The Cognitive Offloading Problem No One in IS Is Studying

Five hundred and ten citations in its first year. That is the number that stopped me. I need to be careful here because I do not have Gerlich (2025) locally and cannot verify the specific findings, but my recollection is that the study found a significant negative relationship between frequent AI tool use and critical thinking scores, with the mechanism being cognitive offloading: the systematic substitution of AI-generated answers for independent reasoning. The speed of that citation count tells you something about how hard the finding hit. Researchers across disciplines recognized the pattern immediately. We have all watched it happen. The question I kept asking after I read the abstract was why the IS field, which has the theoretical machinery to explain this, has said almost nothing about it as an organizational problem.

Lee et al. (2025) ran experiments suggesting that generative AI reduces both cognitive effort and confidence in one's own judgment. Again, I have not verified these findings locally, so I need to hedge. But the direction of the result is consistent with what I have seen in the literature I do have. Kattnig et al. (2024) documented automation bias in AI-assisted decision making. People followed the automated recommendation even when a reliable source told them it was wrong. Participants without the AI aid outperformed those who had it. Strich et al. (2021) found that substitutive AI deskills qualified employees while allowing less qualified ones to mask inexperience. I wrote about that pattern in what happens when AI deskills the people who need it most. These findings line up. They tell the same story from different angles. The tool helps in the short run. The user's capacity degrades in the long run. And the user does not notice because the output looks good enough.

Cognitive offloading is the mechanism. It is not the same as automation bias, though they overlap. Automation bias is the tendency to trust automated output regardless of its validity. Cognitive offloading is the broader pattern of delegating cognitive work to an external system and then failing to re-engage that work when the system falters. You offload the diagnosis to the model. You offload the analysis to the assistant. You offload the writing to the generator. Each offload saves time and feels productive. Each one also reduces the frequency with which you exercise the judgment, the pattern recognition, the diagnostic reasoning that built your expertise in the first place. This is not a claim about laziness. It is a claim about atrophy. Muscles that are not used get weaker. Cognitive skills that are not exercised degrade.

I think what makes this an organizational problem rather than just an individual one is absorptive capacity. Cohen and Levinthal (1990) defined it as the organizational ability to recognize the value of new external information, assimilate it, and apply it to commercial ends. The critical mechanism is prior related knowledge. You cannot absorb what you cannot recognize, and you cannot recognize it without prior knowledge in that domain. Absorptive capacity is path-dependent and cumulative. Organizations that invest in building knowledge can absorb new knowledge faster. Organizations that never built the knowledge base cannot absorb new information regardless of how much they spend. I traced this mechanism in why expensive platforms fail to deliver value.

Here is where cognitive offloading meets absorptive capacity. When professionals across an organization offload diagnostic judgment, analytical reasoning, and evaluative decisions to AI systems, they stop exercising the very cognitive skills that constitute the organization's prior knowledge base. The individual atrophy compounds at the organizational level. The firm that once could evaluate whether an AI recommendation was sound gradually loses that collective capacity. The prior knowledge does not vanish overnight, but it degrades, and Cohen and Levinthal's path dependence works in both directions. Just as investing in an area of expertise builds absorptive capacity that makes future learning faster, disinvesting in that expertise forecloses the development of capability in that area. AI offloading is disinvestment by another name.

March (1991) gave us the vocabulary for thinking about how organizations allocate effort between exploitation and exploration. Exploitation refines what you already know how to do. Exploration builds new capabilities. The returns to exploitation are immediate and visible. The returns to exploration are distant and uncertain. March warned that organizations systematically favor exploitation, and that this favoritism creates a competency trap: the organization gets better and better at what it currently does and loses the ability to adapt when what it does becomes insufficient. Cognitive offloading accelerates the competency trap in a specific way. It turns cognitive exploration into simulated exploitation. The AI does the exploring. The human receives the output and executes on it. The organization records a productivity gain. The exploration capacity atrophies.

Frenkenberg and Hochman (2025) also appear to address cognitive offloading in the context of human-AI collaboration, though again I cannot verify the specifics locally and must hedge. What I can say from the sources I have is that the mechanism is well-established across multiple studies now. Offloading cognitive work to AI reduces the exercise of critical thinking. Reduced exercise degrades expertise over time. Degraded expertise reduces the capacity to evaluate AI output. Reduced evaluation increases over-reliance. The loop tightens with each cycle.

This is where I think the IS field is missing something. We have absorptive capacity for explaining why organizations fail to absorb new knowledge. We have automation bias research for explaining why people trust systems they should not. We have deskilling research documenting that expertise erodes when it is not exercised. Benbya, Pachidi, and Jarvenpaa (2021) acknowledged that automation can render experts redundant or deskill them. Berente et al. (2021) named deskilling and labor substitution as major ethical frontiers. Kattnig et al. (2024) gave us the experimental evidence for automation bias. Strich et al. (2021) showed that AI deskills the qualified while enabling the unqualified to mask. Gopal et al. (2025) warned that accepting AI-suggested constructs without justification optimizes metrics over meaning. What we do not have is a unified account of how these mechanisms combine when cognitive offloading is treated as an organizational-level absorptive capacity problem.

Cohen and Levinthal defined absorptive capacity at the firm level, and they were explicit about this. Individual learning is necessary but not sufficient. The firm-level version depends on shared routines, communication structures, and diversity of distributed expertise. This is precisely what cognitive offloading attacks. When the organization's analysts, strategists, and decision-makers stop doing the cognitive work themselves and start receiving AI-generated answers, the shared routines for critical evaluation degrade. The communication structures that once conveyed nuanced judgment between experts atrophy because the AI short-circuits the need for those conversations. The diversity of expertise that Cohen and Levinthal identified as a driver of absorptive capacity narrows because the AI output homogenizes the reasoning process. Everyone is using the same model. Everyone is receiving structurally similar recommendations. The organization does not just lose the capacity to absorb new knowledge. It loses the diversity of perspective that made absorption possible in the first place.

Zahra and George (2002) reconceptualized absorptive capacity into potential and realized components. Potential is what lets you acquire and assimilate knowledge. Realized is what lets you transform and exploit it. An organization can have high potential capacity and low realized capacity, which is exactly what the productivity paradox describes. Cognitive offloading is an attack on realized capacity specifically. The organization acquires the AI output. It assimilates it. But the transformation step, the integration of new knowledge with prior knowledge to produce something novel, is where the cognitive work happens. Offloading that work to the AI means the organization never builds the realized capacity to do anything with the output except execute it as given.

The implication is uncomfortable. If cognitive offloading erodes organizational absorptive capacity, then the organizations that adopt AI most aggressively may be the ones that suffer the most significant long-term capability loss. They gain efficiency in the short run. They lose the capacity to evaluate, adapt, and innovate in the long run. The offloading feels productive. The AI spending shows up in budgets and dashboards. The cognitive infrastructure degrades quietly. By the time the organization discovers that it can no longer evaluate whether its AI systems are producing good output, the human expertise that would have made that evaluation possible has already atrophied.

I think this is the research gap. Not another study showing that AI makes people lazy. We have those. But a study that treats cognitive offloading as organizational absorptive capacity erosion and traces the mechanism from individual atrophy through routine degradation to collective capability loss. That is something IS theory is built to do. We have the construct. We have the mechanism. We have the path dependence. What we do not have is the empirical work that connects them.