Operations Research 2025
Abstract Submission

2386. Artificial human intelligence: Replicating human decision making for a driving assistant by combining knowledge engineering, computer vision and rule-based inference

Invited abstract in session WC-12: Explainability and Interpretability in Optimization, stream Artificial Intelligence, Machine Learning and Optimization.

Wednesday, 13:30-15:00
Room: H10

Authors (first author is the speaker)

1. Robert Maximilian Neumann
Lehrstuhl für Wirtschaftsinformatik und Operations Research, Martin-Luther-Universität Halle-Wittenberg
2. Taieb Mellouli
Business Information Systems and Operations Research, Martin-Luther-University Halle-Wittenberg

Abstract

A vast majority of modern AI-based approaches to decision making create a so called ‘blackbox’ where the inputs lead to a result with little to no understanding why. This is a common design choice for AI in complex problems, making the learning algorithm learn the desired outcome all the way. We differ from this common approach by splitting the decision making into two phases and using knowledge engineering to define relevant information (features) and a set of rules based on them in order to replicate the decision process made by humans more closely. We apply our concept within a driving assistant whose task is the determination of the adequate speed given an image (and some timeconsistent information).

The foundation of our approach are the rules and the features they require, which we acquired using knowledge engineering, input from human experts and the german traffic law (STVO). In the first phase our approach creates a vector of these abstract features that represent the current situation given in an input image. For this task we use state of the art computer vision algorithms. In the second phase we apply our defined rules to deduce a speed recommendation. The deduction process is split into two parts where the first updates the timeconsistent knowledge base which is then used to infer the actual recommended speed. This approach is inspired by the research of Walther and Mellouli and will in future research be extended by learning algorithms to analyze and enhance the rule set used for decision making.

Keywords

Status: accepted

Back to the list of papers