AI INCIDENTS AND THEIR IMPACT
Recent advances in artificial intelligence (AI) have taken the world by storm, via industry leaders such as OpenAI, DeepMind, and Google, among many others. This has led to popular interfaces used daily such as ChatGPT, Bing, and Bard, not to mention the thousands of AI processes working in the background of industries across the globe. Underneath these developments is a deep concern about potential dangers ushered in by this new technology. There are reports in the global media of incidents in which AI has harmed or near-harmed the real world in its deployments, but these incidents are just beginning to be catalogued and analyzed to find structure, patterns, and trends to guide future strategy for the sake of all humanity.
This project aims to use statistical and machine learning tools to explore important characteristics of AI related incidences reported in the AI Incident Database (AIID), an open source initiave of the Responsible AI Collaborative. Incident reports include information such as location and date of incident report, alleged deployer, harmed parties, and taxonomy of the incident type, related to chatbot, autonomous driving, and deepfake video generation, among many others.
The project intends to further direct machine learning techniques back on the system to identify potential risk factors associated with these incidents by addressing research questions such as: "What types of AI incidents are the most prevalent?" and "Who are the worst alleged offenders of these incidents?" Students are expected to utilize various R and/or Python graphical and analytical tools (such as Pandas, leaflet, and shiny) to explore, visualize, and reveal patterns in the AI incident data. Emphasis will be placed on the use of interactive visualizations to gain deeper insights from the incident data.
Takeaways: This project provides students the opportunity to seek data-driven answers on the patterns and underlying trends of AI incidents, to improve their coding and programming skills using the R and Python programming languages, and to build up essential collaborative, critical thinking, and problem-solving skills.
Prerequisites: A good understanding of introductory statistics is required for participants. Having good programming skills and some knowledge of intermediate statistics will be helpful, but not required.