制,开发商将必须遵守对不同行业和未来用例的不同范围的先发制人的风险评估。同一名开发人员还必须持续监控工具的操作,这可能会产生GDPR冲突,因为它也可能被用于提取个人数据。该声明继续说,对该法案的修改将“与委员会自己的评估——以及对AI法案提供意见的利益相关方的评估——背道而驰,即哪一种立法选项将更好地服务于欧盟,支持可信赖的AI的发展。” 此外,据该组织称,修订后的立法中对人工智能的定义包括一些传统上不被认为是人工智能的软件产品,包括用于开发人工智能系统的api和工具。BSA政策总监Matteo Quattrocchi告诉Tech Monitor:“《AI法案》将受益于AI开发者和部署者之间的更明确的责任分配,确保合规义务分配给最能减轻挑战和担忧的实体,并鼓励AI价值链上的协调。”然而,并不是所有人都同意该组织的评估。欧盟人工智能标准组织的英国行业代表、来自蜻蜓(Dragonfly)的亚当•利昂•史密斯(Adam Leon Smith)表示,由于可能产生意想不到的后果,需要将通用系统纳入立法。“这些组件可能是《人工智能法案》旨在防止的伤害的原因,”里昂·史密斯说。“事实上,如果AI组件的使用方式超出了开发者的想象,那么风险可能会更大。“最新的提案草案包括了对开发商的免责条款,他们只需要在使用说明中包括使用限制。坦率地说,下载没有法律文档的开源软件是不寻常的——所以这是额外的一两行——对开发人员来说不是一个巨大的负担。”
您觉得本篇内容如何
评分

评论

您需要登录才可以回复|注册

提交评论

techmonitor

这家伙很懒,什么描述也没留下

关注

点击进入下一篇

燧原科技打造一站式人工智能算力中心

提取码
复制提取码
点击跳转至百度网盘