A new study from China has made a discovery about how artificial intelligence (AI) models like ChatGPT and Gemini work. Researchers say these large language models (LLMs) can understand and sort real-world objects in a way that is very similar to how humans do. What’s even more interesting is that the AI models seem to do this naturally. They were not trained to do it on purpose.
Scientists from the Chinese Academy of Sciences and the South China University of Technology in Guangzhou conducted the study. They published it in the science journal, Nature Machine Intelligence. The researchers wanted to know if AI could think in the same way as humans when it comes to recognizing and grouping different objects.
To test this, the team used two popular AI models: ChatGPT-3.5, which works with text, and Gemini Pro Vision, which works with images. They gave the models a special task. The AI had to look at a group of three objects and choose which one didn’t belong something called the “odd-one-out” task. These tasks are often used in human psychology to study how people think.
The AI was tested using 1,854 common natural objects. These included things like dogs, chairs, apples, and cars. The models answered more than 4.7 million times. When researchers looked at the answers, they found something amazing.
The AI models had created 66 different “conceptual dimensions” to help them organize the objects. These were not just simple categories like food or animals. The AI also considered things like how something feels, if it is safe for children, or what kind of emotion it might bring up.
This showed that the AI was thinking in a more complex way just like humans do. It was not just copying information it had seen during training. It was forming its own ideas about how objects are connected.
The researchers also looked at multimodal models, which are AI systems that work with both text and images. They found that these models understood objects even better and more like humans.
Because they can use both words and pictures at the same time, they can form a deeper understanding of what something is. The scientists even used brain scan data to compare how humans and AI respond to the same objects. They found that there was some overlap in how both the brain and AI models processed the information.
This showed that AI models are starting to do more than just copy patterns from data. They may be building their own way of thinking about the world. This is important because it could mean we are getting closer to developing artificial general intelligence, or AGI. A machine that can think and reason like a human, not just in one task but across many different kinds of problems.
The researchers said, “These models are showing signs of building an internal model of the world that resembles human thinking. This could bring us closer to machines that can understand the world in a more flexible and human-like way.”
The scientists also made it clear that while AI may seem to understand the world, it does not have real emotions or personal experience. AI models don’t learn by touching or feeling things like humans do.
Instead, they learn by looking at patterns in huge amounts of text or images. That means their version of understanding is not the same as ours. They are still tools that reflect the knowledge humans have already created.
Researchers explained, “LLMs don’t have lived experience. They are not truly aware. What they do is find and repeat patterns that often line up with how we think, but that doesn’t mean they actually feel or know what something is.”
Still, the findings change the old idea that AI is just a fancy calculator. If AI models can sort and group objects in ways similar to humans, then we may need to rethink what these systems are capable of.
This research could also change how we use AI in areas like education, healthcare, robotics, and more. If machines can understand things like people do, they could become better helpers and partners in everyday life. But experts also say we need to be careful. As AI gets smarter, we have to ensure that people use it in ways that are safe and fair.
