ABSTRACT

Figure 1 Causal AI in web and healthcare: From statistical explainability based on data representation and associational support to context explainability based on causal relation and interventional support eventually leading to domain explainability based on causal representation in knowledge graph leading to counterfactual support.

CCS Concepts: • Artificial Intelligence → Causal AI; • Machine Learning → Explainability; • Natural Language Understanding; • Ontologies → Knowledge Graphs;

Improving the performance and explanations of ML algorithms is a priority for adoption by humans in the real world. In critical domains such as healthcare, such technology has significant potential to reduce the burden on humans and considerably reduce manual assessments by providing quality assistance at scale. In today’s data driven world, artificial intelligence (AI) systems are still experiencing issues with bias, explainability, and human-like reasoning and interpretability. Causal AI is the technique that can reason and make human like choices making it possible to go beyond narrow Machine learning based techniques and can be integrated into human decision making. It also offers intrinsic explainability, new domain adaptability, bias free predictions and works with datasets of all sizes. In this tutorial of type lecture style we detail how a richer representation of causality in AI systems using a knowledge graph (KG) based approach is needed for intervention and counterfactual reasoning (Figure 1), how do we get to model based and domain explainability, how causal representations helps in web and health care.

1 TUTORIAL SCHEDULE AND ACTIVITIES

This tutorial has the following four major modules as follows:

  • (10 mins) The current landscape of AI: We begin the session motivating the need for causal AI in healthcare and when analyzing the web based data. We’ll give instances from actual web examples that highlight the limitations of the available statistical and data-powered AI Systems. The opening session will emphasize the need for causal AI for creating comprehensible explainable AI systems.

  • (25 mins) Causal Knowledge Graph Development: CausalKG is a step toward symbolic AI for knowledge-infused learning that is causality-based. We demonstrate that systems using CausalKG don’t just learn from correlations; instead, have a causal understanding of the environment around them [3]. In addition to causality, CausalKG uses KG which also represents space, time, and interactions. The AI algorithms used today rely on uniformly distributed, independent data which are unable to deduce hypothetical, interventional scenarios or out-of-distribution situations. We show how to enable interventional and counterfactual reasoning that may be derived utilizing observational data and domain expert knowledge and how CausalKG can be utilized to infuse current KGs with causal knowledge of the domain. We discuss on building a CausalKG which has the benefit of incorporating causality into reasoning and prediction processes, such as the understanding, planning, and diagnosing of medical conditions for example.

  • (25 mins) Ontology and Knowledge based inference for causal Explanation: We construct an inference system to record explanations based on causal assertions using an ontology in the form of relationships and sub class hierarchy. We begin by ontology construction that enables us to express how one fact leads to another and how one fact explains another. We provide a formal set of pattern extractions, causal relationships that move from causal claims to explanations. These patterns exhibit the ontological principles considered necessary for reaching judgments on explanatory statements and algorithm predictions. We also present Drug abuse Ontology as an ontological framework that recognizes the patterns in web based data and in the mental healthcare domain.

  • (25 mins) Applications in web and healthcare : The COVID19 Pandemic has brought attention to the disparity in access to mental health services. Internet users have turned to online forums like Reddit to share their experiences. There are clinically accepted causal interpretations and semantics for diagnosing mental illnesses. In this session, we propose a causal process reasoning infused machine learning techniques called Causal Process Knowledge Infused Reasoning (CPR). CPR uses mental health causal models to guide the learning of neuro-symbolic approximation functions (reasoners) to capture causal model conditional probabilities. We assess the efficacy of CPR using task descriptions, datasets, and baseline techniques for suicidality context identification and intervention. We demonstrate a system that deploys CPR on Reddit’s social media (SocMedia) posts showing CPR’s ability to scale to web data. We further discuss possible improvements to causal theories that may enhance performance on these tasks and have broader mental health care implications.

