This innovative project showcases the AI’s newfound spatial awareness, a feat not commonly seen in other AI models like Sora by OpenAI, which often display unrealistic architectural designs and glitches.
The study, published in the Nature Machine Intelligence journal on Nature.com, titled automated construction of cognitive maps with visual predictive coding, authored by James Gornet & Matt Thomson from the California Institute of Technology, elucidates the methodology in significant detail. The paper, released recently, also provides access to the code on GitHub and Zenodo.Matt Thomson, one of the researchers involved in the project, shared insights with TechXplore, emphasizing the limitations of current AI models. According to Thomson, these models lack true problem-solving abilities, conceptual navigation skills, and the capacity for creativity. He pointed out that AI tends to rely on memorization rather than synthesis of ideas.
The project was spearheaded by graduate student James Gornet, who advocated for the use of Minecraft and brought together his expertise in neuroscience, machine learning, math, statistics, and biology from the Department of Computational and Neural Systems (CNS) at Caltech. While James Gornet did not comment on the project, Thomson highlighted the importance of their research in advancing AI and potentially gaining insights into brain functions through CNS studies.
