{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_52980"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_52980","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"Assessing Robustness of AI-Based Systems in the Face of Human-Based Exploitative Uncertainty","abstract":"Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","abstract_html":"Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","abstract_has_math":false,"creators":["Chan, Kenneth H."],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Cheng, Betty H.C","Bopardikar, Shaunak D.","Siegel, Joshua E.","Yan, Qiben"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:34Z","subjects":["Computer science","Artificial intelligence"],"languages":["eng"],"rights":["In Copyright"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52980","local:Chan_grad.msu_0128D_21966"],"render_values":[{"text":"etd:52980","href":null,"code":true},{"text":"local:Chan_grad.msu_0128D_21966","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/jv6q-fx29","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/jv6q-fx29","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_52980","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Cheng, Betty H.C","Bopardikar, Shaunak D.","Siegel, Joshua E.","Yan, Qiben"]},{"key":"dc:creator","label":"Author","values":["Chan, Kenneth H."]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Computer science","Artificial intelligence"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52980","local:Chan_grad.msu_0128D_21966","https://doi.org/doi:10.25335/jv6q-fx29"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","As research continues to develop and improve the capabilities of artificial intelligence, its integration in safety-critical, learning-enabled systems is increasingly ubiquitous. Machine learning models extract patterns directly from large collections of data, which can reduce human developer burden by automating intensive tasks. The ability and performance of these machine learning models are closely tied to the completeness of the training data, which often includes only the expected operating contexts due to the sparseness of edge cases. However, systems deployed in the real world often involve operating contexts that include humans. Human-involved operating contexts introduce a large source of uncertainty due to the inadequate coverage of relevant training data that comprises unpredictable, erratic, and even malicious human behaviors. Yet, when deployed in safety-critical settings, these systems must demonstrate safe behaviors and mitigate potential damages in the face of uncertainty in addition to satisfying operational requirements. This dissertation describes automated testing methods that address the robustness of learning-enabled systems with respect to a range of human-based exploitative uncertainties. First, we define a spectrum of exploitative uncertainty based on the intentions of the external human agent(s). Specifically, we broadly categorize human intentions into malicious, selfish, or incidental intentions. Second, we address the problem with a multi-disciplinary and multipronged approach, integrating and synthesizing a number of distinct and disparate techniques to assess and improve the robustness of the system under study. For malicious intentions, we harness evolutionary search to identify different types of diverse failures in learning-enabled components and systems. For selfish intentions, we use a top-down modeling approach with reinforcement learning and game theory to identify different strategies and maneuvers that may lead to undesirable behavior(s). For incidental intentions, we do not specify a priori specific human-based uncertainty, which facilitates the identification of a broad range of behavior(s). More specifically, we use a bottom-up exploratory approach with novelty search and behavior trees to discover diverse behaviors in learning-enabled systems. We demonstrate the capabilities of our approaches by applying the described research to various state-of-the-art, machine-learning-based, and safety-critical cyber-physical systems.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (195 p.) : ill."]},{"key":"dc:title","label":"Title","values":["Assessing Robustness of AI-Based Systems in the Face of Human-Based Exploitative Uncertainty"]}]}],"canonical_facts":{"dc:contributor":["Cheng, Betty H.C","Bopardikar, Shaunak D.","Siegel, Joshua E.","Yan, Qiben"],"dc:creator":["Chan, Kenneth H."],"dc:date":["2026"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Computer Science - Doctor of Philosophy, 2026","As research continues to develop and improve the capabilities of artificial intelligence, its integration in safety-critical, learning-enabled systems is increasingly ubiquitous. Machine learning models extract patterns directly from large collections of data, which can reduce human developer burden by automating intensive tasks. The ability and performance of these machine learning models are closely tied to the completeness of the training data, which often includes only the expected operating contexts due to the sparseness of edge cases. However, systems deployed in the real world often involve operating contexts that include humans. Human-involved operating contexts introduce a large source of uncertainty due to the inadequate coverage of relevant training data that comprises unpredictable, erratic, and even malicious human behaviors. Yet, when deployed in safety-critical settings, these systems must demonstrate safe behaviors and mitigate potential damages in the face of uncertainty in addition to satisfying operational requirements. This dissertation describes automated testing methods that address the robustness of learning-enabled systems with respect to a range of human-based exploitative uncertainties. First, we define a spectrum of exploitative uncertainty based on the intentions of the external human agent(s). Specifically, we broadly categorize human intentions into malicious, selfish, or incidental intentions. Second, we address the problem with a multi-disciplinary and multipronged approach, integrating and synthesizing a number of distinct and disparate techniques to assess and improve the robustness of the system under study. For malicious intentions, we harness evolutionary search to identify different types of diverse failures in learning-enabled components and systems. For selfish intentions, we use a top-down modeling approach with reinforcement learning and game theory to identify different strategies and maneuvers that may lead to undesirable behavior(s). For incidental intentions, we do not specify a priori specific human-based uncertainty, which facilitates the identification of a broad range of behavior(s). More specifically, we use a bottom-up exploratory approach with novelty search and behavior trees to discover diverse behaviors in learning-enabled systems. We demonstrate the capabilities of our approaches by applying the described research to various state-of-the-art, machine-learning-based, and safety-critical cyber-physical systems.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (195 p.) : ill."],"dc:identifier":["etd:52980","local:Chan_grad.msu_0128D_21966","https://doi.org/doi:10.25335/jv6q-fx29"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["In Copyright"],"dc:subject":["Computer science","Artificial intelligence"],"dc:title":["Assessing Robustness of AI-Based Systems in the Face of Human-Based Exploitative Uncertainty"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:34Z"}