2 TUTORIAL ORGANIZER BIOS

Usha Lokala:

She is a Ph.D. student at AIISC. Her research interests include ontology engineering, knowledge graphs and natural language processing. Her interdisciplinary research funded by NIH, NIDA, and NSF applies ontology, deep learning, and natural language processing in the domain of Public Health, Addiction, Social Media Analysis and Age related Cognition. Her work has been published in reputed conferences and Journals (IEEE, Drug and Alcohol Dependence, WWW, CPDD, AAAI ICWSM, JMIR, PLOS One). Lokala’s work [6] on public health addictions won second prize in Opioid Challenge at SBP BRiMS 2018, a computational social science conference. She also hosted tutorials and talks at venues AAAI ICWSM, AI ML Systems, ASONAM, IJCAI Special Edition on Responsible Social Media Mining. For more details, please visit Lokala’s webpage.


Kaushik Roy:

He is a Ph.D. student at AIISC. He completed his master’s in computer science at Indiana University Bloomington and has worked at UT Dallas’s starling lab. His research interests include statistical relational artificial intelligence, sequential decision making, knowledge graphs, and reinforcement Learning. His work is published in reputed conferences in IEEE, KR, AAAI, and ECML-PKDD. He also hosted several tutorials at AAAI ICWSM, KGC, and AI ML Systems Conferences. For more details, please visit Roy’s webpage.


Utkarshani Jaimini:

She is a Ph.D. student at AIISC. Her research focuses on developing a richer representation of causality using knowledge graph (CausalKG) for better explainability with the applications in autonomous driving and healthcare. Her research interests includes causal analysis, applied machine learning to solve healthcare problems, predictive analytic, information extraction, augmented personalized health, internet of things (IoTs), semantic web and semantic cognitive perceptual computing. Her work is published in reputed conferences and journals (ISWC, ML4H, IC2S2, IEEE Internet Computing, JMIR, American Thoracic Society, SLEEP Society meeting). For more details, please visit Jaimini’s webpage.


Amit Sheth:

He is an Educator, Researcher, and Entrepreneur. He is the founding director of AIISC and NCR Chair at The University of South Carolina. Previously , he was the Lexis-Nexis Ohio Eminent Scholar and the executive director of Ohio Center of Excellence in Knowledge-enabled Computing. He is a Fellow of IEEE, AAAI, ACM, and AAAS. He has organized more then 100 international events (general/program chair, organization committee chair), more then 70 keynotes, given more then 45 many well-attended tutorials and is among the well-cited computer scientists. He has founded three companies by licensing his university research outcomes, including the first Semantic Web company in 1999 that pioneered technology similar to what is found today in Google Semantic Search and Knowledge Graph. Several commercial products and deployed systems have resulted from his research. For more details, please visit Dr. Sheth’s webpage.





3 TOPIC AND RELEVANCE

SocMedia emerged as the go-to platform for mental health (MH) support for patients seeking Mental Health Care (MHCare). The transition of patients from clinical settings to SocMedia has been facilitated by peer-support groups and the lack of social stigma. For example, in order to find clues that can demonstrate a correlation or cause between various MH illnesses and a patient’s propensity for suicide, researchers have begun examining SocMedia content. In that case, this tutorial is relevant to develop causal AI systems for a deeper understanding of SocMedia conversations that may result in policy decisions due to the abundance of SocMedia conversations in the areas of health care, mental health, substance use, etc. that are available on the internet whereas surface level deep learning systems cannot achieve such goals.





4 TUTORIAL AUDIENCE AND PREREQUISITES

Researchers from academia, business, and healthcare professionals can participate in this tutorial to discuss the intersection of causal representation, reasoning, semantic linking, NLP, and deep learning. Since the tutorial is taught in a lecture-style format, a fundamental knowledge of causal systems, Knowledge Graph (KG), Ontology, and natural language processing (NLP) is preferred. This will make it possible for the public to understand the limitations of statistical AI and to monitor the development of AI as it moves toward neuro-symbolic AI. The lesson will include enough examples and use cases of causal AI systems and techniques. The present state of AI/ML systems for web based analysis of health care will be explained to newcomers interested in these systems. The methodology and datasets provided in the tutorial will be valued by experienced participants as viable solutions to common technical challenges in the social good domains. We welcome additional studies using our datasets and methodology from other researchers in order to deepen our understanding of causal AI systems for web and health care.





