Artificial intelligence is now beginning to translate that silent storm into words, images and even sound.

According to a report published in the BBC, A 52 year old woman could not speak clearly, not since a stroke left her paralysed 19 years ago. Yet sentences were appearing in front of her, formed from thoughts she never voiced aloud.

Known only as participant T16, she had a tiny grid of electrodes surgically placed in the front part of her brain. As she imagined speaking, the electrodes captured patterns of neural activity. An AI system then converted those signals into written text in real time. She was part of a study at Stanford University, alongside three patients with amyotrophic lateral sclerosis, or ALS. The goal was bold: translate thoughts directly into language.

It was the closest science had come to something resembling mind reading.

From thought to text

The results, revealed in August 2025, marked a turning point. Soon after, researchers in Japan introduced a separate technique sometimes called “mind captioning”. By combining three AI systems with non invasive brain scans, they were able to generate detailed descriptions of what a person was seeing or imagining.

Together, these breakthroughs are reshaping neuroscience. They offer new ways to help people who cannot communicate and raise the possibility that brain based technologies could one day alter how we interact with one another.

Maitreyee Wairagkar, a neuroengineer at the University of California, Davis, believes commercial applications are close. Companies including Neuralink are already working to move brain chips from laboratory settings into everyday use. According to Wairagkar, widespread deployment could begin within a few years.

A long scientific journey

Brain computer interfaces, or BCIs, are not new. In 1969, neuroscientist Eberhard Fetz showed that monkeys could learn to control a meter needle using the firing of a single neuron, if rewarded with food. Around the same time, Spanish scientist Jose Delgado demonstrated remote brain stimulation in animals.

For years, BCIs have helped users control robotic limbs or computer cursors by decoding movement related signals. But translating speech or complex thought has proved more difficult. Early research focused on animals, and speech cannot be meaningfully studied in monkeys.

In 2021, researchers at Stanford University reported that a paralysed man could write by imagining drawing letters in the air. He managed 18 words per minute. Natural speech averages around 150 words per minute, so the gap remained large.

In 2024, Wairagkar’s team advanced the field further. They converted attempted speech from a man with ALS directly into text at around 32 words per minute with 97.5 percent accuracy. It was the first time such technology approached practical daily communication.

How the system works

These systems rely on microscopic electrode arrays implanted on the brain’s surface. The arrays record patterns of neural firing. Machine learning algorithms then identify patterns linked to specific phonemes, the smallest units of speech.

The comparison often made is to voice assistants. But instead of processing sound waves like Amazon Alexa, the AI interprets raw neural signals.

Capturing inner speech

One major limitation remained. Most systems required patients to attempt to speak, even if physically unable. That effort activates the motor cortex, making decoding easier. But it is slow and exhausting.

Researchers at Stanford University wanted to know whether pure internal speech, silent thoughts, could also be captured.

They designed tasks that encouraged participants to count silently in their heads. The system detected traces of number words passing through the motor cortex. When participants imagined full sentences, accuracy reached up to 74 percent in real time. In open ended prompts, however, the output often became nonsense.

The findings suggest that inner speech activates brain networks similar to those used for spoken language, though the signals are weaker.

Beyond plain words

In 2025, Wairagkar’s team made another leap. They decoded not only words but also vocal features such as pitch, rhythm and intonation. Their system produced audible speech from an ALS patient with severe motor impairment.

The participant could raise his pitch to ask a question or vary tone while singing simple melodies. About 60 percent of the output was judged understandable. Although imperfect, it hinted at future systems capable of restoring expressive communication, not just flat text.

Progress may depend on scaling up. Current systems sample only a few hundred neurons. The human brain contains billions. More electrodes and improved hardware could dramatically enhance clarity and speed.

Researchers are also examining other brain areas, including the superior temporal gyrus, which processes sound and may contribute to inner speech. This could help patients whose motor cortex is damaged but who still comprehend language.

Rebuilding what we see and hear

While some scientists focus on restoring speech, others are decoding vision and sound.

In visual reconstruction studies, volunteers view images while undergoing functional magnetic resonance imaging. AI systems then attempt to recreate what the person saw. Advances in generative AI, including models such as Stable Diffusion, have greatly improved results.

In 2023, Yu Takagi of the Nagoya Institute of Technology used this approach with a dataset from the University of Minnesota. The AI often produced recognisable versions of the original images, though some objects remained challenging.

Research has revealed that the occipital lobe handles low level features such as colour and layout, while the temporal lobe processes higher level meaning and object recognition.

Takagi’s later work extended to music. Using a proprietary algorithm from Google, his team attempted to reconstruct music from brain scans. Although quality lagged behind image reconstruction, they captured basic musical character and genre.

Interestingly, music appears to be processed differently from images. Unlike visual information, high level meaning and low level features of music are not clearly separated in the brain.

What comes next

Potential applications stretch far beyond medical use. Scientists speculate about recreating dreams, understanding hallucinations in psychiatric conditions, or even enabling direct brain to brain communication. Ethical and human rights concerns remain unresolved.

As for entertainment applications that stimulate artificial sights or sounds directly in the brain, researchers advise patience. Technically possible in theory, such developments are likely at least a decade or two away.

For now, the most immediate impact is human. For people silenced by injury or disease, AI powered brain interfaces are beginning to return something precious: the ability to be heard.