There is a second reason drug discovery is hard, and it is not economic but mathematical. A drug-like molecule is built from a small set of atoms — mostly carbon, hydrogen, oxygen, and nitrogen — connected in a particular arrangement. Because those atoms can be connected in an enormous number of ways, the set of possible drug-like molecules is not just large; it is estimated to be on the order of \(10^{60}\) or more. For comparison, the number of molecules that have ever been synthesized and recorded is on the order of \(10^{8}\) to \(10^{9}\).
The gap between those two numbers is the core difficulty. Even if every compound ever made were tested against a target, that would sample a vanishingly small fraction of the possible space. Exhaustive search is not merely impractical; it is impossible in principle. The same problem appears on the biological side: a protein can be described by the positions of thousands of atoms, and the number of conceivable protein shapes and binding pockets is likewise far beyond enumeration.
This is why drug discovery is fundamentally a search problem under extreme scarcity of information. Any method that can propose promising regions of the space to test, rather than testing blindly, has potential value. That framing is what makes the question of AI relevant at all.