This conference is from July 6, 2026–July 11, 2026.
The Human-First AI Lab (HAL 2.0) at the University of Nebraska-Lincoln had two papers accepted at the 43rd International Conference on Machine Learning (ICML 2026), one of the top AI conferences in the world. One of the papers, Position: VLM Causal Reasoning Benchmarks Should Probe Temporal Understanding, Not Presume It, received spotlight designation, placing it in the top 5% of submissions. Position papers are reserved for ideas intended to shape the future direction of a field. The second paper, The Abstraction Gap in Vision-Language Causal Reasoning, introduces new diagnostic tools for evaluating machine perception. This year ICML received approximately 24,000 submissions, the largest to date in the history of any AI conference.
Both papers probe whether vision-language models engage in genuine causal reasoning or generate responses that sound right but lack deeper understanding. The spotlight position paper takes an unconventional approach, drawing on Paul Cézanne's painting practice to capture how human visual perception works. Cézanne built visual understanding through sustained attention and repeated observation over time, not by capturing a single frozen moment. The paper argues that visual intelligence in machines requires a similar capacity for temporal understanding, a capacity that current AI benchmarks do not test.
The HAL 2.0 is directed by Dr. M. R. Hasan, Assistant Professor of Electrical and Computer Engineering.
For more information on ICML, visit here.