The Underdetermination Problem

In the AI Story

Underdetermination is the structural feature that makes the Berkeley study simultaneously cited by the triumphalist tradition as evidence of productivity gains and by the elegist tradition as evidence of work intensification. The study documented that AI-assisted workers worked faster, took on more tasks, and expanded into new domains. The triumphalist tradition reads this as capability expansion. The elegist tradition reads it as auto-exploitation. The datum is shared. The interpretation is tradition-dependent.

Laudan was explicit that this interpretive flexibility is the central problem of inter-tradition evaluation. When traditions are flexible enough to accommodate any evidence, evidence alone cannot adjudicate between them. What can adjudicate is comparative problem-solving effectiveness over time — the sustained accumulation of solved problems and the sustained reduction of anomalies, evaluated across multiple domains and over extended periods.

The condition has direct consequences for how the AI discourse should be conducted. Appeals to empirical evidence as if it could settle the dispute — common on both sides — systematically overlook the interpretive frameworks through which the evidence is read. The triumphalist who points to adoption curves as proof of progress is not confronting the elegist; she is repeating claims the elegist tradition can absorb without modification. The elegist who points to burnout data as proof of pathology is not confronting the triumphalist; she is repeating claims the triumphalist tradition can dismiss as temporary adjustment.

The progressive response to underdetermination is not to abandon empirical investigation — the evidence still constrains what interpretations are possible — but to recognize that empirical investigation alone is insufficient. The interpretive frameworks must also be scrutinized, their conceptual tensions identified, their anomalies taken seriously. This is the work the AI discourse has largely avoided, and its avoidance is one of the most diagnostic signs that the discourse is not yet serious about evaluation.

Origin

The concept originates in Pierre Duhem's The Aim and Structure of Physical Theory (1906) and was refined by W. V. O. Quine in "Two Dogmas of Empiricism" (1951). Laudan's contribution, in Progress and Its Problems and subsequent work, was to extend the analysis from individual theories to research traditions and to specify the operational procedures by which underdetermined disputes can nevertheless be rationally evaluated.

Key Ideas

Evidence is not self-interpreting. The same data can support opposed conclusions under different frameworks.

Auxiliary hypotheses matter. A tradition can absorb disconfirming evidence by modifying auxiliary assumptions.

Sustained comparison resolves. What empirical evidence cannot resolve in a single test, comparative problem-solving can resolve over time.

Framework criticism required. Progressive inquiry must examine interpretive frameworks, not just empirical data.