Mentes y máquinas
Pruebas neuronales revelan que el cerebro humano predice secuencias de palabras como la IA
Anonymous
Scientists now have direct neural proof that the human brain predicts language the same way AI does. A study published on January 21, 2026 in Nature Communications found that the brain's language processing mirrors the layered, hierarchical structure of large language models like GPT-4 — not just loosely, but mathematically.
The research was led by Dr. Ariel Goldstein of Hebrew University, alongside teams from Google Research and Princeton University. They recorded brain activity while people listened to a 30-minute podcast — real speech, not lab sounds. Their finding: meaning in the brain "unfolds step-by-step — much like the layered processing inside systems such as GPT-style models," according to ScienceDaily.
Brain Maps Onto AI Layers With Striking Precision
The team used electrocorticography, or ECoG — a method that records signals directly from the brain's surface. This gave far sharper data than standard brain scans. Participants listened to natural conversation while electrodes tracked exactly which brain regions activated, and when.
The results were striking. "Deeper" layers of an AI model matched "deeper" processing regions in the brain, including Broca's area, which handles complex language. This wasn't mimicry. ScienceDaily reported that AI and the brain appear to have converged on the same mathematical solution for processing language — one found by nature, one built by engineers.
Princeton's 100 Hours of Data Backs Up the Finding
Prof. Uri Hasson's lab at Princeton University provided a key dataset: 100 hours of natural conversation recorded via ECoG. This scale was critical. Earlier studies used artificial sequences — think "beep-beep-boop" — which couldn't capture how the brain handles real speech. According to Max Planck Institute, that gap had long blocked a definitive proof.
A separate fMRI study with 304 participants confirmed the same pattern. AI can now identify different types of active neurons based on their predictive signals with 95% accuracy, according to Psychology Today. Dr. Mariano Schain of Google Research led the technical work of aligning AI embeddings with those neural signals.
Not Everyone Agrees: The Brain May Be More Complex
Not all scientists are convinced the comparison holds up. On April 21, 2026, NYU researchers published a study in Nature Neuroscience pushing back. Dr. David Poeppel argued that "the human brain makes predictions by grammatically grouping words" — a very different process from how LLMs predict the next single word in a sequence.
Poeppel's point is that LLMs treat every word with equal weight. The brain, by contrast, builds a grammar-based hierarchy — prioritizing structure over raw probability. Other linguists add that framing the brain as a "simple AI" risks missing the brain's deeper ability to build a full model of the world, not just predict text.
Why This Matters: Speech Restoration and Greener AI
The real-world stakes are high. By understanding the brain's predictive word patterns, researchers can now decode intended speech from neural signals with over 70% accuracy, according to Reddit r/science. This is already being used to build brain-computer interfaces for patients with ALS or aphasia — people who have lost the ability to speak.
On the tech side, researchers at KAIST point out that the human brain runs on just 20 watts of power — less than a light bulb. If AI is built to mimic the brain's predictive coding, the result could be systems that are exponentially more energy-efficient, according to Muy Interesante. In March 2026, KAIST published a further study showing the brain uses a "thinking twice" correction mechanism — yet another feature future AI models may adopt.
Sentiment
NeutralNEGATIVENEUTRALPOSITIVE
Publishers
8
Articles
0
Reach
8
Topics
A
AnonymousAuthor
