Researchers in China have discovered that AI models like ChatGPT process information in a similar way to the human brain.
A study into large language models (LLMs) found evidence that popular artificial intelligence tools built by OpenAI and Google sort information in a spontaneous way, despite not being trained to do so.
The findings, made by a team from the Chinese Academy of Sciences and South China University of Technology, revealed that LLMs “share fundamental similarities that reflect key aspects of human conceptual knowledge” – challenging the assumption that AI systems simply mimic responses through pattern recognition.
The researchers tasked OpenAI’s ChatGPT-3.5 and Google’s Gemini Pro Vision with performing an “odd-one-out” task, which saw the AI create 66 conceptual dimensions in order to categorise objects.
“Remarkably, the dimensions underlying these embeddings were interpretable, suggesting that LLMs and multimodal LLMs develop human-like conceptual representations of objects,” the researchers said.
“This provides compelling evidence that the object representation in LLMs, although not identical to human ones, share fundamental similarities that reflect key aspects of human conceptual knowledge.”
The study detailing the research, titled ‘Human-like object concept representations emerge naturally in multimodal large language models’, appears in the scientific journal Nature Machine Intelligence.
The scientists hope that their findings will help the development of “more human-like artificial cognitive systems” that are able to collaborate better with humans.
Other research teams are already developing more human-like AI systems, with one Australian startup recently unveiling the world’s first commercial biological computer that runs on living human brain cells.
Cortical Labs’ “body in a box” computer uses lab-grown neurons on a silicon chip in order to send and receive electrical impulses.
The startup claims that the biological-based system can learn and adapt more efficiently than traditional computing systems, with an early version demonstrating how 800,000 human and mouse neurons could teach itself how to play the video game Pong.
The neurons exhibited sentience when embodied in the game world, according to a paper published in the journal Cell.
“Our technology merges biology with traditional computing to create the ultimate learning machine,” the company’s website states.
“The neuron is self programming, infinitely flexible, and the result of four billion years of evolution.”
