“I discovered something new,” a friend says on the terrace. I look at her expectantly. Something new! What would it be? A way to be able to concentrate for more than two hours a day? A Sally Rooney with humor? A cafe where you still have a day snack for 10 euros?
Unfortunately, it turns out to be chatgpt. The girlfriend is one late adopterbut she makes up for her backlog by using ‘chat’ for really everything: from dilemmas in her private life to advice on nutritional supplements. I have to admit: this sounds very handy. It can even be a solution, because I camp with a refusal dishwasher. For me, household appliances are something as a problem wolf Bram: deeply incomprehensible and unpredictable, and therefore difficult to approach. Perhaps Chatgpt can unlock the mystery to me.
Yet her enthusiasm bumps into me. I recognize myself in the words of Frank Heinen, in Thursday de Volkskrant: “I would like to look something up on chatgpt, but something in me says that I will open a door that I will never close again.” But why? I have to think about that. In the end I conclude that it is exactly this, thinking, that Large Language Models (LLMs) If Chatgpt not only take over from us, but also take ourselves. After all: who is busy thinking, does (if all goes well) select, summarize, distinguish, verify, make connections. How do you maintain those skills if you outsource them to an LLM?
Fortunately, there appears to be research on this question. This spring it became one chapter published from the book that is still to be published The Artificial Intelligence Revolution: Challenges and Opportunitiesto which a lot is being referred to. The central statement of Barbara Oakley, the author of the chapter: if we do not store any knowledge in it, our brain will deteriorate. Firstly, because your working memory wastes searching for things that you could have known, and secondly because you do not train your brain to organize and connect and connect themselves. By recognizing patterns in the breach of facts, we learn more efficiently and creatively thinking.
Using external memory supports as Chatgpt is called Cognitive offloadingand exists much longer than AI. Think of calculators and birthday calendars: that means we do not have to remember tables and birthdays. But if you keep too much knowledge outside your brain, it therefore influences your cognitive abilities. In the words of the researchers, “metacognitive laziness.”
There have been worries about the stupid man for some time. The IQ became higher and higher worldwide, but has been falling for a few decades in various countries. Last March Financial Times-Journalist John Burn-Murdoch in his article ‘Have Humans passed Peak Brain Power?‘The distraction by technology as the main cause. That certainly plays a role, says Oakley, but the problem is deeper: education has been focusing on skills instead of knowledge, with those metacognitive laziness as a result. AI will make this even worse.
It is in line with what neurobiologist Kenan Malik recently wrote The Observer. LLMs when Chatgpt do not care about our thinking, it is the other way around, according to Malik: they are embraced because we think, and the knowledge you need for that no longer find valuable in themselves. Now this seemed like a typical case of ‘why not both’: thinking can be low in prestige, be replaced by AI, and therefore being practiced even more moderately. But apart from that, it was an interesting point: we throw something away because we do not (there) know the value.
For me as a columnist, the importance of an internal archive is very clear. I can only come up with ideas by related to the things I have seen and read with each other. But the same principle applies to all of us in all kinds of areas. We only catch a lying politician when we know the facts. We understand the news through knowledge of the context. We can estimate distances if we sometimes read card and think about the best route.
Artificial intelligence should not be ‘stool’, in the sense that you can no longer do without, says Oakley. It must be a supplement to your thinking, not a replacement. The New York Times Last showed how that works. The 81-year-old psychologist Harvey Lieberman described in a essay His experiences with Chatgpt as his personal therapist. He was positively surprised: some contributions were moderate or incorrect, but others gave him new insights. Lieberman picked out what he found useful and ignored the rest: he himself remained at the helm as a thinker. The model was “not a stool, but a cognitive prosthesis – an active expansion of my thinking process.”
Lieberman saw chatgpt as a conversation partner, not as an oracle. But for that you must have intellectual self -confidence. You only develop that by continuing to think for yourself.
Floor Rusman ([email protected]) is editor of NRC