Artificial intelligence is a radically new concept for human beings. Whether AI uptake is organic or whether it’s being heavily driven by the companies developing it (it’s the second one), it’s going to have an effect on how people think. Trouble is, OpenAI, Microsoft, Google, and the rest can issue updates faster than academics can conduct studies.
Some have been done, specifically on human cognition. The results are… interesting, even if they’re limited by a) budgets and b) participant sizes. But they don’t paint a pretty picture about what artificial intelligence is doing to human thought processes.
A Microsoft-commissioned study showed evidence that users experience “cognitive decline” through extensive use of AI. A smaller study out of Poland showed that doctors who leaned too heavily on artificial intelligence became worse at diagnoses. There are, of course, newer studies. They show similar problems.
Survey says…
The study found that there are immediate benefits to using AI. This is the bit that artificial intelligence companies sell to customers. It also found that those benefits damped down long-term thinking processes, a trade-off that’s only obvious once the supporting AI system is removed. Its authors say that though “AI assistance improves performance in the short-term, people perform significantly worse without AI and are more likely to give up. Notably, these effects emerge after only brief interactions with AI.”
Sessions measured with AI assistance are as short as ten minutes, but have an immediate effect on persistence. The experiment was done with 350 people, and then repeated with a larger pool of 630 with similar results. Those who had artificial intelligence scaffolding produced worse answers to tasks when it was removed. That group was also more likely to give up on the task entirely.
Given that a large (positive) study on ChatGPT’s effect on education was just retracted due to “discrepancies” that “undermine the confidence the Editor can place in the validity of the analysis and resulting conclusions,” it may be that the less effusive negative study has a point.
From horses to cars
AI boosters could look at these results and argue that artificial intelligence might cause changes, but will also result in new skills or patterns being developed. They might even be right, though the technology hasn’t been around long enough to create a measurable effect in that direction. Any new technology will, inevitably, cause a change in how humans think. The shift from walking to horses and from horses to cars (and so on) changed how humans measure distance and time, for instance.
But those changes still used fundamental thought processes people use for everyday survival. If you didn’t have a car, you could always walk. That’s never changed. It’s just a question of your brain scaling back what’s possible in an hour. But these studies suggest (in this context) people who don’t have a car (the AI system) will simply… not go anywhere.
There is a comparable technological change that lines up with the effect artificial intelligence seems to be having on people. It’s relatively recent, starting with the development of television.
Internet on steroids
The TV changed how human brains work. It has made them measurably dumber, for one. Unless you’re studying film (media studies doesn’t count), television interactions are inherently passive. A similar effect is seen as the internet — initially a reading- and communication-heavy medium — shifted to more passive modes of communication (social media and short videos). That one’s also measurable — attention spans have cratered.
Passive absorption (TVs) and dopamine-heavy concentrated info-bursts (social media/short-form video) are both bad in their own way. But they don’t yank out the underpinning problem-solving functions humans have developed over millennia as severely as handing reasoning, problem-solving, and research functions to an AI system seems to.
Anyone who has ever acquired a skill knows that it’s impossible to download it into your brain. Writing takes time and effort. So does math. Something as commonplace as driving requires a learning process, and even that continues to develop long after you qualify for a license. Martial arts? Musical ability? Reading? All are long-term, persistence-based acquisitions. When an artificial intelligence does similar work for you, you might ‘know’ it in the short term. It’s certainly possible to sound like an expert. But it doesn’t stick around, and the process of acquiring that knowledge seems to actively harm the ability to cement something permanently into a mind.
The hype around AI pushes the idea that humans can work faster and perform better with its assistance. That might be broadly true of users who already have fully developed reasoning systems. Even those users show declines in those systems, with no clear measurement of how bad and how long-term that damage is. Certainly, giving AI to minds who haven’t yet developed or fleshed out those systems seems to be a recipe for disaster.
