When Gordon Moore predicted in 1965 that computing power would double every two years even as costs halved, he captured the astonishing progress possible in electronics. Yet in medicine, the opposite has been happening. Despite vast investments and the proliferation of AI tools, the field of drug discovery is still slow, expensive and high-risk. 

There are several key reasons why AI models struggle to discover truly novel drugs. First, human biology is immensely complex and full of unknowns. Drugs must not only hit a target but also avoid unintended harm and behave safely in humans - something even the most advanced AI has difficulty modelling.

Second, the data available to train AI is limited, fragmented, biased or proprietary. Integrating diverse datasets from chemistry, biology, genomics and clinical trials is a major obstacle. 

Third, many models act as black boxes: they may suggest a molecule, but researchers and regulators still need to understand why it should work, whether it can be synthesized, and how it will behave in the human body. 

In short, while AI has the potential to transform pharmaceutical innovation, it is not yet a silver bullet. For real breakthroughs - new therapies rather than incremental tweaks - we need richer data, better integration of biology and chemistry, transparent models, and time. The promise remains, but the journey is still very much underway.