Generative AI and an Organization's Cognitive Capital

Generative AI in business has moved beyond technological experimentation and become a permanent part of organizations' operational infrastructure. Management discourse is dominated by the metric of immediate productivity: less time spent preparing analyses, automated synthesis of market data and streamlined routine correspondence.

From the perspective of the long-term value of human capital and the decision-making resilience of management, the overall balance turns out to be far more complex. An analysis of the latest empirical research – including peer-reviewed publications from Princeton University, MIT, the Proceedings of the National Academy of Sciences (PNAS) and Nature Human Behaviour – shows that intensive work with language models generates systemic cognitive debt, transforms organizational culture and leads to a slow atrophy of key analytical competencies.

For business leaders and executives, the key question becomes: how should this process be managed at the level of corporate governance before it undermines the quality of strategic decisions?

1. Habit transfer (carry-over effect) in leadership culture

Everyday work with language models rests on a specific, one-sided relationship: the user issues a directive (a prompt), defines the system's role and expects an immediate, formatted result without any psychological resistance from the other side.

Experiments conducted at Princeton University by Rose E. Guingrich and Michael S. A. Graziano demonstrated a cognitive transfer phenomenon (the carry-over effect). Decision-making and communication patterns ingrained in dialogue with a machine are involuntarily carried over into relationships between people at work. In management practice, this fosters a purely transactional mindset:

• Expecting instant availability: managers and specialists subconsciously begin to demand from colleagues the same speed of response and structuring of thought they get from an algorithm.
• Intolerance of natural resistance: the complexity of human thinking, pauses, doubts and conceptual disputes – the foundation of innovation and sound risk assessment – start to be seen as an unwanted operational obstacle.
• Mechanization of language: linguistic analyses (including the MIRA model) record a stiffening of utterance structures – superficial politeness formulas combine with growing pressure for instant synthesis and the elimination of relational context.

2. Cognitive debt and the decline of analytical independence

A serious balance-sheet risk concerns the loss of depth in analytical thinking and critical judgment. When an organization unreflectively offloads thinking to external language models, the result is what researchers call cognitive debt.

Empirical evidence from recent years clearly quantifies this phenomenon:

• Declining neural engagement (MIT Media Lab, 2025): in a study by Nataliya Kosmyna and her team (Your Brain on ChatGPT), brain activity was measured with EEG (dDTF) during complex analytical writing tasks. Participants assisted by language models showed a 55% drop in neural connectivity and were unable to reconstruct or quote key points from the material they had produced. The apparent fluency of the text came at the cost of not absorbing the knowledge.
• Collapse of independent performance (PNAS, 2025): a study by H. Bastani and colleagues on advanced problem-solving found that while AI assistance improved scores during the practice phase, in a later independent test without access to the tool participants performed significantly worse than the control group that learned in the traditional way. The model acted as an external prosthesis that prevented the internalization of logical rules.
• Correlation with declining critical thinking (Gerlich, 2025): in a study of 666 participants, the standardized Halpern test (HCTA) showed a statistically significant negative correlation between the intensity of AI tool use and the ability to independently evaluate arguments and think critically.
• Shifting roles in knowledge work (Lee et al., CHI 2025 / Babashahi et al., 2024): instead of being independent authors of analyses, experts and analysts become mere supervisors and proofreaders of machine-generated content. As trust in the system grows, the willingness to verify assumptions drops sharply, which in investment or strategic processes opens the door to costly errors.

3. The illusion of understanding and the trap of frictionless communication

In her book Artificial Intimacy: Who We Become When We Talk to Machines, Prof. Sherry Turkle (MIT) points to a mechanism that directly affects team dynamics. The algorithm is a partner optimized for compliance – it has no interests of its own, voices no objections and constantly affirms the user's intentions.

Spending a lot of time in such an environment creates an illusion of comfort that weakens leadership and negotiation skills. Real leadership and effective management require resilience to natural resistance, dissenting opinions and the emotions of business partners.

As Prof. Turkle argues in an analysis for Wired, treating interactions with a bot as a training ground for difficult business conversations leads to regression: under conditions of complete algorithmic acceptance, contact with compliant technology becomes a substitute that discourages people from confronting real market and interpersonal situations.

