New ethics paper questions outsourcing the skills sport is meant to test

The customer problem is uncertainty about how far coaching automation should go. Engage the athletic director, coaches, athlete representatives and academic leadership where educational objectives matter. Ask what expertise the program wants people to retain, which decisions AI may assist, and whether a proposed product quietly changes responsibility. A bounded engagement could map one coaching workflow and facilitate a decision-boundary workshop. The value hypothesis is clearer product fit and expectations, rather than improved win rates. This international philosophical argument supports discussion only. It does not prove that an existing tool is unfair, prohibited, effective or unacceptable to a particular sporting community.

Fit is architecture review for coaching copilots or agents. Separate retrieval, summarization, recommendation and execution permissions. Require source-linked suggestions and explicit approval before a system changes a practice plan. Prerequisites include agreed decision authority and a representative set of coaching scenarios. Test whether the interface encourages uncritical acceptance, whether users can inspect underlying evidence, and whether tool permissions match the approved scope. Compare assisted decisions with a documented human process without claiming that agreement proves sporting legitimacy. Security tests remain necessary regardless of the ethical position chosen. Cloud, on-premises and hybrid deployment do not determine whether human expertise has been displaced.

The athletics sponsor should own the decision charter, with coaches responsible for operational use and athlete representatives contributing feedback. Map existing decisions, configure permissions, train staff and rehearse escalation when recommendations conflict with professional judgment. Dependencies include access to tool logs and vendor ability to constrain actions. Governance checkpoints should precede pilot launch and any expansion of automation. Proposed acceptance criteria are documented human authority for every consequential action, zero unauthorized plan changes in scenario tests, and staff ability to explain decisions without merely repeating model output. Monitor skill development over time. Apparent efficiency may conceal overreliance; the paper offers no measured threshold for that risk.