Does AI need rules? – Thoughts on the proposed EU AI Act
Does AI need rules? What is the ‘AI’ the EU is aiming to regulate in its new proposal? How does the regulation affect innovation? What could be the implications internationally and nationally for Finland when EU moves forward with AI regulation?
These questions were in the centre of a panel discussion held in the Think Corner on 30 September 2021. The panel included experts from various relevant fields: Petri Myllymäki from the tech field, Suvi Sankari from the legal field; Olli-Pekka Rissanen with a background in national politics and policy, and Miapetra Kumpula-Natri with experience in EU level policy and law-making. The panel was led by journalist Elina Lappalainen. This discussion was the final instalment of a wider, weeklong program called ‘People’s AI’, which was conducted in Finnish and considered different angels of AI.
In this blog contribution, we summarize the key points of the discussion. As the discussion largely focused on the EU’s new AI regulation proposal, it will be the core here as well. What makes the proposal topical is that currently there are no AI specific laws in Finland nor in the EU. If and when the AI regulation proposal is accepted, it will have vast implications on the Member States and the society as a whole. Therefore, we have collected some key points of the panelists, which help to understand the topics more broadly.
Artificial Intelligence, what now?
A well-known fact, at least for the technology experts, is that AI is not just one thing and talking about it as a singular unit is problematic. It is a complex field of vast ranging technologies as Myllymäki reminded, and defining it is as difficult as defining (human) intelligence. Due to the concept of AI, it is almost impossible to pin down what exactly makes ‘a thing’ AI and experts sometimes disagree between themselves whether a certain technology is AI or not. As AI is a complex set of vast range of technological systems, issues with definitions surface. When the definitions of the objects of the regulation are not clear, then the scope of application is equally unclear; point mentioned by Sankari and Rissanen. Furthermore, according to Myllymäki, regulating AI as distinct technology may be difficult due to the ability to use the same technology for good and bad. Although the starting point for the conversation was that AI is difficult to define, everyone in the panel agreed that it is a good thing that there is will to discuss the use of AI and its regulation.
EU’s AI regulation would be the first AI regulation in the world. Whether the will to regulate surfaced due to the growing fears generated by popular media or something else, the fact that this is on the table is a positive thing. Along with the advantages of the use of AI there are negative aspects as well, and it is important that those risks are acknowledged and mitigated. The increasing use of AI in all spheres of peoples’ lives gives rise to the need to also contain citizens’ fears related to it as well as to support the beneficial use of AI. Further, the importance of keeping EU’s AI markets competitive and support innovation shine through the proposed regulation. AI has reached the point where it is easier to be included in different types of applications and systems, which makes it prone to ever-increasing use. Thus, assessing the benefits and risks of AI is urgent.
The aim of the regulation is for the same rules to apply across all EU Member States. The EU sees a future of only one common AI market for the whole EU. It should be kept in mind, however, that this is only a proposal. It is still in the first round of hearings and the Member States as well as the EU institutions are continuing to negotiate on the substance of the proposal. In the European Parliament, as well, conversations are ongoing regarding the different aspects that should be taken into account, such as the risks the regulation would pose on restricting innovation. There is time to influence the regulation and join in the discussion on what the future of AI regulation should look like. You can find a summary of the proposal here.
Great intention but what about the execution?
The proposal was viewed largely as good when it concentrates on the intended uses of AI. Moreover, its objectives and goals were seen in positive light: upholding the EU’s democratic values and the prospect of EU going values first was appreciated. Kumpula-Natri pointed out the overarching reason for the regulation from the EU is to uphold the democratic values and human rights within the Union. The goals, principles and values are noble, everyone agreed.
Moving outside the normative value discussions, upon which everyone agreed, questions arose regarding the orientation of the proposal. The uncertainty arising from the definitions have a wider effect on the uncertainty of the scope of application. As Myllymäki put it, definitional uncertainties, and the resulting uncertainties related to the proposal’s scope, beg the following question: why would systems, such as social scoring systems, be prohibited only when AI is used in them? Discriminatory technological systems can be created also without AI. Furthermore, the same AI technology can be used for good as for bad. When the proposal focuses on specific technologies that form part of the products the proposal aims to regulate, it creates uncertainty as to what is, in fact, regulated. As mentioned, the definition of AI is not clear. Myllymäki explained that experts occasionally disagree whether a given system is AI or not. Therefore, focusing on the intended uses of AI would clarify the matter to some extent, and not leave the legal issues stuck on technological question. Further, this approach, focus on the intended uses, would potentially provide better grounds for tackling and upholding the normative goals.
