428. Anku Rani. Scoring Cultural Accuracy.
Anku Rani
September 10, 2026
Listen to full episode :
Does a better-looking AI video mean it’s a better video?
Not when culture enters the picture.
In this episode, Gabriella Mirabelli talks with Anku Rani, a doctoral researcher at MIT and an MIT Tata Center fellow, about why AI-generated video that looks best on screen can still be the least culturally accurate. Their conversation covers findings from her research testing three current video generation models against real viewers across nine countries.
Listen and learn:
Why cultural accuracy splits into three distinct measures: who is shown, what they do, and what surrounds them
Why the model that looked best on screen scored lowest for cultural accuracy, an inverse relationship her data confirmed twice
Why viewers from nine countries agreed with the culture score over standard video quality metrics
Why removing a country name from a prompt made a video less accurate, not more universal
A practical listen for any brand generating video across more than one market, Rani's research gives brands a specific recommendation: judge an AI video model on cultural accuracy, not just cinematic polish, before it reaches a global audience. Keep a human in the loop and keep prompting until you get it right.
Anku Rani is a doctoral researcher at the Massachusetts Institute of Technology and is also an MIT Tata Centre Fellow, where she works at the intersection of multimodal AI and Human-Computer Interaction. Her research focuses on culture and video generation. She is also spending this summer training multimodal large language models for quantum computing at IBM Research.
Prior to MIT, Anku conducted research at Adobe Research on visual language models and mathematical reasoning, and at the University of South Carolina's Artificial Intelligence Institute, where she focused on fact verification. Her work has been published in leading conferences, and she actively contributes to the academic community through conference reviews (ACL, EMNLP, AAAI, LREC, COLING, NAACL), workshop organization (AAAI'23, AAAI'24), and program committee service. She is a two-time AAAI scholarship recipient (2023, 2024).
Before moving to the US, Anku worked as a research scientist with a health tech startup and as a machine learning engineer at an edtech startup. She has also served as a data scientist at Verisk Analytics, a US-based InsurTech company, following her postgraduate studies in AI and ML at Plaksha University. Throughout her career, she has consulted for universities, startups, and government organizations.
Follow & Learn More:
Anku Rani