Trustworthiness of AI in Planning and Scheduling: The Experience of the TUPLES Project

Abstract

Artificial Intelligence and hybrid approaches are increasingly relevant in Decision Science at both the scientific and industrial application levels. While symbolic approaches enjoy well-established techniques to verify their behaviour, data-driven or hybrid approaches can pose a risk in terms of robustness under uncertain conditions, and their behaviour can be less predictable and explainable in unexpected conditions. To address this challenge, the European Commission presented in 2019 the Ethics Guidelines for Trustworthy AI [1], and included in the Horizon Europe 2020–2024 programme a financing of 96 million euros for Trustworthy, Human Centred and Responsible AI. Amongst other initiatives, the TUPLES project [2] was initiated to develop the foundations, approaches and tools needed to achieve transparent, robust, and safe algorithmic solutions in Planning and scheduling (P&S), and increase confidence in these systems to accelerate their adoption. TUPLES is planned to release a toolkit including a set of simulation environments, a self-assessment tool to validate the confidence level of a P&S application, and will launch a competition in late 2024 to engage a wider audience and increase its impact [3]. We outline the challenges, approaches and objectives of the project, focusing on the use cases provided by OPTIT in the application sectors of Energy [4] and Waste management [5]. In doing so, we provide a concrete perspective on how Human Centred AI (with a specific focus on robustness and explainability) requires innovative strategies and approaches that will pave the way for significant improvements in the Decision Support Systems of the coming decades.