The algorithm watches, decides, and adjusts—learning as it goes.
In Porto, a neuroscientist is teaching artificial intelligence to listen to the brain and speak back to it — not through fixed commands, but through a living dialogue that adapts in real time. The NeuroSAFE project, funded through a Portuguese-American scientific partnership, seeks to develop algorithms that observe neurons firing, stimulate them thoughtfully, and learn from the response, much as any skilled practitioner learns from practice. The ambition is not merely technical: if the work succeeds, it could reshape how humanity treats some of its most persistent neurological afflictions, from Parkinson's disease to epilepsy, by replacing blunt, one-size-fits-all interventions with therapies that evolve alongside each patient.
- Neurological disorders like Parkinson's affect millions, yet current deep brain stimulation devices operate on rigid, pre-programmed schedules that cannot respond to the brain's constant change.
- Testing adaptive AI algorithms directly on patients is ethically and practically impossible, creating a critical gap between promising theory and safe clinical application.
- Aguiar's team is bridging that gap using brain-on-chip platforms — living neurons grown in the lab alongside sensors — giving the AI a controlled arena to practice, fail, and improve without risk to any human subject.
- A transatlantic collaboration between Porto and Carnegie Mellon University anchors the project scientifically, with roughly €100,000 in competitive funding providing the runway for early-stage discovery.
- The trajectory points toward personalized neurotechnologies — systems that learn a patient's unique neural patterns and adjust stimulation accordingly, potentially transforming treatment for Parkinson's, depression, tremor disorders, and epilepsy.
In Porto, neuroscientist Paulo Aguiar is developing artificial intelligence that can learn to communicate with the brain. His project, NeuroSAFE, recently secured roughly €100,000 through a competitive grant program run jointly by Portugal's Foundation for Science and Technology and Carnegie Mellon University — one of seven selected from 36 applications submitted to the CMU Portugal initiative.
The core idea is both elegant and technically demanding: build algorithms that watch neurons fire, decide when and how to stimulate them, and then refine their approach based on what happens next — all in real time. Unlike today's brain stimulation devices, which follow fixed schedules, NeuroSAFE aims to create something genuinely responsive, a system that evolves the way a musician listens and adjusts mid-performance.
Because experimental algorithms cannot be tested on patients, Aguiar's team will work with brain-on-chip platforms — small laboratory systems where living neurons grow alongside sensors capable of recording electrical activity and delivering stimulation. These controlled environments allow the AI to practice and learn safely, occupying the crucial space between computer simulation and human trials.
The collaboration reaches across the Atlantic, with Yorie Nakahira at Carnegie Mellon's Department of Electrical and Computer Engineering leading the American side. For Aguiar, the funding is as much about building a lasting scientific partnership between Porto and Pittsburgh as it is about this single project — generating the preliminary data needed to pursue larger initiatives and develop computational tools for a new generation of personalized neurotechnologies.
Deep brain stimulation is already used to treat depression, tremor disorders, and epilepsy alongside Parkinson's. A system capable of learning and adjusting in real time could make all of these therapies more precise and more effective. The funding is modest, the work is early, but the direction is unmistakable: toward a future where artificial intelligence does not merely analyze the brain, but actively converses with it.
In Porto, a neuroscientist named Paulo Aguiar is building artificial intelligence that can learn to talk to the brain. The project, called NeuroSAFE, has just received roughly 100,000 euros in funding through a competitive grant program run jointly by Portugal's Foundation for Science and Technology and Carnegie Mellon University. It is one of seven projects selected from 36 applications submitted to the CMU Portugal initiative, and it represents a deliberate step toward therapies that could one day treat Parkinson's disease and other neurological disorders through targeted brain stimulation.
Aguiar leads the Neuroengineering and Computational Neuroscience group at the Institute for Research and Innovation in Health at the University of Porto. His team's goal is straightforward in concept but demanding in execution: develop algorithms that can watch neurons firing, decide when and how to stimulate them, and then adjust their approach based on what happens next—all in real time. The system learns as it goes, much like a musician who listens to their own playing and corrects course mid-performance. This adaptive quality is what makes the work novel. Most current brain stimulation systems operate on fixed schedules; NeuroSAFE aims to create something responsive, something that evolves.
Because you cannot test experimental algorithms directly on patients, Aguiar's team will use what researchers call brain-on-chip platforms. These are small laboratory systems where living neurons are grown and connected to sensors that can both record their electrical activity and deliver stimulation. Think of them as controlled environments where the algorithm can practice, fail safely, learn, and improve before any clinical application is ever considered. The platforms provide the crucial middle ground between computer simulation and human trials.
The collaboration extends across the Atlantic. Yorie Nakahira at Carnegie Mellon's Department of Electrical and Computer Engineering is coordinating the American side of the work. For Aguiar, the funding represents more than just resources for this particular project. It is a chance to deepen the scientific partnership between Porto and Pittsburgh, to generate preliminary data that could support larger initiatives, and to build computational tools that might eventually underpin a new generation of personalized neurotechnologies—systems tailored to individual patients rather than one-size-fits-all devices.
The broader context matters. Since 2017, the CMU Portugal program has funded 42 projects across various domains. This year's round of exploratory competitions, spanning both CMU Portugal and a parallel UT Austin Portugal initiative, is supporting 15 projects in total, with nearly 800,000 euros in combined funding. The money is meant to strengthen Portugal's international scientific standing and its capacity for technological innovation. In the specific case of the CMU Portugal program, the seven selected projects all fall within information technology—a narrow but strategically important field.
What Aguiar and his team are attempting sits at the intersection of neuroscience, engineering, and machine learning. If they succeed in creating algorithms that can safely and adaptively stimulate neurons, the implications extend beyond Parkinson's treatment. Deep brain stimulation is already used for depression, tremor disorders, and epilepsy. A system that learns and adjusts in real time could make these therapies more effective, more precise, and potentially safer. The work is early. The funding is modest by some standards. But the direction is clear: toward a future where artificial intelligence does not simply diagnose or predict, but actively communicates with the living brain.
Citas Notables
The project aims to develop new artificial intelligence methods to control neuron activity in a safe and adaptive way, with algorithms that observe continuously, decide when and how to stimulate, and adjust based on the response.— Paulo Aguiar, project leader
This funding is particularly important because it strengthens collaboration between i3S and CMU, generates preliminary data, and develops computational tools that could contribute to future intelligent and personalized neurotechnologies.— Paulo Aguiar
La Conversación del Hearth Otra perspectiva de la historia
Why does the algorithm need to learn? Can't you just program it to stimulate neurons in a fixed pattern?
Because brains are not machines with identical parts. Two patients with Parkinson's might respond differently to the same stimulus. The algorithm watches what actually happens when it stimulates, then adjusts. It's like tuning a guitar by ear instead of using a chart.
And the brain-on-chip—that's a substitute for testing on people?
Exactly. You grow real neurons in a dish, connect them to sensors, and let the algorithm practice. It fails in a petri dish, not in someone's head. You get clean data, no risk.
How long until this moves to actual patients?
That's the honest answer: we don't know yet. This funding supports the foundational work. If the algorithms prove safe and effective on the chips, then you design clinical trials. Years, probably.
What makes this different from existing brain stimulation devices?
Most of those devices run on a timer. They stimulate on schedule. NeuroSAFE watches the brain's response and adapts in real time. It's reactive, not just rhythmic.
And if it works?
Then you have a tool that could be personalized to each patient, that learns their brain's language. That changes what's possible for neurological disease.