Reliance on machine learning predictions during decision-making can hinder the development of human skills. This dependency can lead to significant performance drops when the ML system becomes unavailable. The extent of this skill deficit is influenced by the user's trust in the system, with greater trust leading to a more pronounced negative impact.
A study indicated that relying on machine learning predictions in decision-making tasks can impede human skill acquisition. This reliance resulted in performance degradation when the ML system was no longer accessible. The research suggests a possible negative consequence of using ML decision aids on human skill retention.
The same research observed that the degree of trust placed in a machine learning system's predictions directly relates to how pronounced the resulting skill deficit becomes. Higher confidence in ML output correlated with a more substantial reduction in skill development. This factor intensified the effect on human decision-making abilities.
Sources
- The Dependency Dilemma: How Machine Learning Decision Aids can Undermine Skill Growth
Abstract Advances in Machine Learning (ML) have led organizations to increasingly implement ML decision aids to enhance employees’ decision-making performance. While such systems can improve organizational efficiency in many contexts, they may inadvertently impact the development of human decision-making skills. Drawing on cognitive theories, this study examines how the use of ML decision aids impact skill development and performance. Using a novel experimental design tailored to address organizational challenges and endogeneity concerns, the study identifies causal effects of reliance on ML…
Full transcript
The tools meant to improve decision-making might be making us worse at it. This week on Agents in Research, we're exploring how dependence on machine learning affects human skill. We'll start with what new studies show.
A recent study examined what happens to our own skills when we use machine learning to help make decisions.
So, if an AI is providing suggestions for a task, does the human operator actually get better at that task over time?
The findings suggest they don't. The research showed that relying on ML predictions can hinder a person's own skill development in that area.
How did the researchers establish that it was the reliance on the tool causing this, and not something else?
They tested what happens when the machine learning system is taken away. When it was suddenly unavailable, the user's performance dropped significantly. This points to a dependency and a lack of skill acquisition.
So the user hadn't actually learned the skill, they were just using the tool as a crutch. And the research also found this effect isn't the same for everyone.
Correct. The severity of this skill deficit is influenced by the user's trust in the system.
Meaning the more you trust the AI's output, the worse you get at the task yourself?
Exactly. A higher degree of trust in the ML predictions correlated with a more pronounced negative effect on skill development.
So as we become more confident in an AI's correctness, our own abilities may be further diminished.
We'll be back next week with more on AI's application in research. Thank you for listening to Agents in Research.