There’s a peculiar possibility hiding in medical research: thousands of approved drugs sitting on pharmacy shelves might already treat diseases we think have no cure. We just don’t know it yet.
According to BBC reporting, scientists at the UK Dementia Research Institute in Edinburgh are using AI to hunt for these hidden solutions. They’re analyzing patient voice recordings, eye scans, and lab-grown brain cells to identify whether existing medications could be repurposed for neurological conditions like motor neurone disease (MND), Parkinson’s, and dementia.
The logic is simple but compelling. Around 1,500 drugs have been developed and approved to treat various conditions over decades. The brain is so complex that we’ve likely never tested most of them against neurodegenerative diseases. What if one of them works?
Racing Against Time
Steven Barrett, a trial participant diagnosed with MND a decade ago, knows firsthand why this research matters. According to BBC reporting, Barrett describes MND as “a horrible disease” that “strips you of who you are” and “rips any sense of future that you may feel that you had planned.” He was looking forward to retirement after a career in the civil service when numbness in his leg signaled the onset of the condition.
Barrett is part of MND-SMART, a trial that tests multiple drugs simultaneously rather than the traditional placebo-controlled approach. As he told the BBC, his motivation extends beyond his own health: “For me the research is much more than taking a tablet, it’s taking a tablet with the intention of delivering outcomes, that may or may not help me but help others.”
That’s the human pressure underneath this scientific work. Drug discovery typically takes over a decade from concept to market. When you’re living with a degenerative neurological condition, a decade might be too long.
How the Machine Learning Works
The Institute’s process is remarkably sophisticated. Researchers gather iris scans, voice recordings, and blood samples from volunteers. They use these samples to cultivate stem cells into groups of brain cells called neurones. Then comes the algorithmic muscle: robots and computers powered by specialist machine learning algorithms test existing drugs on multiple batches of those neurones, looking for compounds that could convert a disease signature into a healthy one.
Prof Siddarthan Chandran, the Institute’s chief executive, told the BBC that this combination of Technology and biology represents a sea change in research capability. “A combination of AI and new technologies mean we can now do things which would have been unbelievable when I was at medical school,” he said.
The advantage of testing existing drugs is obvious. They’ve already cleared safety hurdles. Regulatory approval is simpler. Deployment is faster. In theory, this should accelerate the path from discovery to treatment by years.
The Setbacks Matter Too
But there’s a cautionary tale embedded in this optimism. According to BBC reporting, a recent review of lecanemab and donanemab, once promoted as “breakthrough” drugs for Alzheimer’s, found they didn’t produce meaningful benefits for patients despite slowing disease progression. The review examined 17 studies involving over 20,000 volunteers of amyloid-removing drugs, and the conclusion sparked backlash from other scientists.
This matters because it shows that even when AI and algorithms point toward promising candidates, real-world outcomes can disappoint. Slowing progression isn’t the same as stopping it or reversing it. The gap between laboratory prediction and clinical reality remains substantial.
Prof Chandran remains confident the field is approaching an inflection point. Similar research at MIT has used generative AI to identify novel antibiotic compounds, while Harvard researchers developed a neural network model called TxGNN to surface existing drugs for rare diseases. The momentum is real, even if individual breakthroughs remain fragile.
The Paradox of Complexity
The fundamental puzzle here is frustrating in its elegance. The brain is the most complicated organ in the human body, which means we’ve historically struggled to study it with precision. We couldn’t crunch enough data fast enough. We couldn’t run enough simulations. AI changes that calculus. Suddenly, patterns emerge from noise. Unexpected connections surface. Hidden solutions become visible.
But complexity cuts both ways. The brain’s sophistication also means that a drug working brilliantly on cultured neurones might fail catastrophically in a living patient. Prediction models, no matter how sophisticated, deal in probabilities, not certainties. The machine can suggest. Humans must still verify.
The real question isn’t whether AI can accelerate drug discovery. The research suggests it can. The harder question is whether we can maintain realistic expectations as these tools proliferate. Every accelerated discovery carries the risk of accelerated disappointment. What happens when the algorithm points toward a promising lead, we run trials, and it doesn’t pan out?
That’s the tipping point Prof Chandran mentioned, and it’s worth sitting with: not just the potential for breakthroughs, but the discipline to accept that some promises won’t materialize, even with intelligence artificially enhanced.