However, not all participants agreed with the above-mentioned uncertainties. Kumpula-Natri saw this as a regulation of the intended uses of AI and stated that the AI in the title is primarily a political term. This, according to her ensures that politicians are able to discuss AI in laymen’s terms and participants not need to have deep, technological expertise on the subject. She considered the discussion on what AI is unnecessarily long and further reflected that the issue has been well resolved in the proposal by the list of AI systems, which can later be amended by the Commission. While the use of annexes and delegated acts bears the risk of complexity and even confusion on the legislation, on the other hand, it also helps to keep the legislation up to date and provides the Commission the possibility to modify it later. The possibility of adopting delegated provisions entails also other aspects, which at this time were not in the focus of the conversation.
The issues with definitions and scope of application have also been noted in the Finnish Government’s Union communication to the Parliament. According to Rissanen, more work on understanding the actual meaning of the proposal and its possible effects is needed. The complicated structure of the proposal, its several interfaces with other legislation and challenges described earlier in this blog entry make this undertaking demanding. More generally, Sankari pointed out that the ensuing burden on national legislative work required as well as the resources needed for product market surveillance had not really been discussed yet. Perhaps most problematically, the proposal builds on curing AI ills with for example algorithmic transparency, which research has already called into question (see for example University of Helsinki Legal Tech Lab’s project ALGOT - Potential and boundaries of algorithmic transparency).
From the legal point of view, since the proposal is unusually wide reaching horizontally, as is the use of AI, it would touch upon many substantive areas. Thus, Sankari pointed out that for example, lawyers would need to have expertise on broad spectrum of substantive fields in order to understand and be able to apply the law. Further legal questions could arise, including the uncertainty regarding any space to regulate AI on national level. Handed out as regulation, the law would leave no or possibly only minimal space for Member States to regulate the use of AI. Sankari asked that if the aim of the regulation is to protect people, the lack of consideration in the proposal of how individuals could bring issues regarding the product safety to the authorities, as well as liability for damages and other individual responsibility issues seems curious. Clearly, a lot is yet to be deliberated.
When the regulation of big tech companies and AI is considered, all major players, US and China included, are taking steps to interfere but from different perspectives. Thus, the EU is not the only actor interfering with the AI field. However, there is a risk to regulate companies to death especially if the regulation is unclear, complex and difficult. Even if in reality this would not be the case, there is a risk that it is understood that way. Furthermore, there is a risk to decrease the eagerness to innovate. It may also turn out that the regulation might not have any real-life effects and end up being only a ‘fill the checkbox’ requirement. However, empirical data on these aspects can be collected and analysed only after the regulation has been in force. To conclude with more optimistic view present also in the conversation, these risks can be mitigated and the regulation can encourage companies to operate in the EU.
Concluding remarks
Regulating AI is undoubtedly difficult, which shines through from the problematic nature of the proposal as well as the conversations in national and EU levels. There are aspects which need to be tackled before the legislation is passed. However, the EU is seen as a desirable market area where third party actors want to provide services. It could be expected that, sooner or later, AI will be increasingly regulated globally. Hence, the EU is currently in a position where it can set requirements for the quality of the products brought into the market. In essence, as Sankari put it, the EU has the opportunity to do the whole world a favour by demanding in the regulation that producers produce better products in general. From frustration and worries to trust and belief, it remains to be seen how the regulation turns out. Going values first should not slide the drive for functioning AI regulation too much off the path.
Authors: Sofia Sormunen with Hanne Hirvonen
Sofia Sormunen, MICL and LLM student, is a research assistant at the Legal Tech Lab, and is currently writing her master thesis on the relationship of document registers to EU transparency. Hanne Hirvonen is a doctoral candidate at the Legal Tech Lab, looking into accountability issues in automated public administration.