CAREER: Answer Set Programming for Quantitative Information

University Of Nebraska At Omaha, Omaha NE

Investigators

Abstract

In today’s society, computers are ubiquitously utilized for problem-solving and decision-making, significantly impacting sectors of the economy such as healthcare, national security, and logistics. Artificial Intelligence (AI) has emerged as a pivotal tool in addressing issues too intricate for conventional methods, thereby accelerating the integration of computer-aided decision-making. In high-stakes domains such as healthcare and national security, it is essential that AI systems exhibit not only efficacy but also transparency, ensuring their reasoning processes can be comprehended and audited by humans. The AI methodologies advanced in this project will enable the creation of computer systems that are not only effective and robust but also transparent. This effort will pave the way for a more reliable application of AI systems, bolstering their use in high-stakes environments with increased confidence. Additionally, the project generates an innovative teaching method to equip the next generation of computer scientists with the skills necessary to effectively develop AI systems. Knowledge Representation and Reasoning is a specialized branch of AI concerned with computer systems capable of utilizing knowledge in a format that is human-readable, open to inspection and comprehension, and effectively used by computers. This project is centered on enhancing (i) Knowledge Representation languages and (ii) automated reasoning techniques for these languages. It aims to address the critical need for handling qualitative and quantitative data, key components in numerous real-world applications, while preserving the inherent human-readable characteristic of these systems, a critical factor for high-stakes applications. Existing methods have primarily concentrated on either employing high-level languages that are easily comprehensible to humans or on effectively managing quantitative data. This project aspires to bridge these two approaches by designing a methodology to develop systems that are both human-readable and effective in handling quantitative data. Furthermore, the project aims to enhance these systems with human-readable explanations, providing auditors with essential insights into the rationale behind the solutions generated by the computer systems. With those insights, auditors can ensure that provided solutions align with their values or expert expectations. The key to achieving this goal is to use a fully declarative language such as Answer Set Programming and enhance the solving capabilities of its tools with dedicated external solvers to deal with variables with large domains. Traditionally, this has been the goal of Constraint Answer Set Programming, where those dedicated external solvers are exposed as extensions to the language. To benefit from the solving performance of those extensions, application designers are required to rewrite part of the encoded knowledge using the dedicated extensions, thus obscuring the declarativeness achieved when using the original language. This project aims to automatically connect the declarative language with the external solvers without requiring the use of dedicated extensions, thus enhancing solving performance without hindering declarativeness. To achieve this goal, the project will significantly advance the formal foundations of (Constraint) Answer Set Programming by developing grounding-independent characterizations of these languages and semantics-preserving translations that synergize the strengths of both approaches to overcome their individual limitations. Additionally, the project is set to cultivate an innovative teaching methodology to equip the forthcoming generation of computer scientists with the skills necessary to effectively utilize this new synergy between Answer Set Programming and Constraint Answer Set Programming. This project is jointly funded by the Division Of Information & Intelligent Systems / Robust Intelligence (IIS/RI) and the Established Program to Stimulate Competitive Research (EPSCoR). This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.