Ozan Sonmez argues that the future of AI is highly uncertain — technologists and ethicists disagree about outcomes and appropriate ethical boundaries. He highlights regulation trends, especially data-protection measures (GDPR and related legal work) as the most enforceable response, and doubts broader regulation’s feasibility. He illustrates models’ opacity with an anecdote: AI can predict gender from an isolated image of a retina with high accuracy even though humans cannot explain how, underscoring limited interpretability and that many practitioners themselves don't fully understand why models behave as they do.
Chapters
Not even the technology enthusiasts agree on what AI will create. Nobody in the ethics side really can put their finger on what should be the ethical use of AI. A lot of debate is going on and we'll keep on going, and that's how new technology is being adopted. So when you look into regulation, I think the only thing that could have happened, and I'm glad it happened, is like data protection. Everything around will be enforced now, because now people will start saying, okay, now that copyright will extend to feeding into algorithms. And it's easy to do it because there's infrastructure. They've been working on it for years and years and years. GDPR and data price law didn't come into effect in a short amount of time. But anything other than that, I think it's completely impossible to take it out from our sphere of influence. Major players might be maybe regulated. But again, most of this is like complete black boxes. Not even the people who kind of initiate this and send it out to the world understand why they does it. They do it that way. One funny story was like, they, it's. Sorry. As I read, it's impossible to detect the gender of an individual if you only have their eye. Like if you only have the retina with nothing, nothing there. It's just there on the out. Nobody can understand if it's a man or a woman. AI can understand that. They tested it and it's like freakingly accurate. Like anything above 50% accuracy has already been more successful. Like you can go random 50, 50 is okay. AI is very accurate. They don't know why, they don't know what it uses to understand from the looks, from the picture of the eye. If it's the gender, there's nothing. And people are surprised. Like, they don't know. They don't understand most of the stuff. When you ask people doing AI research and applications, like, I don't know, some are saving themselves by saying, I don't know. That's. I don't think that's honest. Most of them are 100% honest. But most people doing research, most of people doing these models actually don't know.
