Inside Baseball: The Automated Ball-Strike System as an Object Lesson in Technological Rule Enforcement

URL SCAN: Inside Baseball: The Automated Ball-Strike System as an Object Lesson in Technological Rule Enforcement

FIRST LINE: "Clearly-defined rules are often assumed to be straightforward to automate and evaluate."

THE DISSECTION

This is an STS (Science and Technology Studies) autopsy of MLB's Automated Ball-Strike System, submitted to the FAccT (Fairness, Accountability, and Transparency) discourse. The authors document that automating a "clearly-defined" rule—calling balls and strikes—took seven years and still involved contested translation between rulebook and operational reality.

The paper's core argument: evaluation frameworks that measure distance between a formalized rule and its technological implementation are analytically bankrupt. The "ground truth" of the strike zone was always a hybrid artifact—rulebook definition plus umpire enforcement judgment. Technology doesn't discover the rule; it must construct a version of it, navigating stakeholder conflicts (pitchers, hitters, umpires, broadcast interests, pace-of-game concerns).

THE CORE FALLACY (RELATIVE TO DT)

The authors believe they are examining a technical translation problem that reveals social complexity. They are wrong about the direction of causation. The complexity is not revealed by technology—it is produced by the attempt to operationalize contested human standards. The seven-year struggle is not a cautionary tale about implementation patience. It is empirical confirmation that formalized human rules encode implicit judgment that cannot be cleanly extracted and delegated to algorithmic systems.

This has direct implications for the Discontinuity Thesis. DT's mechanics assume AI systems can automate cognitive work at scale. This paper demonstrates that "cognitive work" in the rule-enforcement sense was never as clean as it appeared. Every domain humans thought was "objective" (strike zones, credit risk, diagnostic criteria) involves the same contested hybrid: formal definition + interpretive enforcement. AI will not simply "do it better"—it will have to construct its own contested version and impose it.

HIDDEN ASSUMPTIONS

1. Automatable in principle: The paper assumes the project of automation is correct; only the methodology of evaluation needs fixing. It never asks whether large-scale rule enforcement can be cleanly delegated to technical systems at all.
2. Stakeholder legitimacy: The paper treats "balancing stakeholder values" as a solvable engineering problem. It is, in fact, a permanent site of political conflict that technical systems cannot resolve—only displace.
3. The umpire as noise: Implicit framing positions human umpires as the source of inconsistency to be engineered out. But the paper's own evidence shows umpires were the * mechanism* through which the rule survived contact with physical reality. Removing them doesn't create objectivity; it creates a different contested artifact.
4. Convergence assumption: The paper implies that with sufficient social science attention, these systems can be "experienced well" in practice. It does not interrogate who experiences what, and at whose expense the necessary trade-offs are made.

SOCIAL FUNCTION

This paper is transition management. Specifically, it is a contribution to the FAccT community's project of making AI governance legible and tractable—managing the social legitimacy of algorithmic systems before they produce mass-level disruptions. It performs the intellectual work of "we can fix this" while the underlying mechanism (contested ground truth resisting clean automation) proceeds regardless.

It is not copium exactly. The authors are empirically rigorous. But the framing—which asks for better evaluation of sociotechnical systems—accepts the premise that these systems will be deployed at scale and focuses on optimizing their social reception. This is elite technician classwork: making automation more palatable to those it displaces or subjugates.

THE VERDICT

Under DT mechanics, this case study reveals a critical lag mechanism: the seven-year struggle is not an anomaly. It is the normal timeline for automating any human-standard domain that encodes contested judgment. This lag will play out across every cognitive domain—legal judgment, medical diagnosis, credit allocation, hiring, content moderation—with the same contested "ground truth" dynamics.

The implication: the displacement of human judgment by AI systems will be far slower, messier, and more politically volatile than DT's timeline assumes if the relevant institutions resist or contest the automation. Where they do not resist—where the ground truth is less contested or the displaced parties have no leverage—the transition will be swift.

The paper is empirically valuable as confirmation that automation of human-standard domains is not a straightforward engineering problem. It is a political and interpretive problem that technical systems must create a resolution for, not simply discover. The Sovereign class in the DT framework will be those who control that resolution—who decide what "the strike zone" actually is when the umpires are gone.

Survival Implication: Anyone whose economic function depends on exercising judgment within human-standard domains (law, medicine, finance, journalism) should note: your "obsolescence" is not guaranteed to be immediate. The contested ground truth problem means your domain will go through a messy translation period. Use it. Build leverage during the lag. The question is not whether the rule gets automated—it is who gets to define the automated version.