Column Judith Witte: Halluci­nating

Food production in factories isn’t the only thing getting ‘smarter’. On a smaller scale, our homes and kitchens are gradually becoming thoroughly digital environments too, with ever-smarter appliances.

I say ‘our’, but what I actually mean is ‘your’. My husband and I are a little slow on the uptake in this department, I’m afraid. We’re probably the only ones on our street without a robotic lawn mower, and we still operate our thermostat and lights manually. You know, by physically pressing on/off buttons. When I want to whip cream, I reach for a whisk. Mashed potatoes turn to glue if you use an immersion blender, so I put mine through a potato ricer. Still, a hypersonic new addition recently arrived on our kitchen counter. My coffee grinder is called the ‘Smart Grinder Pro’. It uses beeeeep-technology to grind my coffee beans ever so cleverly to exactly the right consistency.

Because I love that liquid black gold so very much, I studied the little booklet that came with the machine and learned absolutely everything about grind settings and how they affect the intensity of my cup of coffee. I started experimenting with the settings and writing down my findings. The old-fashioned way: with a pen in a notebook. My state-of-the-art coffee grinder isn’t connected to the cloud either. Not yet, anyway. The good news: that means it can’t be hacked. No scoundrel is going to run off with the data I’ve so carefully entered!

Breaking into systems you have no business accessing is a serious criminal offense in most countries. Unfortunately, that doesn’t deter cybercriminals or AI agents, judging by the barrage of hacks and data breaches in recent months. Customer data from Basic-Fit, Bol, Booking, Rituals, Odido and Bijenkorf, among others, is out in the open. You’d have to live under a rock to avoid becoming a victim.

Meanwhile, the average mortal, most still digitally illiterate, hunches over a keyboard in bewilderment, trying to write a prompt for some chatbot or other. Will it produce a sensible answer this time? Or something that sounds utterly convincing but is factually wrong? It could easily happen, because the language model’s goal is to produce a plausible, fluent answer, not to verify the absolute truth. There’s even a word for this gibberish: hallucinating (named Word of the Year 2025 by Van Dale). A chatbot can also be guilty of sycophancy: the phenomenon of technology being excessively agreeable and flattering toward the user. The danger of brain rot now looms. You’re more likely to develop this new condition if you spend a whole lot of time joyscrolling.

I read in my newspaper (the paper kind) that the UN is seriously asking whether ‘AI will bring about humanity’s downfall’. A valid question, and one we’re all putting off answering. We’d rather bury our heads in the sand. Lost in thought, I take a sip of my coffee…

Wow! It tastes so good today, I must be hallucinating.

Judith Witte
[email protected]