The Specific Event

Mental health workers and policymakers are currently pushing back against the deployment of algorithmic triage systems in healthcare settings, according to recent reporting. The core complaint, as articulated by workers and their advocates, is that AI-driven triage tools are producing worse patient outcomes. Collective bargaining agreements and proposed legislation are both being used as vehicles to install safeguards. This is a specific, live policy conflict, not a hypothetical debate about AI ethics. Workers are at the table arguing that algorithmic decision-support is actively harming the people it was designed to help.

The story has been framed primarily as a labor rights issue, or alternatively as a patient safety issue. Both framings are defensible. But neither one names the coordination problem that is actually driving the harm, and until that problem gets named correctly, the safeguards being negotiated will likely address symptoms rather than causes.

What Is Actually Failing Here

The failure is not that the algorithms are making bad decisions in some isolated, technical sense. The failure is a coordination failure between workers and the systems they are expected to operate with. Kellogg, Valentine, and Christin (2020) documented this class of problem in their review of algorithms at work: the introduction of algorithmic management into professional contexts does not simply automate routine judgment, it restructures the entire competence landscape within which workers operate. The workers who functioned effectively under prior arrangements find that their accumulated expertise maps poorly onto the new arrangement.

Mental health triage is a particularly sharp case of this because the prior competence was irreducibly relational and contextual. A clinician doing intake assessment was drawing on structural knowledge about how presentations relate to diagnoses, how presentation severity changes under different conditions, and how to weight competing signals in real time. Algorithmic triage systems encode a different structural logic, one derived from population-level statistical patterns rather than the relational schema a trained clinician develops through practice. When workers say the system is hurting patients, they are, in part, reporting a schema mismatch. Their adaptive expertise - what Hatano and Inagaki (1986) distinguished from routine expertise precisely because it enables flexible response to novel cases - is being overridden or ignored by a system that cannot recognize when a case is genuinely novel.

The Awareness-Capability Gap in a Healthcare Setting

There is a secondary failure layered on top of this schema mismatch, and it is the one my research framework is most directly concerned with. The workers pushing for safeguards clearly have awareness that the algorithm is producing poor outcomes. They can observe the mismatch. What they are not being given is the structural knowledge required to engage with the algorithmic system effectively, to understand which features of a case the triage tool weights heavily, where its decision boundaries are, and when its outputs should be treated as informative versus unreliable. This is the awareness-capability gap operating in a high-stakes setting: the workers know something is wrong but lack the schema to diagnose what is wrong precisely enough to correct for it in practice.

The legislative and bargaining responses being proposed appear, based on available reporting, to be oriented toward human override mechanisms and mandatory review requirements. These are procedural solutions. They give workers the authority to contradict the algorithm but not the structural knowledge to know when contradiction is warranted. Schor et al. (2020) would recognize this pattern: the platform relationship creates dependence not just on the platform's outputs but on the platform's logic, and workers who cannot access that logic are in a precarious epistemic position regardless of what their formal authority allows.

The Deeper Organizational Theory Problem

What is happening in mental health triage deployment illustrates a point that organizational theory has been slow to absorb. The introduction of algorithmic coordination into a professional setting is not simply a technology adoption event. It is a coordination regime change. The new regime presupposes a different competence distribution, and if that competence is not built into the workforce before deployment, the resulting variance in outcomes will look like individual performance differences when it is actually a structural gap. Power-law distributions of patient outcomes across different clinical sites using the same triage tool would not be surprising under this analysis. And they would not be correctable by giving workers more override authority.

The correct intervention is schema induction: structured training that gives workers an accurate model of how the triage algorithm reasons, what it optimizes for, where its training data came from, and what kinds of cases fall outside its competent range. This is not the same as teaching workers to game the system, and the distinction matters. Gentner's (1983) structure-mapping theory predicts that workers who understand the relational structure of the algorithmic logic will be able to transfer that understanding to novel patient presentations, while workers trained only in procedural override protocols will be limited to the specific cases those protocols anticipated. The safeguards being negotiated right now are almost entirely procedural. That is the gap worth naming.

Why This Matters Beyond Healthcare

Mental health triage is an unusually visible case because the stakes are high and the workers have organized effectively enough to generate policy attention. But the same coordination failure is present in any professional context where algorithmic decision-support has been layered onto an existing practice without attention to the competence regime the algorithm presupposes. The workers are right that something is wrong. The policy conversation needs to move from safeguards against bad outputs to the prior question of what structural knowledge workers need to function as genuine partners in an algorithmically-mediated system rather than as supervisors of outputs they cannot reliably evaluate.

Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive Science, 7(2), 155-170.

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.

Schor, J. B., Attwood-Charles, W., Cansoy, M., Ladegaard, I., and Wengronowitz, R. (2020). Dependence and precarity in the platform economy. Theory and Society, 49(5-6), 833-861.