Human and AI Decision Making in a Game of Innovation and Imitation
Author
Simmering, Paul Ferdinand
Term
4. term
Publication year
2018
Submitted on
2018-02-05
Pages
95
Abstract
Dette speciale undersøger, hvordan AI indgår i ledelsesbeslutninger med et forretningsspil som afprøvningsramme. Seks personer konkurrerede mod en AI‑agent, der kombinerede Monte Carlo Tree Search (en søgemetode, der afprøver mange mulige fremtidige træk) med et neuralt netværk, som forudsiger udfald. Under spillet tænkte deltagerne højt for at afdække deres ræsonnementer. Studiet fandt strukturelle paralleller i menneskers og AI’ens tankeprocesser, men også kvalitative forskelle: fordi AI’en ikke tog højde for gensidighed (den forventede give‑og‑tage‑dynamik mellem parter), traf den andre valg end mennesker. Resultaterne peger desuden på mulige problemer med værdi‑overensstemmelse, hvor AI’ens mål ikke fuldt ud matcher menneskelige mål. Samlet peger fundene på både fælles træk og betydningsfulde forskelle, når AI bruges til at understøtte ledelsesbeslutninger.
This thesis examines how AI participates in managerial decision‑making using a business game as a testbed. Six people competed against an AI agent that combined Monte Carlo Tree Search (a method that probes many possible future moves) with a neural network that predicts outcomes. While playing, participants spoke their thoughts aloud to reveal their reasoning. The study observed structural parallels between human and AI thought processes, but also qualitative differences: because the AI did not account for reciprocity (the give‑and‑take people often expect), it made different choices. The results also point to potential value‑alignment issues, where the AI’s objectives may not fully match human goals. These insights highlight both overlap and gaps that matter when using AI to support management decisions.
[This abstract was generated with the help of AI]
Keywords
Documents