Dr. Licato Wins NSF Award to Study Human Reasoning and AI

licato grant

Computer Science and Engineering assistant professor Dr. Licato received an National Science Foundation (NSF) award of $150,000 for his project "EAGER: Grounding Natural Language Inference in Cognitive Processes."

Dr. Licato, who is the director of the Advancing Machine and Human Reasoning (AMHR) Lab, will study Type 1 and Type 2 reasoning, made famous by the book “Thinking, Fast and Slow.” This work will help improve the reasoning processes of Large Language Model AIs like ChatGPT. 

Current approaches to determining semantic textual similarity (STS) in the field of natural language processing tend to over-rely on measures derived from substring overlap or similarity. The project will result in new methods for studying and automatically assessing both kinds of inferential relationships, and then develop new ways of measuring text meaning by introducing a six-option answer format that more richly captures naturalistic inference types. The aim is to develop automated methods to estimate how people naturally reason about the inferential relationships between texts through a well-known distinction from cognitive psychology: Type 1 and Type 2 reasoning.

Type 1 reasoning processes are quick, automatic, and often unconscious, such as recognizing a friend's face, whereas Type 2 processes are slower and more deliberate.

The stated objectives of the project are to:

1. Develop new methods for studying inferential properties that emerge at the sentence- rather than word-level of natural text.

2. Develop new ways of measuring how similar two pieces of text are, helping us fight plagiarism.

3. Carry out empirical data collection and study designs that strongly consider issues of fairness, ethics, accountability, and transparency.