Examination of risks in AI/ML applications

Abstract

Artificial Intelligence (AI) and Machine Learning (ML) systems powered by Natural Language Processing (NLP) and Computer Vision (CV) are permeating across industries and in our daily lives. Due to novelty of technology, AI/ML applications expose organizations to social, legal, and financial risks. Notable examples: Amazon’s AI hiring tool, Microsoft’s chatbot, Uber’s autonomous car accident. New regulations are being introduced across the world to govern AI applications. This dissertation explores the causes of these risks and mitigation strategies through three essays. Essay 1 uses grounded theory approach to propose a unifying theoretical framework for unintended consequences in AI projects. In this essay, 840 quotes from key informants about 30 unique AI cases using multiple news articles for each case were analyzed. The analysis of media discourse revealed signals of intended actions concerning the implementation of AI tools, which led to unintended consequences through various linking mechanisms. Essay 2 provides a conceptual framework using socio-technical systems theory to study effects of risk factors on AI project risk assumed by organization in developing and implementing AI systems. Essay 3 attempts to explore risk factors disclosed by AI oriented organizations in their annual disclosures using a dataset of 112 SEC annual 10-K filings. Together, this dissertation attempts to contribute to risk management literature in context of AI. Expected findings can inform organizations of critical sources of risk in AI projects and help mitigate them.