Artificial Intelligence Begins to “Read the Brain”

For a long time, the possibility that a machine could interpret what happens in the brain belonged to the realm of science fiction. Today, different lines of research are beginning to transform that possibility into an experimental reality. 

The starting point is that neuronal activity produces electrical signals and magnetic fields that can be recorded from outside the body. In contactless capacitive electroencephalography (EEG), a conductive sensor is placed at a short distance from the skin, separated from it by air, and detects variations in the brain’s electric field through capacitive coupling. It requires neither gel nor direct electrical contact, while magnetoencephalography (MEG) records the extremely weak magnetic fields associated with neuronal activity. 

These signals contain information about brain processes that, although weak, noisy, and difficult to interpret, can be analyzed using machine learning and artificial intelligence.

In recent years, machine learning has made it possible to move beyond the simple classification of brain states toward the decoding of perceptual information. There are already studies using electroencephalography (EEG) signals to identify features of the images a person is viewing and even to generate images from those signals. A study published in 2023, for example, developed a method based on diffusion models—generative models primarily used for image generation—to reconstruct images from electroencephalography. Other studies published in 2024 used neural networks and diffusion models to directly transform electroencephalography signals into visual representations; one of them also demonstrated the possibility of applying the procedure to magnetoencephalography data. In 2026, the work known as NeuroVision continued this line of research through a deep-learning-based system for reconstructing images from electroencephalography.

Comparison between the real image and
the image obtained through Brain-IT (1)

This development is particularly significant because electroencephalography has much lower spatial resolution than functional magnetic resonance imaging. However, it offers an enormous advantage: it records brain activity with very high temporal resolution and can be used through relatively accessible, non-invasive devices. For this reason, EEG is one of the technologies of interest for future brain–computer interfaces. In this context, research conducted by the United States Defense Advanced Research Projects Agency (DARPA) becomes particularly relevant. Its Next-Generation Nonsurgical Neurotechnology (N3) program sought to develop an interface capable of establishing bidirectional communication with the brain without necessarily relying on conventional implants. What is particularly interesting is that the program was not based on a single physical principle. Different research groups explored approaches based on magnetism, electricity, ultrasound, and optics. Among the technologies investigated were magnetoelectric nanotransducers capable of converting electrical phenomena into magnetic ones and vice versa; ultrasound combined with optical techniques to access deep brain regions; optical detection of changes associated with neuronal activity; magnetic fields and ultrasound for inducing stimulation; and extremely sensitive magnetometers capable of detecting the tiny magnetic fields produced by neurons. The common goal was to achieve two complementary operations: detecting what is happening in the brain and, eventually, introducing information back into it.

Another major line of research is being developed in Israel by scientists at the Weizmann Institute of Science. The system known as Brain-IT uses functional magnetic resonance imaging (fMRI) recorded while a person observes images and, through artificial intelligence, reconstructs a visual representation of what the person is seeing. The work by Roman Beliy, Amit Zalcher, Jonathan Kogman, Navve Wasserman, and Michal Irani was accepted at the International Conference on Learning Representations (ICLR) 2026. The system uses a transformer to organize information from functionally related groups of brain voxels and subsequently employs a diffusion model to generate the reconstructed image.

For this reason, saying that this technology “reads the mind” is attractive from a journalistic perspective, but scientifically it is an oversimplification. The system cannot freely access a person’s thoughts. It reconstructs visual information from brain patterns recorded under specific experimental conditions. Nevertheless, the advance is conceptually extraordinary: a signal produced by the brain can be transformed into data; those data can be interpreted by artificial intelligence; and the result can once again be converted into a visual representation. More recent research is even exploring the possibility of obtaining linguistic descriptions and answers to questions about perceived images directly from brain activity.

In this way, a new technological architecture is beginning to take shape. First, the brain produces electrical and magnetic activity; then, sensors capture these signals; machine learning identifies patterns; artificial intelligence interprets them; and generative models can convert this information into images, words, or actions. Research conducted by the Defense Advanced Research Projects Agency adds the possibility of the reverse process: using physical principles such as magnetism, ultrasound, electricity, or optics to transmit information into the brain.

The real technological leap, therefore, does not simply consist of “reading the mind.” It consists of moving from one-way communication toward a brain–artificial intelligence circuit: detecting, interpreting, responding, and eventually stimulating. The brain ceases to be merely the organ that controls a machine and begins to become part of a communication system in which artificial intelligence can interpret neural signals and, in certain experimental technologies, return information to the nervous system. It is precisely at this convergence of neuroscience, nanotechnology, and artificial intelligence that a new stage in the relationship between the natural mind and the artificial mind may begin.

Bibliography

1. Roman Beliy,  Jonathan Kogman, Michal Irani  Amit Zalcher y Navve Wasserman. BRAIN-IT: Image reconstruction from FMRI via brain-interaction transformer.  Published as a conference paper at ICLR 2026.  arxiv.org/abs/2510.25976v2

2. D’Andrea A. L. (2026 ) Nanopsicología. La psicología del siglo XXI. Editorial Autores de Argentina. www.amazon.com/dp/B0GHLJGCK9

3. D’Andrea, Alberto L. (2026). Del ion al amor: correlatos electromagnéticos de la mente. Biotecnología & Nanotecnología al Instante.

https://infobiotecnologia.blogspot.com/2026/01/correlatos-electromagneticos-de-la.html