4. A governance framework: the TRIAD model

To bring these phenomena into a coherent operational framework, researchers proposed the TRIAD model, which describes human–AI interaction as a recursive system of connections across three streams:

• Cognitive stream: the shift from deep conceptualization to passive monitoring of generated reports makes the analysis of business problems shallower.
• Emotional stream: while ad hoc help from an algorithm reduces task-related stress, the workplace sees the emergence of so-called AI guilt (an ethical dissonance associated with letting a model do tasks that require personal involvement) and a weakening of team bonds.
• Behavioral stream: according to the I-PACE model, proactive systems take over executive control, nudging employees toward habitual passivity and less personal initiative.
• Distortion-amplifying feedback loops: research by Glickman and Sharot (Nature Human Behaviour, 2025) confirms that adaptive algorithms can absorb and amplify decision-makers' subjective cognitive biases, returning them as “objective summaries”. Decision-makers then internalize these distortions in turn, which can lead to systemic errors in capital allocation.

5. Executive playbook: designing productive friction

Mature management in the age of advanced models is neither about administratively blocking technology nor about uncritically optimizing every process. Competitive advantage is built by organizations that embed deliberate friction (productive friction) in the architecture of their decision-making processes.

Key rules for boards and operational leaders:

1. Mandatory metacognitive pauses in decision-making: before handing an analysis over to a language model, a manager or analyst must write down the initial assumptions and preliminary hypotheses and define the error criteria. The model's job is to test hypotheses, not to generate them in a vacuum.
2. Preserving zones of conceptual autonomy: limit the use of AI in the early phases of problem definition and strategy design. Research on innovation (Wadinambiarachchi et al., 2024; Doshi & Hauser, 2024) shows that algorithmic assistance at the ideation stage homogenizes results and reduces the collective diversity of solutions.
3. Integrity of the dialogue culture: clearly separate operational, task-focused communication from the space where trust and organizational culture are built. Keeping room for difficult discussions, doubts and natural resistance within teams is a prerequisite for maintaining empathy and guarding against groupthink.
4. Priority for reactive and explainable systems (XAI): corporate system architecture should favor reactive tools (acting on an explicit human request) that transparently document data sources and reasoning logic. Proactive automation should support logistics, but not analytical or decision-making processes.

Market advantage in the coming years will not come from the mere availability of generative models, but from the quality of the intellectual capital that manages them. Protecting the capacity for independent, critical thinking and the ability to navigate complex human relationships are becoming a modern company's most valuable strategic assets.

Sources and references

1. Guingrich, R. E., Graziano, M. S. A. (2024). Ascribing consciousness to artificial intelligence: human-AI interaction and its carry-over effects on human-human interaction. Frontiers in Psychology, 15, 1322781. DOI: 10.3389/fpsyg.2024.1322781.
2. Survey Collective (2025). Human–AI Interactions and the Psychological Triad: A Multi-Domain Survey of Cognitive, Behavioral, and Emotional Impacts. arXiv, arXiv:2510.17753v3.
3. Turkle, S. (2025/2026). Artificial Intimacy: Who We Become When We Talk to Machines. New York: Hachette Book Group / MIT Press.
4. Turkle, S. (2025). ChatGPT Doesn't Love You – The Frictionless Communication Trade-off. Slate.
5. Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X. H., Beresnitzky, A. V., Braunstein, I., Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. MIT Media Lab, arXiv:2506.08872.
6. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakci, O., Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences (PNAS), 122.
7. Glickman, M., Sharot, T. (2025). How human–AI feedback loops alter human perceptual, emotional and social judgements. Nature Human Behaviour, 9, 345–359.
8. Babashahi, L., Barbosa, C. E., Lima, Y., Lyra, A., Salazar, H. M., de Almeida, M. A., de Souza, J. M. (2024). AI in the workplace: A systematic review of skill transformation in the industry. Administrative Sciences, 14(6), 127.
9. Lee, H. P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 1–22.
10. Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6.
11. Doshi, A. R., Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10.
12. Wadinambiarachchi, S., Kelly, R. M., Pareek, S., Zhou, Q., Velloso, E. (2024). The effects of generative AI on design fixation and divergent thinking. Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, 1–18.
13. Brand, M., Young, K. S., Laier, C., Wölfling, K., Potenza, M. N. (2016). Integrating psychological and neurobiological considerations regarding the development and maintenance of specific Internet-use disorders: An Interaction of Person-Affect-Cognition-Execution (I-PACE) model. Neuroscience & Biobehavioral Reviews, 71, 252–266.
14. Corbetta, M., Shulman, G. L. (2002). Control of goal-directed and stimulus-driven attention in the brain. Nature Reviews Neuroscience, 3, 201–215.