Almost Working: Why Artificial Intelligence Fails at the Threshold of Drug Discovery

Almost Working: Why Artificial Intelligence Fails at the Threshold of Drug Discovery

Artificial intelligence has transformed many areas of science. Algorithms can now predict protein structures, design new molecules, and explore chemical space at extraordinary speed. For a moment, it seemed inevitable that AI would finally unlock the long-standing bottleneck of drug discovery.

And yet the breakthrough has not arrived.

Despite remarkable advances in machine learning, robotics, and computational chemistry, clinical success rates have remained stubbornly unchanged. Many programs appear promising in early stages-statistically convincing, biologically plausible, technologically sophisticated-only to fail later in development.

Why?

In Almost Working, Larry Lim Kheng Cheong argues that the problem is not a lack of intelligence, data, or computing power. The deeper issue lies in the nature of biology itself. Artificial intelligence excels in smooth, continuous systems that can be optimized step by step. Living systems do not behave this way. They operate through fragile thresholds, context shifts, and irreversible biological decisions.

Drug discovery is not a gradual engineering problem. It is a landscape of cliffs.

Drawing on historical experiments, clinical failures, and the logic of complex systems, this book offers a new way to understand why the AI revolution in medicine has been slower than expected-and what it may take to move forward.