The Development of Decision Support Systems in the 1960s as Antecedent of "AI-Rationality"

Nohr's article argues that decision support systems and business simulation games developed in the 1950s and 1960s established a mechanistic, reductive conception of decision-making that later became a foundational epistemology for contemporary artificial intelligence. Its central claim is that the "decision" concept links game design, managerial training, and early computer science, and that the field's genealogy runs from post-war military logistics and RAND Corporation research through MIT's Beer Game and the SAGE air defense program. The analysis is historical and genealogical rather than quantitative, tracing how institutions framed decision-making as something to be formalized and automated.

The work's key observations concern the ambition to "de-subjectify" human judgment. Military and corporate projects treated the human operator as a replaceable switching element or black box, and cybernetic thinkers such as Wiener, Beer, and Forrester encouraged the belief that subjective choices could be rendered objective and calculable. Projects like Newell and Simon's General Problem Solver, along with RAND's System Development Corporation and Management Information Systems efforts, typically cost millions of dollars and drew large interdisciplinary teams, a level of institutional investment the article links to the persistence of this framework despite its acknowledged flaws.

The article also highlights the limits of this approach as identified by contemporaries. Mathematical game theory modeled the rational decision-maker as a manipulable variable, and the author draws on Shubik's retrospective critique and Wiemer's work on evolutionary game theory to show that these models often yielded outcomes poorly matched to human behavior, as in the prisoner's dilemma. Kirsch's 1973 criticism that such projects oversimplified motivation and drew inadmissible human-machine analogies is treated as an early warning. The more scientifically rigorous business games became, the less effective they were at training real managers.

Ultimately, the article concludes that decision research has changed little since this period, sustaining a mechanistic conception of decision-making that remains a core AI competence. It frames the shift from classical to evolutionary models as raising concerns about algorithmic authority over choices once attributed to conscious individuals, a question it presents as central to understanding contemporary AI-driven rationality.