Artificial intelligence is poised to transform how drugs are tested by creating digital versions of patients that can predict treatment outcomes before human trials begin. Rather than waiting years for results, pharmaceutical companies could use AI-generated synthetic control arms and predictive models to accelerate clinical trials, potentially addressing a decades-long productivity crisis in drug development .

Why Is Drug Development Getting Slower Despite More Investment?

The pharmaceutical industry faces a troubling paradox. Over the past several decades, research investment has climbed steadily, yet the number of new drugs approved per billion dollars spent has actually declined. This phenomenon, known as Eroom's law (Moore's law in reverse), reflects a fundamental challenge: biology remains poorly understood, making it difficult to predict whether a treatment will work before testing it on real patients .

The current drug development pipeline is grueling. A typical journey from discovery to approval takes roughly 10 years and costs enormous sums. Researchers must first identify a disease target, design a drug to hit that target, test it in laboratory settings and animal models, and then move to human clinical trials. The brutal reality: approximately 90% of drug candidates fail during clinical trials, the final and most expensive stage .

Meanwhile, global healthcare demand continues climbing due to aging populations, lifestyle changes, and better diagnostics. More people than ever need pharmaceutical treatments, yet the pipeline cannot keep pace with demand. This growing gap between need and supply creates what experts call unmet medical needs .

How Could AI-Powered Synthetic Patients Speed Up Drug Testing?

Rather than relying solely on traditional clinical trials with human participants, AI could generate synthetic control arms, essentially digital twins of patient populations. These simulated patients would be based on real-world data and biological understanding, allowing researchers to model how different groups might respond to a treatment before exposing actual people to experimental drugs .

Predictive modeling powered by AI foundational models, which are large AI systems trained on vast amounts of biological data, could help researchers understand disease progression and treatment outcomes with greater accuracy. This approach addresses the core problem that has plagued drug development: the inability to rationally predict whether a treatment will benefit patients before clinical testing begins .

Ways AI Could Improve Drug Development Efficiency

• Synthetic Control Arms: AI generates virtual patient populations that simulate how real patients might respond to treatments, reducing the need for large traditional control groups and accelerating trial timelines.
• Predictive Modeling: Machine learning models trained on biological data can forecast disease progression and treatment efficacy, helping researchers make smarter decisions about which candidates to advance.
• Preclinical Acceleration: Foundational AI models can streamline early-stage research workflows, compressing timelines before drugs ever reach human testing phases.

Dr. Jean-Philippe Vert, co-founder and chief scientific officer of Bioptimus, a company developing foundational AI models for drug discovery, explained the stakes clearly. With over 25 years of experience at the intersection of machine learning and life sciences, Vert has witnessed both the promise and limitations of traditional approaches .

"There's a famous quote in computer science, originating from Moore's law: year on year, computers get faster, have more capacity, and more memory, so information technology has been growing at a fast pace with exponential increases in capabilities. In drug discovery, people refer to Eroom's law, the inverse of Moore's. Productivity has been decreasing; for example, the number of new drugs per billion dollars spent is decreasing and we are getting worse at finding new drugs," said Dr. Jean-Philippe Vert.

Dr. Jean-Philippe Vert, Co-Founder and Chief Scientific Officer at Bioptimus

Vert emphasized that the core challenge lies in biology's complexity. Scientists can observe many phenomena, but they struggle to make rational predictions about disease evolution and treatment outcomes. This is where AI's pattern-recognition capabilities could prove transformative .

What Does This Mean for Patients and the Pharmaceutical Industry?

If AI-powered synthetic trials and predictive modeling succeed, the implications would be profound. Faster drug development timelines could bring life-saving treatments to patients years earlier. The current 90% failure rate in clinical trials could potentially improve if AI helps researchers identify promising candidates more accurately before human testing. Additionally, reduced trial timelines could lower development costs, potentially making drug development more economically viable for diseases affecting smaller populations .

The shift represents a fundamental reimagining of how pharmaceutical research works. Rather than treating drug development as a purely experimental discipline reliant on wet-lab testing and human trials, AI could help transform it into a hybrid approach combining modeling, technology, and engineering. This mirrors how biology itself has evolved since the human genome was sequenced in 2000, when the field suddenly needed computational tools to make sense of three billion genetic letters .

As healthcare systems worldwide grapple with rising demand and limited resources, AI-driven efficiencies in drug development could help address the growing gap between medical needs and available treatments. The coming years will reveal whether these tools can deliver on their promise to reverse decades of declining productivity in pharmaceutical innovation.