I’ve been meaning to post on this since the article came out a month ago (see link below). Back then, this article did trigger a bunch of discussion. But I’m not sure folks really nailed what the studies actually suggest.
Basically, there was a string of studies that observed the effort and output while using a genAI tool. And there were two key findings that were not surprising to me.
Recent studies suggest that tools such as ChatGPT make our brains less active and our writing less original. Source: A.I. Is Homogenizing Our Thoughts
Cognitive offloading
Researchers measured brain activity in those who were using genAI to complete writing task. They found less brain activity relative to those who were not using genAI.
This makes sense, the genAI tool takes over some of the cognitive effort needed to complete a task. This isn’t much different from a car taking away some of the effort to get from one point to another.
Indeed, human progress is chock-full of all the ways we’ve offloaded effort.* And offloading such effort, we’ve been able to do more, be more productive.
Homogenization
One other finding from these studies is that the output of these task ended up being similar, dare I say, bland. Again, not surprising. While we might think different people are driving the tasks, in reality, they are offloading the effort to the same ‘entity.’ If we all use Claude.ai, then the output might be similar because, in reality, the output is coming from the same place: Claude.ai.
Added to that, these big genAI systems, when left their own devices, seem to go towards averages, a consequence of them hovering up as much as possible and making sense of it. That process tends to teach them what’s common rather than what’s special.
Looking in the wrong place
When the article came out, most folks did the aggressive pointing, saying, ‘oh, no, our brains will be mush’ and ‘oh, no, all writing will converge towards a boring average grey goo’ and ‘look at all the pretentiousness of all those em-dashes and lofty, awkward words.’
Yet, these are not the real things these studies reveal.
What these studies reveal is: we cannot surrender all cognitive effort. As I keep saying, you cannot use genAI on autopilot.
My own experience with genAI tools is that the less effort one puts in, the worse the output is. There’s only so far one should be willing to cognitively offload.
Indeed, if you offload it all (go on autopilot), then you do get that boring, homogenous, identi-krap. What do you expect?
Pilot in the cockpit
In summary, I am not surprised that when we use genAI on autopilot, our cognitive functions are quiescent and we spit out bland and similar output.
With genAI, you must be in the pilot seat, you must use your own brain.
As I learned from some smart folks recently: We must still be the ones making the decisions and putting in the effort. That’s how we’ll separate the slop from the good stuff.
*🤔 Geez, in what ways have we offloaded cognitive effort? Ask the Homeric bards in Ancient Greece what they think of books. Socrates, for one, was not too enthused.
