Column

</ Thank you, CHATGPT! >

Deborah Nas

Deborah Nas is part-time Professor of strategic design for technology-based innovation in the Faculty of Industrial Design Engineering (IDE) and innovation expert. She studied IDE at TU Delft

“Please summarise this…”. “Please analyse this…”. The results appear amazingly fast, and after a few follow-up questions, I instinctively respond with “Thank you!” I’m not alone: 70% of users are polite to chatbots like ChatGPT. Why do we thank algorithms running in distant data centres? Is it our polite upbringing, or perhaps a secret fear that when AI rules the universe, it might get back at us?

It’s anthropomorphism–our tendency to attribute human traits and emotions to non-human entities. Like that university friend who affectionately called his old car “grandpa,” or encouraged his printer after repeated jams to “come on, work this time!” With AI, however, something is fundamentally different – it talks back in a way that feels astonishingly personal. And as technology becomes more human-like, our relationship with it evolves. For my new book on ‘AI Companions’ I’m exploring a future where AI becomes so personal that it can fulfil the role of friend, lover, coach, guru, or even God. I’m fascinated, but also deeply concerned. Already people share their deepest doubts, fears, and desires with digital conversation partners. They’re always there for you, never judge, and say exactly what you want to hear. But although all psychology books ever written were included in their training data, they’re not experienced therapists.

Even more impactful is the fundamental shift in technology design: previously, systems only said what they were programmed to say. With generative AI, the technology powering these digital conversation partners, we create models that can say anything, and then instruct them what not to say. We try to rein them in with guardrails, hoping that works in every situation. That’s hard. And as companies race to market, they’re bound to miss things – sometimes with huge consequences.

So when your AI someday offers life advice, consider – just maybe – consulting a human too. Or don’t. What could possibly go wrong…?