Visual General Intelligence: A White Paper (arxiv.org)

A new white paper titled "Visual General Intelligence" presents a vision-centered approach to understanding intelligence and its potential pathway toward Artificial General Intelligence (AGI). The authors explore how visual experiences, such as images and videos, could contribute to the development of general intelligence similar to advancements seen in language models like GPT, which have showcased the ability to transfer knowledge across tasks through extensive training on diverse text data. This inquiry highlights the importance of visual modalities, setting the stage for a paradigm where visual intelligence is foundational in the AGI landscape. Significantly, this framework encourages a comprehensive examination of how computer vision research can evolve in the AGI era, proposing new benchmarks and learning paradigms while establishing the interplay between visual and language-based knowledge. By fostering a collaborative dialogue among experts from various fields, the paper aims to clarify the principles that should guide computer vision research, ultimately contributing to a more holistic understanding of intelligence that encompasses both visual and textual forms, thus enriching the AI/ML community's approach to achieving AGI.