EU’s AI Act challenge: balance innovation and consumer protection

Members of the EU Parliament have agreed on a first draft for regulating the use of AI. The AI ​​Act is now taking the next procedural step to be negotiated and worked out with individual member states. In the end, there should be an EU-wide body of law to regulate the use of AI technologies, such as ChatGPT.

Essentially, the AI ​​Act is about categorizing AI systems into specific risk classes ranging from minimal, to systems with high risks, and those that should be banned altogether. And when it comes to AI systems making consequential decisions for people, especially high standards should be applied, particularly to the transparency of data a particular AI was trained for its decision-making, and how the algorithms work to ultimately make decisions. In this way, EU politicians want to ensure that these AI applications function securely and reliably, and don’t violate fundamental human rights.

Until a final set of rules is in place, though, there will be an inevitable amount of discussion within the various EU bodies as there is no consensus. Italy, for instance, has recently taken a tougher stance and banned Open AI’s generative AI tool ChatGPT due to a lack of age controls for use and possible copyright infringement in the training data. In the meantime, however, Italian authorities have allowed ChatGPT use again under certain conditions.

Other EU countries followed the initiative of the Italian data protection authorities as well. Germany raised that ChatGPT should be banned if it can be proven the tool violates applicable data protection rules.

Too much regulation hampers innovation

While consumer advocates are calling for strict rules to protect citizen rights, business representatives warn that overly strict regulation of technology could lead to more innovation slow down. According to advocates of a less strict interpretation of the AI ​​Act, the EU could fall behind in an important future-oriented industry.

In an open letter, representatives of the Large Scale Artificial Intelligence Open Network (LAION eV) called on EU politicians to moderate AI regulation to proceed. The intention to introduce AI supervision is welcomed, it says, but such oversight must be carefully calibrated to protect research and development, and maintain Europe’s competitiveness in the field of AI. Signatories include Bernhard Schölkopf, director at the Max Planck Institute for Intelligent Systems in Tübingen, and Antonio Krüger, head of the German Research Center for Artificial Intelligence (DFKI).

LAION demands that open-source AI models in particular shouldn’t be over-regulated. Open-source systems in particular allow more transparency and security when it comes to the use of AI. In addition, open-source AI would prevent a few corporations from controlling and dominating the technology. In this way, moderate regulation could also help advance Europe’s digital sovereignty.

Too little regulation weakens consumer rights

On the other hand, the Federation of German Consumer Organizations (VZBV) calls for more rights for consumers. According to a statement by consumer advocates, consumer decisions will in future be increasingly influenced by AI-based recommendation systems, and in order to reduce the risks of generative AI, the planned European AI Act should ensure strong consumer rights and the possibility of independent risk assessment.

“The risk that AI systems lead to false or manipulative purchase recommendations, ratings and consumer information is high,” said Ramona Pop, board member of VZBV. “The Artificial intelligence is not always as intelligent as the name suggests. It must be ensured that consumers are adequately protected against manipulation and deception, for example, through AI-controlled recommendation systems. Independent scientists must be given access to the systems to assess risks and functionality. We also need enforceable individual rights of those affected against AI operators.” The VZBV also add that people must be given the right to correction and deletion if systems such as ChatGPT cause disadvantages due to reputational damage, and that the AI ​​Act must ensure AI applications comply with European laws and correspond to European values.

Self-assessment by manufacturers is not enough

Although the Technical Inspection Association (TÜV) basically welcomes groups in the EU Parliament to agree on a common position for the AI ​​Act, it sees further potential for improvement. “A clear legal basis is needed to protect people from the negative consequences of the technology, and at the same time, to promote the use of AI in business,” said Joachim Bühler, MD of TÜV.

Bühler says it must be ensured that specifications are also observed, particularly with regard to transparency of algorithms. However, an independent review is only for a small part of AI ​​systems with high risk intended. “Most critical AI applications such as facial recognition, recruiting software or credit checks should continue to be allowed to be launched on the market with a pure manufacturer’s self-declaration,” said Bühler. In addition, the classification as a high-risk application should be based in part on a self-assessment by the providers. “Misjudgments are inevitable,” he adds.

According to TÜV, it would be better to have all high-risk AI systems tested independently before launch to ensure the applications meet security requirements. “This is especially true when AI applications are used in critical areas such as medicine, vehicles, energy infrastructure, or in certain machines,” said Bühler.

AI should serve, not manipulate

While discussions about AI regulation are in full swing, at a meeting in Takasaki, Japan, at the end of April, the G7 digital ministers spoke out in favor of accompanying the rapid development of AI with clear international rules and standards according to a statement by the Federal Ministry for Digital Affairs and Transport (BMDV).

“We in the G7 agree that when it comes to regulating AI, we must act quickly,” said Volker Wissing, Germany’s Minister of Transport and Digital Infrastructure. “Generative AI has immense potential to increase our productivity and make our lives better. It’s all the more important that the large democracies lead the way and accompany its development with clever rules that protect people from abuse and manipulation. Artificial Intelligence should serve us, not manipulate.”

But it’s questionable whether it will happen as quickly as Wissing would like, seeing as the AI ​​Act has been in the works in Brussels since April 2021. After the agreement in the EU Parliament, trialogue negotiations between the Council, Parliament, and Commission could begin in the summer of 2023. It’s anyone’s guess when a final set of rules can be put in place and converted into applicable law. It’s also questionable if technological development of AI will outpace attempts at AI regulation by then.