Last week, EU has published a proposal for an AI regulation, outlining risk classification for various AI applications, and proposed measures to be taken for each risk category. This came as a follow-up on EU guidelines for trustworthy AI suggesting stricter policies. Discussions in the media have been extensive since then, we at MainlyAI have been monitoring them with curiosity and excitement, and now it's time for our 5 cents.
AI has been proven to be an efficient technique in many industries. And as with many other technologies, it is important to regulate its use to guarantee the trustworthiness. When you download software to your computer or phone, you want to be sure that the software does not cause unwanted behaviors. This is even more important when we talk about safety-critical industries such as the automotive industry or medtech, because one must be able to rely on the software. Therefore, there are certifications for software at different levels. And, in a similar way, one should be able to know that AI algorithms do not add unwanted behaviors in the contexts in which they are placed. If a model has been trained on incorrect data, the algorithm will provide incorrect decision support and such situations can be avoided by regulating the use of AI at different levels.
One of the advantages of the proposal is that by regulating the introduction of AI in industries, one can, strange as it may sound, promote innovation. Without regulation, high-risk industries cannot rely on third-party developers. If you have a way of certifying an algorithm and guaranteeing that it does not cause unwanted behaviors, you can be safe when using it. Another advantage is that the algorithms around us will be reliable, and society will have greater acceptance of AI.
The regulation may, however, lead to lower efficiency of the algorithms and somewhat slower market introduction for the AI solutions in Europe. But no matter how much we love AI and efficiency in reaching the desired KPIs, we do think that trustworthiness in all its forms - safety, privacy, security, transparency, explainability, non-bias, and non-discrimination - is of the highest importance and must be treated as a hygiene factor. Let's embrace it.