[Research Seminar] “Talking Terms: Agent Information in LLM Supply Chain Bargaining”
Speakers: Sam KIRSHNER
University of New South Wales
——————-
Thursday, September 19th, 2024
10:30–12:00 in L712 & on Zoom
We investigate the performance of large language models (LLMs) in autonomous
supply chain contract negotiations.Our objectives are to determine whether
LLMs behave likehumans, a critical factor for adoption, and to explore howfirms
strategically manage information when deploying thesebots. To address these
objectives, we conduct twoexperimental studies, where GPT assistants act as
participantsin a supply chain bargaining scenario. Study 1 examines theimpact of
information asymmetry by comparing outcomesunder public and private
supplier cost scenarios. The findingsshow that LLMs, while mimicking human
behavior, areinfluenced by the availability of information, leading toenhanced
supply chain efficiency but less fairness in profitdistribution compared to human
negotiators. In Study 2, weconsider information interventions for the supplier’s
agent,such as cost deception, to assess their effects on negotiationoutcomes.
Interestingly, cost deception improved supplierprofits at the expense of retailers
and overall supply chainefficiency. These results further highlight fairness
andefficiency trade-off and raise novel ethical implications of theimpacts of AI
deception. Beyond offering timely insights intocutting-edge AI integration into
supply chains, our approachand findings provide many new avenues for research
thestrategic and ethical dimensions of LLMs.