Researchers discover that AI chatbots and a human brain disorder share a hidden similarity—opening new doors in both neuroscience and artificial intelligence.
Key Points at a Glance
- LLMs like ChatGPT can appear fluent while providing incorrect or incoherent answers.
- This behavior mirrors Wernicke’s aphasia, a human brain disorder.
- Energy landscape analysis reveals structural signal similarities between AI and aphasic brains.
- The findings could lead to improved diagnostics for aphasia and better AI model design.
- Researchers urge caution, noting that AI and human brains are still fundamentally different.
At first glance, it sounds like the premise of a sci-fi thriller: artificial intelligence begins to mimic a human brain disorder. But that’s exactly what researchers from the University of Tokyo are exploring—not with the goal of sounding alarm bells, but rather opening new windows into both neuroscience and machine intelligence.
In a study published May 15, 2025, in Advanced Science, scientists at the International Research Center for Neurointelligence (WPI-IRCN) discovered a surprising parallel between how large language models (LLMs) like ChatGPT process language and how brains affected by Wernicke’s aphasia—a neurological condition—produce speech. Both systems can appear fluent, articulate, even confident. Yet under the surface, something is misfiring.
Wernicke’s aphasia is a condition in which individuals can speak in flowing sentences, but the words often make little sense. This is not due to a lack of vocabulary, but a disconnection between meaning and language production. The researchers noticed this oddly echoed the “hallucinations” of AI—responses that sound convincing but are factually incorrect or nonsensical.
“You can’t fail to notice how some AI systems can appear articulate while still producing often significant errors,” said Professor Takamitsu Watanabe, the study’s lead author. “That prompted us to wonder if the internal mechanisms of these AI systems could be similar to those of the human brain affected by aphasia.”
To investigate, the team turned to an unlikely ally from physics: energy landscape analysis. Originally used to map the states of magnetic materials, the method has been adapted to visualize brain activity. Imagine a ball rolling across a landscape: deep valleys represent stable states, shallow areas chaotic ones. When the brain (or an AI) is functioning well, the ball tends to settle in meaningful valleys. But in Wernicke’s aphasia—and, as it turns out, in some LLMs—the landscape is flatter and less structured. The result? The ball, or information signal, bounces erratically, creating fluent but disjointed output.
Applying this analysis to both brain scans and AI models, the researchers found strikingly similar patterns of instability and signal drift. In the case of aphasia, this leads to language that loses meaning; in LLMs, it leads to confident-sounding but incorrect responses.
This discovery is more than just an academic curiosity. For clinicians, it could offer a new method of diagnosing and understanding aphasia—not through visible symptoms alone, but via the dynamic patterns of brain activity. For AI developers, the implications are equally profound: by understanding the structural weaknesses that lead to ‘AI aphasia,’ engineers may be able to improve model architectures and reduce erroneous outputs.
Still, Watanabe is careful to draw the line: “We’re not saying chatbots have brain damage,” he notes with a smile. “But they may be locked into a kind of rigid internal pattern that limits how flexibly they can draw on stored knowledge.”
In other words, the analogy between artificial intelligence and human cognition isn’t perfect—but it’s productive. As LLMs become more ubiquitous, trusted with everything from medical advice to legal counsel, understanding the quirks of how they process and misprocess language is not just useful—it’s essential.
This research adds a compelling dimension to the broader conversation about AI reliability and transparency. It also reaffirms something more fundamental: the human brain, for all its imperfections, still holds mysteries that even our most advanced machines echo but do not fully replicate.
The intersection of AI and neurology might just be the next great frontier—not in replacing human minds, but in learning from their patterns, strengths, and even their faults.
Source: University of Tokyo
