Alan Ritter

Alan Ritter is an associate professor in the College of Computing at Georgia Tech. His research on natural language processing aims to solve technical challenges that help machines read the web and engage in safe and helpful dialogue with people.

Alan's work on AI-Caring is focused on developing safe dialogue assistants that have the potential to help older adults live more independently through assistance with activities of daily living. For example, Dr. Ritter's research group is developing instruction-grounded chatbots that have the potential to provide assistance in the kitchen [1]. The project identified several deficiencies in existing chatbots, such as ChatGPT, to provide recipe-grounded conversational assistance. In particular these models have a tendency to present instructions to users in the wrong order. To address this challenge, Dr. Ritter's research group has developed new methods that track the current dialogue state (current step of the recipe the user is working on), and also identify user intent. A demo of a system based on this work can be viewed at the following URL: https://tinyurl.com/ChattyChef

Although large language models present opportunities for older adults to sustain independence and improve quality of life, they also present new risks. For example, despite efforts to align model outputs with human preferences, they can sometimes generate offensive replies. In addition to directly generating toxic language, chatbot outputs can be inappropriate in more subtle ways that depend on context and are difficult to detect. For example, a study conducted in Dr. Ritter's group showed models are two times more likely to agree with offensive user inputs as compared to safe inputs [2]. This appears to be caused to the echo chamber effect observed in online social media data included in the model's pre-training data, which models learn to imitate.

In addition to AI-Caring Dr. Ritter has also served as a PI on several DARPA and IARPA programs including CAUSE, BETTER and AMD. As part of DARPA's AMD program (Accelerated Molecular Discovery), Alan led the development of SynKB (https://tinyurl.com/synkb), a system that extracts a high-quality knowledge base of chemistry reaction details [2], which was shown to be competitive with commercial chemistry databases, such as Reaxys. Alan's work on extracting and analyzing reports of new cybersecurity threats [3], developed as part of IARPA‚Äôs CAUSE program, was covered by WIRED (https://www.wired.com/story/machine-learning-tweets-critical-security-f…). Dr. Ritter is the recipient of an NSF CAREER award and an Amazon Research Award.

[1] Improved Instruction Ordering in Recipe-Grounded Conversation
Duong Minh Le, Ruohao Guo, Wei Xu and Alan Ritter
ACL 2023

[2] Just Say No: Analyzing the Stance of Neural Dialogue Generation in Offensive Contexts
Ashutosh Baheti, Maarten Sap, Alan Ritter, Mark Riedl
EMNLP 2021

[3] SynKB: Semantic Search for Synthetic Procedures
Fan Bai, Alan Ritter, Peter Madrid, Dayne Freitag, John Niekrasz
EMNLP 2022 (Demo Track)

[4] Analyzing the Perceived Severity of Cybersecurity Threats Reported on Social Media
Shi Zong, Alan Ritter, Graham Mueller and Evan Wright
Proceedings of NAACL 2019