Four Filters on One Path | Visual AI Learning Guide

Look at the path running left to right, from a designed molecule to a medicine a patient takes. The four barrier categories sit on that path as gates, not as four separate rooms. A candidate passes through the scientific gate, then the clinical gate, then the economic gate, then the regulatory gate. Each gate can close on its own, and closing any one of them ends the journey for that candidate. The important consequence is that the number of medicines coming out the far end is set by the narrowest gate, not by how good the average gate is. If you make molecule design much better, you push more candidates up to the first gate, but they still have to get through the slow, expensive clinical gate and the evidence gate after it. That is why a real improvement in one area often produces almost nothing visible in the clinic.

0:00 / 0:00

Auto-play

A drug candidate does not face the scientific, clinical, economic, and regulatory barriers as separate problems to be solved in any order. It meets them in sequence along a single path, and each one can end the journey.

The scientific filter asks whether the predicted molecule actually behaves well in a living body. The clinical filter asks whether it produces a measurable benefit in humans without unacceptable harm. The economic filter asks whether someone will pay for the years of work required to find out. The regulatory filter asks whether the resulting evidence meets the standard required for approval, and whether the design process can be explained and reproduced.

Because these filters are sequential rather than parallel, the path is limited by whichever filter is currently tightest. Improving molecule design does not help if trials remain slow and expensive; making trials cheaper does not help if the evidence standard cannot be met by an opaque design process. This is why a single breakthrough in one area tends to produce little visible change: the candidate simply arrives at the next filter and waits.

A useful way to hold this together is to think of the path as having a throughput set by its narrowest stage, not by its average stage. Progress in AI design raises the number of candidates entering the path, but the number of medicines leaving it is governed by the slowest and most selective stage downstream.