Human Learning and Generative AI: What Are the Differences? Similarities? Complementarities?

Artificial Intelligence (AI) based on Deep Learning is trained on thousands of images or texts: it generates a result – an image, a word – by evaluating the probability that it is relevant. For example, in the sentence “The cat entered the kitchen through the…”, the word “cat flap” appears not because the machine understands, but because its statistical probability is very high in that context.

At a superficial level, human learning resembles this process: we encode information, store it, and then retrieve it. But the human brain is not a passive database: it sorts, forgets, reorganizes, based on relevance. That’s why knowledge acquired for an exam often fades quickly afterwards (Crowder, 1976). As Dehaene (2014) points out, the human brain learns by ranking, categorizing, and discarding information deemed irrelevant. AI, on the other hand, keeps everything—unless programmed to delete.

On the Usefulness of Retaining Information

In both cases, the question arises: is it useful to store or retain knowledge? For humans, vital necessity is a good reason. If you know that sticking your fingers into an electrical outlet can cause severe burns or death, you're very likely to remember it. On the other hand, remembering the date of the invention of the iron is of secondary importance—unless that information becomes important to you, like winning a quiz where that date matters. For AI, the concept of death or the date of the iron's invention carry equal weight—unless an algorithm is designed to prioritize them.

In this context, storing Health, Safety, and Environment (HSE) data on a machine is simple. However, it is the human who must hold the knowledge because they are the one at risk of accidents. The challenge is to make sure each of us retains the life-saving gestures or actions to prevent accidents at any time. Using AI to know how to respond in an emergency may be useful, but not without risk: AI servers could be temporarily inaccessible.

The Paradox of HSE Knowledge

Despite the life-and-death stakes in HSE, the human memory process doesn’t always follow. A denial mechanism may kick in, leading us to believe that certain HSE information doesn’t concern us directly—or only very partially. For example, many drivers exposed to road safety campaigns might think accidents only happen to others, thus ignoring the messages. This denial acts as a form of protection from threat. Therefore, messages must be paired with strategies that increase the likelihood of real learning.

What Strategies Help People Learn Better?

Among available strategies, emotional engagement can be leveraged. Trying to scare, delight, or surprise a learner can significantly improve memory retention. Damasio's research showed how emotions influence the limbic system’s role in memory (Damasio, 1995). Unfortunately, it’s no guarantee—because the brain continues to filter and forget non-essential information.

Another strategy involves re-exposing learners to the same content at regular intervals to reinforce retention (Cepeda et al., 2006; Rohrer & Taylor, 2006; Dehaene, 2014). This can be effective. However, it’s essential to find the right scenarios and frequency for these refreshers. Discussing HSE once a year during a “Safety Day” is likely insufficient. It would be more productive to give each person regular opportunities to engage with and be stimulated by HSE content.

Offering varied formats for these reminders can be a good strategy. If you always use the same simulation, fatigue and loss of attention/motivation can set in (Söbke et al., 2020). Instead, you might alternate between full-scale simulations, digital simulations, escape games, or serious games—complemented by mobile apps or other tools to quiz users on their knowledge and the soft or hard skills to apply.

Two Key Levers to Reinforce Learning

The most effective way to anchor learning is to spark learner motivation. Motivation is essential for true learning. But it goes beyond mere engagement. A serious game might generate engagement… but the trainer’s real challenge is to turn that into true motivational drive.

Two key strategies stand out:

Naturally, combining both strategies is possible—and likely to increase the chances of fostering learner motivation.

The Importance of Evaluation

Finally, even if learners are motivated, the message must be interpreted as intended by its author. Learners may misinterpret content due to personal filters shaped by past experiences, prior knowledge, mood, context, group dynamics, and so on. All these factors may lead to different interpretations of the same experience.

Therefore, being motivated and retaining information is not enough—if that information is misunderstood (Alvarez, 2023). That’s why all learning must be accompanied by evaluations (Biggs & Tang, 2011) to verify whether learners’ interpretations align with the intended message.

On this note, it's interesting that large language models (LLMs) are often accused of generating “hallucinations”—in other words, inventing responses. But these are not mistakes in the human sense. Machines do not “err” the way living beings do. When a machine hallucinates, it simply means the statistical output does not match human expectations. The AI can then offer another answer and see if it satisfies the user. If so, the system may refine its output in the future. But it won’t understand its answer.

In contrast, for humans and other living beings, understanding one’s error is essential to learning (Astolfi, 2008). This understanding is often linked to survival.

References

• Alvarez, J. (2023). Serious game, un “carcan ludique” ?, éditions Loco
• Astolfi, J.-P. (2008). Error: A Tool for Teaching. ESF Sciences humaines
• Biggs, J., & Tang, C. (2011). Teaching for Quality Learning at University (4th ed.). Open University Press
• Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks. Psychological Bulletin, 132(3), 354–380
• Crowder, R. G. (1976). Principles of Learning and Memory. Lawrence Erlbaum
• Damasio, A. R. (1995). Descartes’ Error: Emotion, Reason, and the Human Brain. Odile Jacob
• Dehaene, S. (2014). How We Learn: The New Science of Education and the Brain. Odile Jacob
• Prince, M. (2004). Does active learning work?. Journal of Engineering Education, 93(3), 223–231
• Rohrer, D., & Taylor, K. (2006). The effects of overlearning and distributed practice. Applied Cognitive Psychology, 20(9), 1209–1224
• Söbke, H., Arnold, U., & Stefan, M. (2020). Gamified Learning: What makes an educational game motivating and effective? GALA 2020 Proceedings