Neuromodulation – influencing the brain’s activity to treat disorders – has undergone a dramatic transformation over the past few years. Today, it’s being heavily shaped by technological advances, such as artificial intelligence (AI), which are changing the way we conceive of and interact with the human brain.

Dandelion Science, a neurotechnology company based in both Switzerland and the US, is at the forefront of exploring these new possibilities. Fresh from announcing a major new partnership with the Wyss Center for Bio and Neuroengineering, as well as a new grant from Innosuisse to further Dandelion’s ongoing collaboration with the École Polytechnique Fédérale de Lausanne (EPFL), Adam Hanina, Dandelion’s CEO, spoke to us about the future of neuromodulation.

The urgent need for better, more precise therapies

Over 30% of people develop neurological conditions in their lifetime, according to the latest research from the World Health Organization – and problems with functions like visual processing are even more prevalent. Despite this widespread need, millions of affected individuals either don’t receive the treatment they need or else receive treatment that is ineffective.

Adam explained that the brain has traditionally been treated like any other biological organ, meaning that mental health disorders like depression and anxiety, as well as neurological conditions such as dementia and migraines, have often been addressed with pharmacological treatments. This approach, referred to as chemical neuromodulation, uses drugs to alter brain activity by interacting with neurotransmitter systems. But while medications have their merits, their effects are often diffuse, impacting both targeted and non-targeted brain regions. It’s clear that better therapies are urgently required.

An expanding neuromodulation landscape

The diagram above presents a detailed hierarchy of neuromodulation techniques, ranging from broad, untargeted interventions to cutting-edge computational methods that offer precise control over brain circuits.

The next tier after chemical neuromodulation in the hierarchy involves energy-based techniques, such as electrical, magnetic or ultrasound stimulation. These methods aim to alter neural activity by delivering external energy to specific areas of the brain. Techniques like transcranial magnetic stimulation (TMS) or deep brain stimulation (DBS) use electrical or magnetic fields to target particular regions of the brain, while ultrasound can influence deeper brain areas without invasive surgery. TMS is used to treat major depression while DBS has proved to be an effective, albeit highly invasive, treatment for movement disorders associated with Parkinson’s disease, essential tremor, dystonia and other neurological conditions.

Compared with drugs, these energy-based approaches provide more regional specificity, as they focus on particular circuits, but the types of stimulation they involve tend to be limited. As a result, they are not always able to modulate highly complex neural dynamics and/or may affect regions adjacent to the target, potentially causing unintended outcomes.

Progressing upwards, brain–computer interfaces (BCIs) offer a direct, hardwired link to the brain. These systems sense neural activity in real time and provide feedback to the brain, facilitating a loop where the brain’s responses can be influenced and adjusted dynamically. BCIs use electrodes or sensors to interact directly with neural circuits, allowing for more precise interventions. State-of-the-art BCIs typically have up to 100 channels – and some currently being developed have up to 1000. Indeed, advanced systems like Neuralink and Synchron enable users to control electronic devices such as smartphones and laptops through thought.

Still, while BCIs can effectively sense and manipulate neural activity within a confined region, they are not yet capable of widespread, finely tuned neuromodulation across the brain. Indeed, they have shown some promising results when it comes to enhancing neurological rehabilitation and treating conditions like epilepsy, but they remain hampered by their spatial resolution and the technical challenges involved in real-time brain feedback. In addition, a major limitation is the need for invasive hardwiring, which constrains their ability to influence broad areas of the brain, increases cost and reduces accessibility.

Notably, each of these approaches has been conditioned by and remains dependent on the technologies available. Whatever the modality of delivering neuromodulation stimuli to the brain, the challenge is how to identify stimuli with real therapeutic potential; after all, the aim of all neuromodulation techniques is to take advantage of brain plasticity by stimulating healthy circuits and retraining them to take over functions from circuits that have suffered damage. Given the infinite universe of neural dynamics, which are complex and operate at rapid time scales at high dimensions in space, a manual search for useful stimuli is clearly impossible.

But with the emergence of generative AI, the field of neuromodulation is, as Adam put it, ‘experiencing a genuine revolution’.

Enter AI: The path toward generative neuromodulation

Technologies pioneered by OpenAI have proven highly effective in tackling complex challenges, with ChatGPT generating realistic text and Sora creating videos from text commands.

These groundbreaking tools also hold immense potential in neuroscience, a field dedicated to understanding and influencing the intricate ways the brain processes information. For example, generative AI is now being applied to create more sophisticated models of brain function, offering a deeper understanding of disease severity and classification.

As Adam explained to us, generative AI is one of the cornerstones of what Dandelion Science terms ‘Generative Neuromodulation’. Instead of using text commands, generative AI, guided by neural objectives, can create complex sensory stimuli for therapeutic purposes. This approach harnesses cutting-edge AI to explore the brain’s spatial and temporal dynamics – how neurons synchronise their activity across time and space to carry out tasks. In Adam’s words, ‘we finally have the computational power to systematically search for and deliver high-dimensional stimuli that can control neural dynamics’.

Through this method, high-dimensional stimulation can be delivered either non-invasively through the body’s natural sensory pathways, such as the eyes or ears, or potentially even through BCIs. By providing inputs that closely mirror the brain’s own computational processes, this technology may offer a more precise and effective way to influence neural function.

Adam compared this to noise cancellation: ‘Either you cancel background noise by drowning it out – but by doing so you also interfere with the music you are listening to – or you take a precision approach. In our case, we are building a data-driven platform that can generate therapy from the dysfunction itself.’

Generalisability and the future of accessible therapy

One of the most exciting aspects of AI-driven generative neuromodulation is its potential to be widely applicable. Many brain disorders share similar underlying neural processes, so by learning how to influence these processes, we have the potential to develop therapies that work for a broad range of conditions.

As Adam emphasised, disorders that once required invasive therapies may be treatable with non-invasive stimulation, delivered through everyday devices like smartphones or wearables. As he put it ‘this scalability could significantly increase access to care by offering sophisticated neuromodulation treatments to millions of people’.

There are important regulatory and ethical considerations to navigate as these neuromodulation technologies continue to advance. Regulations play a vital role in protecting patients, and as these innovations become more widely adopted, it’s essential to establish clear and responsible guidelines around areas like data ownership and the extent of neural access. As Adam put it, ‘it’s crucial to have safeguards in place in tandem with ensuring patients benefit from these transformative treatments.’

AI and neuroscience: A symbiotic evolution towards brain-inspired innovation

As AI transforms neuroscience by helping to decode the brain’s complex systems, it’s likely that the brain’s own efficient computational mechanisms will in turn inspire new advances in AI. Adam outlined that ‘the brain operates on just 20 watts of power, which presents a model of efficiency for AI researchers to emulate’. In fact, the fields of neuromorphic computing and AI modelling are already embracing brain-inspired architectures.

In this vein, Adam shared his ultimate vision: ‘Neuromodulation is, at its core, a tool to explore and reshape how the brain functions. As our ability to investigate grows more sophisticated, we move closer to unlocking the brain’s programming language. Beyond the profound therapeutic and economic implications, in a world often divided, I believe that discovering a universal operating system of the brain could remind us of our shared humanity – that despite our differences, we are more alike than we may realise.’