EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

806. Conversational Decision Support Systems: a telling-to-modeling AI-enhanced approach

Invited abstract in session TB-45: Artificial Intelligence and Machine Learning for Decision Support, stream Decision Support Systems.

Tuesday, 10:30-12:00
Room: 30 (building: 324)

Authors (first author is the speaker)

1. Mariusz Kaleta
Inistitute of Control & Computation Engineering, Warsaw University of Technology

Abstract

In the full cycle of the Operations Research methodology, artificial intelligence is most prominently utilized in the algorithmic aspect. Meanwhile, modeling is traditionally considered an art exclusive to human being experts. What if we could simply tell the computer the problem and, in return, receive an optimization model and its solution? We propose a new class of Decision Support Systems (DSS), termed Conversational Decision Support Systems (C-DSS). These systems, empowered by agents based on large language models, perform the art of modeling and elevate human-machine interaction to the interface of problem definition and mathematical modeling, using natural language as the primary mode of communication. In our concept, a decision-maker interacts with the C-DSS by discussing the problem to obtain decision proposals.

First, we analyze how general-purpose language models handle problems that can be modeled as mixed-integer problems, requiring various modeling techniques such as linearization of convex/concave functions, logical conditions, and indicator constraints, among others. Additionally, we enhance the general-purpose language model with specialized knowledge using the Retrieval Augmented Generation method. The generated mathematical programming models require verification, which can also be facilitated by another language model-based agent. Therefore, we propose an architecture of C-DSS powered by several specialized language model-based agents.

Keywords

Status: accepted

Back to the list of papers