"With the "EU AI Act", the European Union is committed to creating one of the world's first legislations on artificial intelligence (AI) for the benefit of society. But is it even necessary? We believe: No. A reply of the Post by Svenja Hahn from October 10, 2023.
The planned EU AI Act is taking more and more shape, and many who are working on it are proclaiming it as good news: it is intended to serve as a global model to reduce the possible risks of AI and the dangers that can be associated with its use prevent. It is intended to strengthen trust in AI on the user side. At the same time, it should not represent over-regulation that could inhibit innovation in Europe and put even more strain on already struggling medium-sized businesses. So far, so questionable.
In our opinion, what is particularly questionable is the seemingly banal consideration at first glance as to what should actually be regulated horizontally around AI that is not already regulated. Where is the regulatory gap that the EU believes it sees, that it fears and therefore vehemently wants to close? We don't see this gap. Instead, we see that for numerous industries there are already a lot of vertical requirements of national and international origin that manufacturers must comply with. The Machinery Directive and the Medical Devices Regulation are particularly well-known here.
If there were regulatory gaps, it would be important to first clearly identify them and then specifically close them. But this should then be regulated in general, i.e. vertically related to a specific application or domain, and not with a focus on a specific technology such as AI. We see AI as one tool among many that can be used for product creation or use. It is a means to an end and not the root of evil. You can see this, for example, in mass surveillance in public spaces: AI can be used for this, and the AI Act could prohibit that. But mass surveillance can also be implemented without AI - so it would be more important to regulate or ban mass surveillance itself and not a tool that could be used but does not have to.
In our opinion, the classification of AI applications, as provided for in the AI Act, is also problematic. It focuses on banned and so-called high-risk applications. With this assignment we see many open questions. This starts with the fact that the definition of what AI actually is is not clear, even between the EU players involved. There are three different definitions of AI, depending on whether you research the Council, the Commission or the Parliament. In some cases, these definitions go so far that, strictly speaking, all statistical software would be regulated as AI applications, which is technical nonsense.
Similar doubts arise when one looks at what should be considered a high-risk application. We see a lot of trivial stuff here. For example, who would have thought that AI-supported automated admissions and testing at universities would be considered high risk? This sometimes questionable assignment results in requirements that are currently far too high and prevent the use of AI in many places, where it could provide relief and efficiency. Risk classification is also not trivial for specific use cases. Applied AI found in a study that the classification of 40 percent of the use cases examined was unclear.
The classification also raises the question of who should actually certify all the AI-based “high-risk systems”. It looks to us as if, as with the medical device regulation, the EU is creating an unnecessary and costly bottleneck for itself that is preventing innovation. And here we are back to the medium-sized businesses mentioned at the beginning: Large companies may still be able to cope with such expenses, but the small ones will be deterred or immediately left behind. The introduction of the General Data Protection Regulation (GDPR) is a warning example of this.
It is obvious that with all this effort, China and America will certainly leave us technologically behind (if they have not already done so).
One final technical point about Ms. Hahn's article: She explained that the AI Act should regulate applications and not the technology itself, as the latter is changing too quickly. In fact, so much is currently happening around AI that the AI Act will permanently appear outdated. This could be observed in the way "General Purpose AI" - a term created by the EU but which actually aims at the core of every machine learning process - and foundation models should be taken into account in a technology-specific manner. Not to mention that generative AI technologies have been around for decades and have only now found their way into widespread use.
So what needs to be done to meaningfully address the disruptive possibilities of AI on the EU side? For us, there are clear tasks to be completed here: These include, firstly, financial and intellectual support, secondly, education and thirdly, disseminating best practices. In concrete terms, this means: Let's create a climate that sees the topic as an opportunity and not as an enemy. This also includes competitive financial support for AI technologies. One should remember here the “Large European AI Models” (LEAM:AI) initiative to develop large, data-rich AI models. The idea is good - there is no money for it.
If you look at the EU subsidies for many other industries, these priorities should give you food for thought. Improved education on all topics related to digitization is necessary so that our society can have a say, participate in action and help shape it, because AI is here and we will all have to learn to deal with it and work with it. And finally, creating role models: AI is not a panacea. Making it tangible what AI can and cannot do in which industries creates realistic ideas and scope for action. Let's all help shape the technology instead of unnecessarily blocking it with a lot of bureaucracy." [1]
1. AI Act: KI fördern statt kaputtregulieren. Frankfurter Allgemeine Zeitung (online)Frankfurter Allgemeine Zeitung GmbH. Oct 25, 2023. Von Patrick Glauner und Marco Huber
Komentarų nėra:
Rašyti komentarą