Danielle R. Thomas
I am a Systems Scientist/Faculty within the Human-Computer Interaction Institute at Carnegie Mellon University. Concurrently, I am the Research Lead at PLUS—Personalized Learning Squared, a research project led by Prof. Ken Koedinger at CMU, in collaboration with Carnegie Learning, Inc. and Stanford University.
My research interests focus on improving student learning outcomes and equitizing educational access through 1) the advancement of AI in education, 2) the development of hybrid human-AI tutoring systems, and 3) understanding how inequity impacts students' learning.
As a former middle school teacher, school admin, and teacher educator, my first-hand experiences fuel the mission of the PLUS project of doubling math learning for 10,000 students by 2026!
Check out my CV for information about my experiences, education, and notable publications. Over the past year I have first-authored papers in conference proceedings, such as Artificial Intelligence in Education (AIED), Learning Analytics and Knowledge (LAK), and the International Journal of STEM Education.
What's happening now...
Hot off the press...
We conduct a two-study quasi-experiment to determine the impact of hybrid human-AI tutoring among students with and without disabilities in general education classrooms. We find positive effects among students. In particular, students with disabilities may benefit more from the motivational benefits of human tutor interaction. [preprint]
Danielle R. Thomas, Erin Gatz, Shivang Gupta, Vincent Aleven, & Kenneth R. Koedinger
In The 25th Artificial Intelligence in Education (AIED) Conference, July 7-13, 2024, Recife, Brazil (2024)
How do you respond to students saying, "I am dumb" or "I can't do this?" We (and generative AI!) assess the performance of 60 tutors within an online lesson tasking tutors to respond to situations involving students engaging in negative self-talk. We (and AI) find evidence of tutor learning, with GPT-4 demonstrating high absolute performance, paving the way for assessing tutors at scale. [preprint]
Danielle R. Thomas, Jionghao Lin, Shambhavi Bhushan, Ralph Abboud, Erin Gatz, Shivang Gupta, & Kenneth R. Koedinger
In The 11th ACM Conference on Learning @ Scale (L@S), July 18-20, 2024, Altanta, Georgia (2024)
This work overviews the progress of the PLUS project towards using generative AI for tutoring feedback and assessment. While using generative AI shows promise as a low-cost and efficient method for these uses, ethical considerations and practical implications are discussed to ensure fair and responsible use. [preprint]
Danielle R. Thomas, Erin Gatz, Shivang Gupta, Jionghao Lin, Cindy Tipper, & Kenneth R. Koedinger
In The 17th Annual Learning Ideas Conference, June 12-14, 2024, New York, NY (2024)
We introduce hybrid human-AI tutoring and implement the model across three diverse schools. We find positive impacts on learning outcomes with evidence suggesting lower achieving students may benefit more from tutoring than higher achieving students—a promising finding. [link]
Danielle R. Thomas, Jionghao Lin, Erin Gatz, Ashish Gurung, Shivang Gupta, Kole Norberg, Stephen E. Fancsali, Vincent Aleven, Lee Branstetter, Emma Brunskill, Kenneth R. Koedinger
In The 14th Learning Analtyics and Knowledge (LAK) Conference, March 18-22, 2024, Kyoto, Japan (2024)
In this systematic review, we determine the average STEM student outperforms ~70% of their peers. Most notably, underrepresented minority students benefit given one caveat—they must be given the opportunity. [Journal article link]
Danielle R. Thomas & Karen H. Larwin
International Journal of STEM Education (2023)
This workshop highlights the challenges and opportunities of AI-in-the-loop math tutoring and encourages discourse in the AIED community. Access papers and presentations here.
Vincent Aleven, Richard Baraniuk, Emma Brunskill, Scott Crossley, Dora Demszky, Stephen Fancsali, Shivang Gupta, Kenneth R. Koedinger, Chris Piech, Steve Ritter, Danielle R. Thomas, Simon Woodhead, Wanli Xing
In The 24th Artificial Intelligence in Education (AIED )Conference, July 3-7, 2023, Tokyo, Japan (2023)
We introduce Personalized Learning Squared (PLUS), a human-AI tutoring platform designed to improve tutoring efficiency. PLUS leverages student-facing AI-powered math software and a tutor-facing personalized dashboard to provide the right support, to the right student, and at the right time.
Danielle R. Thomas, Shivang Gupta, Erin Katz, Cindy Tipper, Kenneth R. Koedinger
in 16th Annual Learning Ideas Conference, NYC (2023)
We introduce a method of providing explanatory feedback to human tutors on their responses to open-ended questions leveraging LLMs using named entity recognition.
Jionghao Lin, Danielle R. Thomas, Feifei Han, Shivang Gupta, Wei Tan, Ngoc Dang Nguyen, Kenneth R. Koedinger
Workshop at 24th Artificial Intelligence in Education (AIED) Conference (2023)
Towards the Future of AI-Augmented Human Tutoring in Math Learning
We compare the performance of humans and GPT-4 in identifying criteria of praise by tutors to students. GPT-4 performs moderately well is some areas but underperforms in recognizing sincerity and authenticity- not surprising, yet paves the way for future work.
Dollaya Hirunyasiri, Danielle R. Thomas, Jionghao Lin, Kenneth R. Koedinger, Vincent Aleven
Workshop at 24th Artificial Intelligence in Education Conference (2023) Towards the Future of AI-Augmented Human Tutoring in Math Learning
We introduce an AI-based method of autograding online tutor lessons. Comparing two methods of training set creation using learnersourced tutor responses and by prompting ChatGPT. Our findings show a constructive use of ChatGPT for pedagogical purposes that is not without limitations. [Video presentation]
Danielle R. Thomas, Shivang Gupta, Kenneth R. Koedinger
In The 24th Artificial Intelligence in Education Conference, July 3-7, 2023, Tokyo, Japan (2023)
We show tutors perform ~20% better from pretest to posttest on our short scenario-based lessons similar to situational judgement tests. How would you respond to a student who has just made a math error?
Danielle R. Thomas, Xinyu Yang, Shivang Gupta, Adetunji Adeniran, Elizabeth McLaughlin, Kenneth R. Koedinger
In The 13th International Learning Analytics & Knowledge Conference, Austin, TX (2023)
Comparing the achievement of 70 students participating in a hybrid tutoring program compared to a matched control, we found the learning gain among participating students was nearly double that of students not participating.
Danielle R. Chine, Cassandra Brentley, Carmen Thomas-Browne, J. Elizabeth Richey, Abdulmenaf Gul,... Kenneth R. Koedinger
In The 23rd Artificial Intelligence in Education Conference, Durham, UK (2022)
