Since mid-June, the European Parliament has been pending a regulation to regulate artificial intelligence (AI), the so-called AI Act. Especially the classification of new services such as ChatGPT by OpenAI has been against the initial draft of the European Commission anxiously awaited.
Announcement
“The idea back then was that rating systems in particular should be regulated,” says Sandra Wachter of the Oxford Internet Institute. Because these systems would often be used as an aid to decision making. “Should I hire this person, should I give him a loan, should I let him go to university? Then ChatGPT came along and then the world was totally different.”
Problem of bias and ambiguity in AI regulation
The lawyer Sandra Wachter has been dealing with the ethics of AI, explainable AI and how AI should be regulated for many years. In the new episode of the podcast, he discusses, among other things, the problem of the bias of large AI models and what ambiguities the draft AI Act still contains in an interview with TR editor Wolfgang Stieler.
Wachter thinks it’s good that the inclusion of generative AI in the law didn’t mean the risk-based lawsuit was dropped. Furthermore ChatGPT and Co. are not classified as a high-risk technology from the outset, but should be strictly regulated. To ensure transparency, training data should be widely open. In this way, the causes of possible distortions, i.e. bias or bias, should become clear and correctable.
Here you will find an overview of our three podcast formats: the weekly news podcast “Weekly” and the monthly podcasts “Unscripted” and “Deep Dive”.
On the other hand, he criticizes the fact that the manufacturers of such systems certify in a “self-assessment” that their products comply with the AI law. Even more so is, for example, “that you should make an effort or do your ‘best effort’ to make sure you address the bias,” Wachter criticizes. He also provides insight into what the law means for users and businesses and how much influence lobbyists have on regulation.
Announcement
More on this in the whole episode – as audio stream (RSS Feeds):
(wst)