Our tutors have co-presented multiple tutorials as follows:

  • Talk: Causal Knowledge Graph Explainability using Interventional and Counterfactual reasoning (webpage link).
  • Tutorial : Neuro-symbolic AI for Mental Healthcare (webpage link)

  • Knowledge-infused Reinforcement Learning (webpage link)

  • Explainable AI using Knowledge Graphs (webpage link)

  • Knowledge In-Wisdom Out-Explainable Data for AI in Cyber Social Threats and Public Health (webpage link)

    Knowledge-infused Deep Learning (webpage link)

  • Knowledge-infused NLU for Addiction and Mental Health Research





6 TUTORIAL SOCIETAL IMPACTS

Through the development of domain-specific data sets, cuttingedge frameworks, and computational methods for comprehending user language and asynchronous conversations on a variety of platforms, we investigated a new healthcare dimension of SocMedia like Mental Health Care, and User interactions. Through the use of these methods, we hope to (a) build the new clinical process guidelines for patients and develop task-based Order Sets (b) create actionable Order Sets that are integrated into the system and aid medical professionals in making choices and avoiding delays and discrepancies in diagnosis and treatment. (c) create process-guided explanation models that are simple for the end-user to comprehend (d) create causal AI system using a knowledge-graph-based methodology for better explainability, support for intervention and counterfactuals in social good domains.





7 SLIDES





8 VIDEO





REFERENCES

  • [1] Manas Gaur, Ankit Desai, Keyur Faldu, and Amit Sheth. 2020. Explainable AI Using Knowledge Graphs. In ACM CoDS-COMAD Conference.
  • [2] Manas Gaur, Ugur Kursuncu, Amit Sheth, Ruwan Wickramarachchi, and Shweta Yadav. 2020. Knowledge-infused deep learning. In Proceedings of the 31st ACM Conference on Hypertext and Social Media. 309–310.
  • [3] Utkarshani Jaimini and Amit Sheth. 2022. CausalKG: Causal Knowledge Graph Explainability using interventional and counterfactual reasoning. IEEE Internet Computing 26, 1 (2022), 43–50.
  • [4] U Lokala, M Gaur, K Roy, and A Sheth. 2021. Communication: Knowledge-infused NLU for Addiction and Mental Health Research. In ASONAM 2021- IEEE ACM International Conference on Advances in Social Networks Analysis and Mining. The Hague.
  • [5] U Lokala, M Gaur, K Roy, and A Sheth. 2021. Talk: Knowledge-infused Natural Language understanding for Public Health, Epidemiology, Substance Use, and Mental Health. In Workshop on Mining Actionable Insights from Social Networks. 15th International Joint Conference on Artificial Intelligence.
  • [6] Usha Lokala, Francois R Lamy, Raminta Daniulaityte, Amit Sheth, Ramzi W Nahhas, Jason I Roden, Shweta Yadav, and Robert G Carlson. 2019. Global trends, local harms: availability of fentanyl-type drugs on the dark web and accidental overdoses in Ohio. Computational and mathematical organization theory 25, 1 (2019), 48–59.
  • [7] Kaushik Roy, Manas Gaur, Qi Zhang, and Amit Sheth. 2022. Tutorial: Knowledgeinfused Reinforcement Learning. The Knowledge Graph Conference (2022).
  • [8] Kaushik Roy, Usha Lokala, Manas Gaur, and Amit Sheth. 2022. Tutorial: Neurosymbolic AI for Mental Healthcare. International Conference on AI-ML Systems (2022).
  • [9] Amit Sheth, Kaushik Roy, Manas Gaur, and Usha Lokala. 2021. Tutorial on Knowledge In-Wisdom Out-Explainable Data for AI in Cyber Social Threats and Public Health. AAAI ICWSM (2021).