Human-Centered AI Research Intern, Causal Reasoning Models

Toyota Research Institute•Los Altos, CA

7d•$45 - $65•Hybrid

About The Position

Toyota Research Institute (TRI) is seeking a highly motivated and talented PhD research intern to join their Adaptive Behavior Systems team within the Human-Centered AI Division (HCAI). This interdisciplinary team, composed of ML researchers and social and behavioral scientists, aims to enable decisions through human-centered AI that advance collective well-being for people and the planet. The intern will focus on pushing the boundaries of causal reasoning in Large Language Models (LLMs), collaborating to study and develop methods for testing and improving causal reasoning in these models. The project is geared towards publication in a top-tier venue and will involve strategy discussions connecting research to business impact at Toyota. This is a paid, 12-week internship opportunity during Winter 2027, with a hybrid in-office work arrangement.

Requirements

Ph.D. student in related fields - AI/ML, computer science, data science, statistics, or related field.
Publication record or demonstrated research experience in causal inference (structural causal models, identification, counterfactual reasoning).
Hands-on experience with one or more of: LLM evaluation and benchmarking; mechanistic interpretability; and methods for improving LLM reasoning, including prompting and prompt optimization, post-training (supervised fine-tuning or reinforcement learning with verifiable rewards), pretraining, or architectural changes.
Experience with Python programming and DL frameworks like Pytorch.
Interest in human-centered research (e.g. behavioral science, computational social science), including qualitative and/or quantitative methods such as experiments and user studies.
Excellent communication and teamwork skills.

Responsibilities

Perform research and publish in a relevant venue. Publication target venues include NeurIPS, ICML, ICLR, CLeaR, UAI, and COLM, with ACL or EMNLP where the contribution is primarily linguistic, and TMLR as a journal option. Findings may also be presented at ACIC or EuroCIM. The exact topic is to be finalized with the mentor.
Take ownership of the project from project inception and ideation to validation of the developed methods.
Collaborate cross-functionally with researchers in multiple fields to research and develop technology that leverages generative AI to understand human behavior.