Regulating Chatbots and Deepfakes under the EU AI Act
The EU is concerned that certain AI systems intended to interact with natural persons or to generate content may pose specific risks of impersonation or deception. To deal with these concerns it is introducing transparency obligations for the following:
- Providers of chatbots and digital assistants;
- Users of deepfakes;
- Users of biometric categorisation systems, like facial recognition; and
- Users of emotion recognition systems, like those used in the entertainment industry.
Providers of chatbot and digital assistants must ensure their systems are designed and developed in such a way that end users are informed that they are interacting with an AI system, unless this is obvious from the point of view of a notional reasonably well informed, observant and circumspect, individual. This information should be provided to the person using the technology in a clear and distinguishable manner at the latest at the time of the first interaction or exposure.
Based on these criteria it would be reasonable to assume that providers of digital assistants do not have much to do here. Some work is required though. For instance, the recitals to the proposed AI Act call out specific consideration for vulnerable groups – for instance information and notifications should be provided in accessible formats for persons with disabilities.
Creators of deepfakes will need to think a little more on this point. What is obvious to one group of users may not be so in a more general context. Bear in mind also that the aim and intention of the deep fake is irrelevant. Use of a deepfake alone will trigger compliance.
There are two exceptions to this new law. One is to permit law enforcement authorities deploy this technology in limited circumstances and, the other is to allow deepfakes for evidently creative, satirical, or parody reasons. It will be interesting in time to see how this second exception will be treated and how broad its scope will be. An analogous exception under EU copyright law is generally accepted to be relatively narrow in scope.