Regulation of AI applications in the US is already in place or is about to be introduced, but it seems that Europe still has a long way to go from the first draft of the “AI Act” to its entry into force…
According to the US National Consumer Protection Agency, the Federal Trade Commission (FTC), regulation of AI applications is already status quo (“The reality is[in the US]AI is regulated”). Unfair and deceptive trade practices should also apply to AI applications, and the FTC’s authority should be extended to companies that manufacture, sell, or use AI. One of our top priorities is also setting the tone for the promotion of AI applications. Guidelines on how to use user data for training algorithms and AI applications have already been made by him three years ago, and transparency is once again a key principle.
21, there has also been a draft EU-wide AI regulation.st April 2021 “Europe will become a trusted global center for artificial intelligence (AI).”Since then, the draft has been discussed and negotiated under various Chairs of the EU Council. The draft primarily represents a ban that either bans the use of certain AI applications outright or ties their use to more extensive preconditions and safeguards. The European Parliament is now due to vote on the draft by the end of April, and its representatives will join her tripartite session with the European Commission and the European Council in May to reach a final position. can decide.
There is also currently a growing debate about how the so-called Large Language Models (LLMs) on which applications such as ChatGPT are based should be classified within the regulatory framework, and how they should be classified. There seems to be “High Risk” applications under draft.
In summary, the current draft takes a risk-based approach, distinguishing between AI applications that are classified as (i) unacceptable risk, (ii) high risk, and (iii) low or minimal risk.
AI applications (or “AI system(as defined in the draft) are mainly associated with unacceptable risks and therefore where their use violates EU values, e.g. fundamental rights (e.g. human behavior or social scoring operations) are prohibited.
In previous discussions on the draft, chatbots were generally categorized as low-risk or minimal-risk applications, with only “minimal” transparency obligations as a legal consequence.
On the other hand, AI applications categorized as high risk typically have to meet a number of additional requirements, such as:
- According to Article 10”Data and data governanceThe draft states, “These applications can only be trained on data with certain qualities further specified in the draft.
- Applications should also be designed and developed in such a manner.
- (i) among other things, automatically recording/recording an ‘Event’ (Article 12 “Record keeping“);
- (ii) the operation is sufficiently transparent to allow users to properly interpret and use the results (Article 13 “Transparency and provision of information to users“);
- (iii) the application can be effectively monitored by a natural person during the period of use (Article 14 “human oversight“); and;
- (iv) achieve an appropriate level of accuracy, robustness and cybersecurity in light of its intended purpose and perform consistently in these respects throughout its life cycle (Article 15 “Accuracy, Robustness, Cybersecurity”).
- Article 16”Obligations of providers of high-risk AI systems” also imposes extensive obligations on providers of such applications – for example, they must perform conformity assessment procedures.
However, the absence of European regulation does not mean AI is not regulated in the EU or Germany at this time. Naturally, in individual cases, laws already in force may also apply to AI applications. We can already see this under the terms of the GDPR. After ChatGPT was banned in Italy (as reported by us), German data protection authorities are now also investigating its use in Germany, and the European Data Protection Board (EDPB) decided last week that: bottom. Launch a task force to facilitate cooperation and exchange information on enforcement actions that may be taken by data protection authorities. Also, under the new Digital Services Act (DSA), certain regulations may apply to AI applications (and possibly LLMs, depending on how they work). Prohibition of “dark patterns” (Article 25)Online interface design and configurationRegarding the transparency rules (Article 27) of the recommendation system using “DSA” or AI applicationTransparency in recommender systems“DSA).
As a result, the general question of AI regulation in Europe and Germany will continue to be exciting in the coming weeks, especially with the European Parliament’s vote on a draft AI regulation due at the end of April. Furthermore, the deadline set by the Italian data protection authority against the European representative of his OpenAI company, the provider of ChatGPT, expires at the end of April. By then, the provider will have to put forward measures to correct the identified deficiencies. Otherwise, poor compliance could result in a fine of up to €20 million or his 4% of the group’s global turnover.
