Do cognitive forcing functions help all decision makers equally?

The same cognitive-forcing-functions experiment (N=199) reports that the intervention does not benefit all participants equally: "our results show that, on average, cognitive forcing interventions benefited participants higher in Need for Cognition more," where Need for Cognition is "motivation to engage in effortful mental activities." The paper frames this explicitly as a case of intervention-generated inequality (IGI), citing the general definition that IGIs "arise when the benefits of an intervention disproportionately accrue to a group that is already privileged in a particular context" — so that "even though everyone might benefit to some extent, the gaps between different groups increase."

The reasoning follows directly from the dual-process account used to explain the forcing functions themselves: because the interventions work by compelling costlier System 2 engagement, their effectiveness should scale with how motivated a person already is to engage in effortful thinking — the trait the paper measures as Need for Cognition. The paper treats this as grounds for a "limited internal audit" of its own designs, explicitly checking "whether our interventions benefited equally people with different levels of Need for Cognition" rather than assuming a uniform effect, and frames it as a broader methodological obligation: "as researchers, we should audit our work for emerging inequalities among relevant groups."

This is the equity-focused counterpart to Do forcing functions reduce AI overreliance without harming team performance?: that note establishes that forcing functions work in aggregate but at an acceptability cost, while this finding shows the aggregate effect conceals a split by cognitive motivation — the people most willing to exert effort are the ones the intervention helps. It parallels Does using LLMs actually improve strategic decision making? and qualifies Can AI guidance reduce anchoring bias better than AI decisions? in the same way: all three treat who benefits from an AI-assistance design as a variable to be checked, not a constant to be assumed.

The excerpt's audit is explicitly "limited" and considers only one individual-difference variable (Need for Cognition) on one Mechanical Turk sample; it does not establish which specific design features of the forcing functions interact with NFC, whether the same gap would appear for other traits (domain expertise, AI literacy, prior trust in AI), or how to redesign forcing functions to close rather than create this gap. The authors call for "further research... to account for individual differences in cognitive motivation and explore the right amount and timing of cognitive forcing" rather than offering a fix, so the finding should be read as a flagged risk for any forcing-function design, not a solved problem.