I am going to offer you Goldman’s deal below. Would you take it?
“If I had a magic drug that was so fantastic that if you took it once you would win every competition you would enter from the Olympic Decathlon to the Mr Universe, for the next five years but it had one minor drawback; it would kill you five years after you took it, would you still take the drug?”
This dilemma was proposed to world-class athletes in the 1980s and 1990s by a physician called Bob Goldman (hence the name ‘Goldman’s Dilemma’). Shockingly, more than half the athletes said they would take it. It seems that when it comes to achieving glory, for many, no limits or boundaries apply.
This offers an interesting lens to look at how students use AI. In this analogy, let the pill that Goldman described be Artificial Intelligence, or ChatGPT. Let’s slightly rephrase Goldman’s dilemma so it makes sense in the classroom:
“If you used AI, and it meant that you got the highest grade possible on every assignment, but it damaged your actual long-term learning and retention, would you use it?”
Students might already be choosing the ‘pill’, often without realising it.
Read more about Goldman’s Dilemma, including:
How does AI impact long-term memory?
There is no denying that AI is helpful in making processes more efficient, such as potentially cutting lesson planning time by 30%. There is also no escaping that students are undoubtedly using it to support their work, with a recent survey showing that 92% are using it in some form.
A recent study from MIT looked at the impact of AI use on memory. Participants were split into groups, in one they weren’t allowed access to any external information when writing an essay, making it a completely independent task. In the other group, participants were allowed to use ChatGPT to assist in their writing process.
After completing the essay, participants were quizzed about their experiences and the essay content. When participants were asked “Can you quote any sentence from your essay without looking at it?“, only 11% who wrote it themselves failed to answer, compared to a staggering 83% of the ChatGPT group.
This offers a striking insight into how AI may affect retention and raises the possibility that students could produce work they barely remember. It almost feels like Goldman’s Dilemma in action.
Why does this occur?
Learning is difficult. Meaningful learning requires hard, cognitive work so that information is embedded. Using technology or AI might be seen as the path of least resistance. It feels easy and gives us the illusion that we are learning, but does the information truly stick?
This concept is called Cognitive Offloading. When we rely on technology to remember information for us, we offload the cognitive demands and challenges of remembering information ourselves because we know that the technology can do this for us. While it feels easier, it comes at the expense of actual learning. Desirable difficulties, such as Retrieval Practice have been shown repeatedly to enhance retention and long-term memory.
AI Tutors: A Better Way?
AI should support the above processes, not replace them. Ultimately, the question is not whether students should use AI, but whether they can use it without choosing the educational equivalent of Goldman’s ‘pill’.
Fortunately, recent evidence suggests some positive potential. Consider the following two studies:
Study 1 – In this study, researchers focused on using an AI tutor chatbot to help students decide what they should practise next in order to improve. Over the period of 5 months with 770 students, they found that performance in their final exams significantly increased by “the equivalent to as much as six to nine months of additional schooling”.
Study 2 – In this study, which was a randomised control trial (RCT) across UK classrooms, students who used a generative AI model that had been trained in pedagogy performed at least as well as students who didn’t have access to this (and when it came to solving novel problems, actually did better). The researchers highlighted that these comparable outcomes are particularly promising, as they could potentially be scalable across education.
Final Thoughts
Education is often seen as a means of developing knowledge and preparing students for the real world. However, if widespread use of AI shifts the focus towards simply achieving grades and not on learning, this is highly problematic. For knowledge to stick, cognitive effort is required; this is a non-negotiable aspect of how we learn. Hopefully over the coming years we will see a shift of AI being used with students to help facilitate this, not shortcut it.
From the world of athletics to the classroom, Goldman’s Dilemma highlights an important trade-off between short-term success and long-term outcomes. Is AI use worth it if students aren’t truly learning? For educators, the challenge is clear: figuring out ways for AI to be used in a way that helps learning, not hinders it. Without careful guidance, we as a society run the risk of promoting superficial performance rather than meaningful learning.
