Does AI enter strategy where reasoning is deepest or most measurable?

This introductory essay to a Strategy Science special issue on AI and strategy states its "core insight" directly: "AI will not enter strategy where required reasoning is deepest but where its performance is most measurable." The authors (Csaszar, Lee, Zemsky, and Zenger) build this claim on a "dual-ladder framework: a causal ladder that maps the cognitive hierarchy of strategic tasks and a delegation ladder that specifies when organizations will grant AI autonomy over those tasks." The two ladders are deliberately separate instruments — one asks what strategizing requires, the other asks what organizations will trust AI to do regardless of how it got there.

The causal ladder, extending Pearl's ladder of causation to strategy, runs from Level 1 (analytic-predictive: "what should I do given that I observe X") through Level 2 (intervention-oriented: "what happens if I do X") up to levels the excerpt names but does not detail, which require counterfactual reasoning and new causal-model construction beyond observed data. At Level 1, the essay argues prediction alone can suffice for action — "the strategist need not know the precise causal mechanism behind sales growth; simply recognizing forecast growth is sufficient" — which is exactly the register where the essay says "AI is particularly powerful." The delegation ladder, by contrast, is "a performance-first view" asking "when organizations will trust AI with strategic discretion regardless of how the AI arrives at its outputs." Whether the two ladders move together — AI gaining strategic discretion as it gains causal reasoning — or apart — AI gaining discretion through means other than causal reasoning — is, the authors say, "an empirical question that the field must answer," and they hold both ladders open because they do not yet know which premise is correct. McBride's companion piece in the same issue, cited here, maps associative intelligence onto Level 1 and causal intelligence onto Level 2, and characterizes current AI as "highly capable associative reasoners... yet limited causal discrimination and minimal interventional capacity" — the gap the delegation ladder has to cross if it is to track causal depth at all.

The essay's own account of what humans keep — "defining which questions to ask, identifying relevant performance dimensions, judging which forecasts matter" at Level 1, and in the conclusion, "specifying objectives and constraints, adjudicating among competing causal models... bearing accountability for outcomes that cannot be delegated to algorithms" — reads as an abstract version of what How should AI agents and humans divide research tasks? reports concretely from one R&D project: AI proposes and executes, humans retain the final call. That paper's finding supplies a data point for the "should" side of this essay's "can/could/should" framing, even though the two differ in domain. The essay's ease with Level 1 — that forecasting without causal mechanism is "sufficient" for action — also sits in tension with Why do accurate predictions lead to poor decisions?, which formalizes exactly why models built to fit data do not thereby support good decisions; that note qualifies how comfortably "measurable performance" at Level 1 should be equated with good strategic choice.

As an introductory essay, this excerpt asserts the measurability claim and the convergence/divergence question; it does not test either. The special issue's empirical contributions are gestured at (prediction tasks in acquisition decisions, AI-generated strategic narratives) but their results are not given in full here, so the central insight stands as a framework for organizing future research rather than a demonstrated finding. What follows at the strength the excerpt supports: a sharper, falsifiable hypothesis about where AI delegation in strategy will expand next (toward measurable-performance tasks first) rather than a settled account of what AI can or should do in strategic decision making.