In my first 30-minute writing sprint, I claimed that not all friction is bad, specifically in the context of outsourcing cognitive functions to AI or a large language model (LLM). So, how do we know what friction is good?
To me, the answer lies in reframing the question to accommodate the idiosyncratic nature of potential friction. The better question is how do we determine if this friction is actually practice for a skill that you want to preserve or enhance? If the skill is something you want to retain and improve, then outsourcing to AI is likely not the best decision. If the use of AI poses no threat to the struggle that makes you stronger, perhaps it is a promising candidate for cognitive outsourcing.
About a year ago, there was a widely published paper called Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. Those who used ChatGPT to assist with their essay writing recorded worse memory recall compared to those who did not use ChatGPT (brain-only group and search engine group). Unsurprisingly, the brain-only group had the strongest memory recall. A quick example of this is I can easily recall what I wrote for my previous post because it was brain-only. If I had used an LLM to generate a 1,500+ word post about using generative AI for content creation, there's a good chance I wouldn't remember it as well.
The abstract for Kosmyna et al's paper states:
Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels. These results raise concerns about the long-term educational implications of LLM reliance and underscore the need for deeper inquiry into AI's role in learning.
There are certain struggles that make you stronger, which is similar to saying there is certain friction that promotes learning.
Let's look at another example related to work. AI generated emails based on transcripts are possible with basic tools, or directly in your email client. For some, the skill of writing a perfect email is treasured, while others actually hate it. Those who outsource their email writing will likely not improve at writing emails and based on the study mentioned, they might have worse memory regarding what the email was actually about.
That said, those who outsource the task to AI can also learn something from the generated copy and use it to improve. I once worked with someone who was talented in his trade but English was not his first language, and he told me that the AI generated emails helped him learn how to communicate in English and use jargon related to his trade.
The question is, what's the trade-off? How does the person who hates writing emails determine if the struggle they experience is actually practice for a skill that they ought to enhance? What do they gain by sacrificing the skill and accepting degraded memory recall? Do they gain time for higher value decisions? Do they have their thoughts communicated in an accurate and concise way that they could never achieve even with years of practice?
The main takeaway is when you are faced with the decision to potentially outsource a task to AI in an effort to reduce friction, ask yourself, "Am I sacrificing a struggle that would ultimately make me stronger at something that I want to be strong at?"
As my grad advisor, Gus Skorburg, put it in a recent substack article, LLMs "are the greatest tool ever invented for learning. They are also the greatest tool ever invented for not learning."
Back