Human resources and people managers are being urged to approach the use of artificial intelligence in the workplace with care, particularly when it is introduced to enhance efficiency and strengthen human capital. New research from the University of Bath School of Management suggests that while AI can offer clear operational benefits, organisations must actively safeguard creativity and critical thinking. Without deliberate strategies, there is a risk that the very capabilities that underpin long-term performance and innovation could be weakened rather than strengthened.

According to Dirk Lindebaum, author of the study On the Dangers of Large-Language Model Mediated Learning for Human Capital, AI is often presented as a straightforward solution for improving productivity. It promises faster problem-solving, tailored responses, and streamlined workflows. However, he cautions that such claims should not be accepted uncritically. The apparent efficiency gains may obscure deeper consequences for how employees learn, think, and engage with their work. Human knowledge, the research emphasises, is not a single, uniform construct but a collection of distinct forms, each interacting differently with AI technologies.

Some types of knowledge appear partially compatible with AI, particularly encoded knowledge—such as formal rules, procedures, policies, and datasets—and embedded knowledge, which includes structured routines and digitised processes. In these areas, AI can support tasks like updating documentation, improving compliance, and refining workflows, offering what may seem like quick and attractive gains for managers. Yet even within these domains, risks remain, as employees may gradually disengage from the underlying processes.

Over time, this reliance can erode familiarity and reduce the depth of expertise needed to manage exceptions, adapt to change, or identify errors. What initially appears to be an efficiency gain may, in practice, result in a more fragile workforce with diminished practical understanding. More concerning, the study identifies several forms of knowledge that are fundamentally incompatible with AI, including embodied knowledge developed through hands-on experience, encultured knowledge shaped by shared norms, and embrained knowledge involving analytical reasoning and problem-solving.

These forms of knowledge depend on real-world interaction, sensory engagement, and repeated practice, and cannot be effectively acquired through exposure to AI-generated outputs alone. The researchers warn that over-reliance on AI for thinking, interpretation, and decision-making may lead to a gradual decline in these critical capabilities. As employees begin to outsource cognitive tasks to automated systems, organisations risk creating a dependency that undermines resilience, decision quality, and ultimately long-term performance.

To mitigate these risks, the study recommends that HR leaders design work environments that preserve experiential learning and human interaction through mentoring, shadowing, and collaborative problem-solving. It also proposes the creation of “learning vaults”—protected spaces within organisations and educational settings where critical and creative skills can be developed without heavy reliance on AI. Similar in spirit to the Svalbard Global Seed Vault, these environments would safeguard essential human capabilities and ensure that employees retain the adaptive, experience-based knowledge required to sustain strong human capital in an increasingly automated world.

More information: Dirk Lindebaum et al, On the Dangers of Large-Language Model Mediated Learning for Human Capital, Human Resource Management Journal. DOI: 10.1111/1748-8583.70036

Journal information: Human Resource Management Journal Provided by University of Bath