A human-centric artificial intelligence approach towards equality, well-being, and responsibility in sustainable communities

Shang, Lanyu

Permalink

https://hdl.handle.net/2142/125695 Copy

Description

Title

A human-centric artificial intelligence approach towards equality, well-being, and responsibility in sustainable communities

Author(s)

Shang, Lanyu

Issue Date

2024-07-11

Director of Research (if dissertation) or Advisor (if thesis)

Wang, Dong

Doctoral Committee Chair(s)

Wang, Dong

Committee Member(s)

• Chin, Jessie
• Cai, Ximing
• Wei, Na

Department of Study

Information Sciences

Discipline

Information Sciences

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Artificial Intelligence
• Human-Centric AI
• Community Sustainability

Abstract

Community sustainability fosters peace and prosperity by meeting the needs of all individuals while preserving natural systems, and this dissertation focuses on three fundamental aspects of equality, well-being, and responsibility to foster sustainable communities. To enhance these aspects within sustainable communities, this dissertation designs a human-centric artificial intelligence for sustainable communities (HAI4SC) approach by leveraging the complementary strengths of artificial intelligence (AI) and human intelligence (HI). In particular, AI has shown superiority in processing large amounts of data, identifying latent patterns, and making predictions, which can help address the scalability and complexity of sustainability challenges. On the other hand, HI excels in providing context, domain expertise, and human-centric insights, which are essential for understanding the complex social and physical factors that influence community sustainability. For example, while AI models could efficiently analyze a vast amount of social media posts to classify misleading content in online communities, they often fall short in identifying misinformation in new or emerging public health crises when there is a lack of timely training data. In contrast, the HI from the domain knowledge contributed by healthcare experts could provide critical context and insights to further adapt such AI models to the new domains, ultimately enhancing community resilience in combating emerging health misinformation. Additionally, by leveraging the mobility and cognitive capability of humans, HI could gather on-the-ground information and localized context to address critical problems in sustainable communities, such as assessing household groundwater contamination and understanding the societal impact of environmental crises. However, individual human inputs from HI often suffer from uncertainty and noise due to the inherent subjectivity and variability in human observations. To complement such limitations of HI, AI could help mitigate these issues by effectively capturing hidden patterns and explicitly quantifying uncertainty of individual observations from the collective yet noisy inputs contributed by humans, improving the accuracy and trustworthiness of community-driven assessment solutions. By harnessing the complementary strengths of HI and AI, this dissertation develops a holistic human-centric AI approach for community sustainability by addressing three fundamental challenges: multimodality, adaptability, and trustworthiness. First, human-centric AI applications for community sustainability often encounter varied data modalities (e.g., text, images, and videos) contributed by human participants, making it challenging to fully understand and analyze complex multimodal content, especially for the malicious or harmful content that is intentionally crafted to undermine the equality and well-being of human communities. Second, given the highly dynamic and evolving nature of human communities, data in human-centric AI applications also spans various domains (e.g., topics, events), and human-centric AI solutions need to remain adaptable to these discrepant domains. Third, it is also essential for such human-centric AI solutions to be trustworthy when supporting critical decision-making for individuals in sustainable communities. To address the multimodality challenge, we develop a multimodal information fusion system that introduces novel generative learning and cross-modal attention designs to seamlessly integrate the multimodal content in enhancing information credibility in sustainable communities. To address the adaptability challenge, we design an adaptive cross-domain analytic framework that explicitly incorporates domain knowledge from well-studied source domains and learn domain-invariant representations that can be adapted to the target domain for improving the resilience and well-being of sustainable communities. To address the trustworthiness challenge, we develop a trustworthy social-physical knowledge distillation scheme that designs novel uncertainty-aware graph learning and attentive time series forecasting mechanisms to effectively model the diverse and uncertain human inputs from community stakeholders to strengthen community responsibility and sustainability. We evaluate the HAI4SC approach with a series of real-world case studies in sustainable communities. By addressing these challenges within the human-centric AI systems, the HAI4SC approach in this dissertation is the first of its kind in effectively harnessing the complementary power of HI and AI toward improving equality, well-being, and responsibility in sustainable communities, significantly advancing existing human-centric AI solutions.

Graduation Semester

2024-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/125695

Copyright and License Information

Copyright 2024 Lanyu Shang

Owning Collections