Researchers have found evidence that large language models (LLMs) can understand and process natural objects similarly to humans. This groundbreaking discovery suggests a potential shift in how we perceive AI's cognitive abilities.
A team from the Chinese Academy of Sciences and the South China University of Technology in Guangzhou conducted the study. They explored whether LLMs could develop cognitive processes similar to human object representation. This includes recognizing and categorizing objects based on function, emotion, and environment.
The researchers tested models like ChatGPT 3.5 and Gemini Pro Vision. They assigned them "object elimination" tasks using text and images. The AI systems analyzed 4.7 million responses related to 1,854 natural objects, including dogs, chairs, and cars.
The results revealed that the AI models created 66 conceptual dimensions for organizing objects. This mirrors how humans categorize and understand the world around them. Multimodal models, which combine text and images, showed even greater alignment with human thinking.
Furthermore, brain scan data showed overlap between how AI and the human brain respond to objects. This suggests that future AI could have more intuitive and human-compatible reasoning. This is crucial for fields like robotics, education, and human-AI collaboration.
However, it's important to note that LLMs do not understand objects in the same emotional and experiential way humans do. This research opens new avenues for developing AI that can better interact with and understand the world.