What the Data Actually Shows

SurveyMonkey released findings this week from what they are calling a "curiosity report," and the headline result deserves careful attention: AI tools are making individual work faster while simultaneously weakening critical thinking, collaboration, questioning, and original idea generation across teams. This is not a speculative warning about future displacement. It is a measurement of something happening right now inside organizations that have already deployed AI at scale. The finding is uncomfortable precisely because it runs counter to the dominant narrative that productivity gains from AI adoption are straightforwardly positive.

The report's framing around "curiosity" is itself telling. Curiosity is a proxy for the generative cognitive behaviors that produce novel outputs, the asking of questions that do not yet have answers, the willingness to explore rather than confirm. What SurveyMonkey appears to have measured is the organizational cost of offloading that function to a tool. Speed increases; generativity decreases. That tradeoff is not random noise. It maps onto a structural dynamic that my dissertation research addresses directly.

The Awareness-Capability Gap at Organizational Scale

My research on the Algorithmic Literacy Coordination (ALC) framework focuses primarily on platform workers, but the SurveyMonkey findings extend the relevant phenomenon into a different organizational context. The awareness-capability gap I study at the individual level - where workers develop awareness of how algorithms function without developing the capacity to respond to them effectively - appears to have an organizational analog. Firms deploy AI tools, their employees become aware that those tools are shaping outputs, and yet the organization as a whole does not develop the structural understanding needed to use those tools without degrading the competencies that made their outputs valuable in the first place.

This is not the same as technological deskilling in the classical sense (Braverman, 1974). The concern is more specific. Hancock, Naaman, and Levy (2020) introduced the concept of AI-mediated communication to describe interactions where AI systems filter, generate, or evaluate the messages humans exchange. Their framework anticipates exactly the kind of outcome SurveyMonkey measured: when AI mediates the communicative work of questioning and ideation, the human capacity for that work atrophies at a rate that exceeds what individual awareness of the tool can compensate for. Workers know they are using AI. They cannot fully observe what they are losing by doing so.

Folk Theories Versus Structural Schemas in Organizational Deployment

The distinction I draw in my dissertation between folk theories and structural schemas is relevant here. A folk theory of AI adoption at the organizational level sounds like this: AI handles repetitive tasks, freeing workers for higher-order thinking. That is the story most organizations tell themselves when they deploy these tools. The SurveyMonkey data suggests this folk theory is wrong in practice, not just in edge cases. The tasks being offloaded are not merely repetitive. They include the low-stakes generative behaviors, drafting a rough question, sketching an underdeveloped idea, attempting a synthesis that might fail, that build the cognitive infrastructure for higher-order output over time.

A structural schema of the same deployment would ask a different set of questions. What is the topology of cognitive labor being redistributed, and which of those redistributions are reversible? Kellogg, Valentine, and Christin (2020) showed that algorithmic systems at work do not simply automate tasks; they reorganize the social and cognitive structure of work itself. The SurveyMonkey findings are consistent with that framework. The problem is not that AI tools are present. It is that organizations lack accurate structural schemas for understanding what those tools are doing to the distribution of cognitive work across their teams.

Why Procedural Training Makes This Worse

The counterintuitive prediction from ALC theory is that general schema-inductive training produces better transfer outcomes than platform-specific procedural training, even when procedural training produces faster initial performance. The organizational AI deployment context is a direct test of this prediction at scale. Most enterprise AI training programs I have reviewed are procedural by design: here is how to use this tool, here is the correct prompt format, here is the workflow integration. These programs increase speed. They do not build the structural understanding that would allow workers to recognize when offloading a cognitive task to AI is degrading rather than augmenting their output.

Hatano and Inagaki (1986) distinguished between routine expertise, which performs well within familiar parameters, and adaptive expertise, which transfers across novel configurations. Procedural AI training produces routine expertise with a specific tool. It does not produce the adaptive expertise needed to evaluate when that tool should not be used, or when its outputs require the kind of critical scrutiny that the SurveyMonkey data suggests workers are becoming less inclined to apply. The speed gains are real. The question is what organizations are purchasing that speed with, and whether they have accurate instruments for measuring the cost.

References

Hancock, J. T., Naaman, M., and Levy, K. (2020). AI-mediated communication: Definition, research agenda, and ethical considerations. Journal of Computer-Mediated Communication, 25(1), 89-100.

Hatano, G., and Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, and K. Hakuta (Eds.), Child development and education in Japan (pp. 262-272). Freeman.

Kellogg, K. C., Valentine, M. A., and Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.