AI for Me, Not (Yet) for Thee? Desirable Difficulties and Deliberate Friction with LLMs

My AI hypocrisy

Forklifts in weight rooms

Retrieval strength, programmed instruction, and performance

Storage strength, desirable difficulties, and learning

“the conditions that produce the most errors during acquisition are often the very conditions that produce the most learning”

Soderstrom, Nicholas C., and Robert A. Bjork. “Learning Versus Performance: An Integrative Review.” Perspectives on Psychological Science 10, no. 2 (2015): 176–99. https://doi.org/10.1177/1745691615569000. [online]

Desirable difficulties deliberately slow down learning:

  • Spaced practice
  • Interleaving
  • Generation

Bjork, Elizabeth Ligon, and Robert A. Bjork. “Making Things Hard on Yourself, but in a Good Way: Creating Desirable Difficulties to Enhance Learning.” In Psychology and the Real World: Essays Illustrating Fundamental Contributions to Society, edited by Morton Ann Gernsbacher, Richard W. Pew, Leaetta M. Hough, and James R. Pomerantz. Worth Publishers, 2011.

LLMs and generation

LLMs provide on-demand retrieval

 

Human generation replaced with machine generation

 

Answer Machines

“Using ChatGPT to complete assignments is like bringing a forklift into the weight room; you will never improve your cognitive fitness that way.”

Chiang, Ted. “Why A.I. Isn’t Going to Make Art.” The New Yorker, August 31, 2024. https://www.newyorker.com/culture/the-weekend-essay/why-ai-isnt-going-to-make-art. [online]

Forklifts outside weight rooms

Forklifts are useful though

Forklifts are dangerous

“In my 30-year career writing software professionally, Databot is both the most exciting software I’ve worked on, and also the most dangerous.”

—Joe Cheng, Posit CTO

“to use Databot effectively and safely, you still need the skills of a data scientist: background and domain knowledge, data analysis expertise, and coding ability.”

When you know what you’re doing, LLMs are powerful.

You cannot LLM your way into durable domain knowledge, data analysis expertise, or coding ability—that still requires some struggle

Maintaining forklift skills

29 CFR §1910.178

Continue learning with deliberate friction

Even if you know what you’re doing, there’s value in slowing down and purposely introducing Bjork-style desirable difficulties into your workflow.

Some of my strategies:

  • Give Claude Code subsets of my data and code, then implement manually
  • Treat LLM output like a StackOverflow answer or random blog post—sometimes explicitly
  • Use the learning-opportunities and learning-goal Claude Code skills

Use LLMs! (I do!)
But don’t sacrifice learning