IPTC Research Included in the 2026 AI Index: A Reference for Global AI Trends

20/04/2026

Can AI Really Tell Time?

with the Information Processing and Telecommunications Center (IPTC) from the Universidad Politécnica de Madrid (UPM) and collaborating academic institutions, investigates whether multimodal large language models (MLLMs) genuinely understand how to read time on analog clocks. While MLLMs have shown impressive capabilities in tasks such as visual question answering and image interpretation, the authors reveal a striking limitation: these systems often fail at the seemingly simple task of telling time from analog clock images.

Reading an analog clock requires identifying visual elements (clock hands), interpreting their spatial orientation, and translating this into a correct time representation. Despite the apparent simplicity, the study demonstrates that current MLLMs struggle with this multi-step reasoning process. Through systematic evaluation, the researchers show that these models lack robust generalization and abstraction abilities, even in controlled and basic scenarios.

The findings highlight a broader challenge in artificial intelligence: scaling models and training data does not necessarily lead to deeper conceptual understanding. Instead, the authors argue for new approaches that prioritize higher-level reasoning and abstraction during training and fine-tuning.

Potential applications of this research include improving multimodal AI systems used in real-world scenarios such as assistive technologies, robotics, autonomous systems, and visual inspection tools, where precise interpretation of visual information is critical.

We are also proud to see our IPTC colleagues recognized in the AI Index Report 2026, published by the Stanford Institute for Human-Centered AI (HAI). This influential report provides a comprehensive overview of global AI progress, trends, and impact, highlighting leading research contributions and shaping the future direction of AI development.

Being cited in this report places IPTC among the elite group of institutions shaping the global understanding of AI development. You can find the cited article in the AI Index Report 2026 at Chapter 2. Technical Performances, p. 96.

Bibliographic reference:

Fu, T., González, M., Conde, J., Merino-Gómez, E. & Reviriego, P. Have Multimodal Large Language Models Really Learned to Tell the Time on Analog Clocks? in IEEE Internet Computing, 19 (4), pp. 48-54. https://doi.org/10.1109/MIC.2025.3618144

Authors:

Javier Conde Díaz: GS / ORCID / LinkedIn

Pedro Reviriego Vasallo: GS / ORCID / LinkedIn

LinkedIn: https://www.linkedin.com/company/iptc-upm/

For more information: www.iptc.upm.es