Artificial intelligence will transform human health, but not because we can simply feed biology into a superintelligent algorithm and wait for cures to emerge. That's the counterintuitive message from Daphne Koller, founder and CEO of insitro, a drug discovery company backed by Andreessen Horowitz (a16z). In a detailed essay published on a16z's platform, Koller challenges what she calls the "AI Magic Wand" narrative that has captivated Silicon Valley: the idea that powerful enough AI will automatically unlock treatments for the hundreds of millions of people currently without effective medicines.

The gap between AI's capabilities and biology's complexity is vast. Koller, who has spent nearly 30 years at the intersection of artificial intelligence and biology, explains that the tech industry has latched onto an intoxicating but fundamentally flawed promise. "The logic is seductive," she writes. "The human body is a system we can already read from and write to, so like other knowledge problems, a powerful enough AI should be able to solve disease." But this assumption collapses under scrutiny.

Where Is AI Actually Helping in Drug Discovery?

The explosion of AI tools in drug discovery has been real and impressive. Following breakthroughs like AlphaFold, a landmark artificial intelligence system that predicted protein structures, researchers have developed AI capable of designing novel proteins, small molecules, RNA therapies, and even gene therapies at unprecedented speed. The problem is where this innovation is concentrated.

The vast majority of AI work in drug discovery has focused on what Koller calls "mechanism-to-drug," the stage where scientists already know what biological target they want to hit and need to design a molecule to hit it. AI excels here. But this represents only one of three essential stages in drug discovery:

• Disease-to-mechanism: Identifying a biological pathway or molecular interaction where therapeutic intervention will actually alter the course of disease in humans
• Mechanism-to-drug: Creating a molecular intervention, such as a small molecule or antibody, that achieves the desired effect with acceptable safety and pharmacological properties
• Drug-to-patient: Designing a clinical development program that identifies the right patients and assesses both beneficial and adverse effects

AI's dominance in stage two has created a false sense of progress. "AI will certainly generate new and better molecules at an unprecedented rate," Koller notes, "but will it generate drugs that unlock diseases for which there is currently no meaningful treatment?"

Why Understanding Disease Remains the Real Bottleneck?

Here's where the narrative breaks down. More than 90 percent of drugs that enter clinical trials fail, a dismal statistic that has barely improved in several decades. In the vast majority of cases, the molecule itself was engineered well. The mechanism it targeted was wrong. "We are doing a pretty good job at manufacturing keys," Koller explains, "but they are generally for the wrong locks".

"The real bottleneck in making a novel medicine is disease understanding: identifying a biological mechanism whose modification actually changes the course of disease in patients. That, far more than molecular design, is where drug discovery succeeds or fails," Koller stated.

Daphne Koller, Founder and CEO of insitro

The industry's response to this uncertainty has been destructive. There are currently 38 biological targets with over 50 drug programs each, representing a concentration of capital on the few mechanisms where researchers have strong conviction. Meanwhile, the number of novel targets the industry advances each year has collapsed from approximately 100 in 2015 to about 30 in 2024. This represents a profound failure to help the hundreds of millions of people for whom medicine currently offers nothing.

What Would It Actually Take to Bridge the Biology Gap?

A second wave of AI optimism has emerged around large language models, LLMs, which are AI systems trained on vast amounts of text to recognize patterns and generate human-like responses. The theory goes that these systems, with their superhuman reasoning capabilities, will connect dots across the published scientific literature and reason their way to new mechanistic hypotheses. This too rests on a flawed assumption.

The assumption is that the scientific community has already collected, or will soon collect, enough data about human biology to contain the answer. We just need better reasoning to extract it. But human biology is staggeringly complex. It spans multiple interconnected layers, from DNA to proteins to cells to entire organisms. Individual components respond dynamically to subtle changes in related components or the environment. Moreover, biology wasn't engineered; it is the result of billions of years of evolution, producing staggering variation across countless genes, cell types, and cellular states.

"There is too much of it, too idiosyncratic, to reason about in the abstract," Koller writes. "You have to measure it." The largest cell atlases assembled to date, spanning hundreds of millions of cells, remain orders of magnitude too small to cover the biological space that needs to be mapped. Until recently, these atlases held almost no causal, perturbational data, the measurements most critical for understanding what an intervention would do in a living system.

How Can AI and Biology Actually Work Together?

This doesn't mean AI won't transform drug discovery. Rather, it means the path forward requires a fundamentally different approach. The real opportunity lies in using AI to derive novel insights from properly measured biology, not to reason about biology we barely understand.

Koller's argument carries particular weight because insitro itself is built on this principle. The company focuses on measuring biology at scale, using high-throughput cellular experiments and machine learning to identify disease mechanisms that have been missed by traditional approaches. This is unglamorous work compared to the narrative of superintelligence solving disease, but it's where the actual breakthroughs will come from.

The stakes are enormous. Only about one-quarter of diseases have approved therapies, and most of those merely slow disease rather than stop it. For the majority of human illness, medicine still has little to offer. AI could help change that, but only if the industry stops chasing the magic wand narrative and starts doing the hard work of measuring and understanding human biology first.