The way the argument about usefulness of animal experimentation for medical purposes is put by its defenders is often fallacious for a very simple reason: it is on qualitative, and not quantitative, grounds.
(In addition, it goes against the nature of advanced science, historically, to reason in qualitative and not quantitative terms. The most developed sciences, like physics, and fields of science reach their maturity by going from being qualitative to being quantitative, especially in recent times.)
When people say ‘drug x was developed through animal experiments’ or ‘treatment y was found researching on animal models’, the point is not even whether what they say is true or not.
The relevant question is: how many (how many millions, more likely) animals were used, how many experiments that had no useful result were necessary before arriving at that particular single result?
It is a statistical problem that we should address.
Because, had we used a different method of research, the statistical utility (I mean, for example, the percentage of successes) could have been higher.
We must always use this yardstick, this criterion for comparison.
(This is the way that control groups are used in tests: I’m here transferring a scientific technique to a meta-scientific context.)
It may very well turn out that, when compared with other methods already in existence or that we know could be developed, the percentage of successes in medical animal experimentation, among the number of all experiments performed, is extremely low.
It could also turn out that the corresponding percentage of misleading results (eg penicillin, or the role of smoking in lung cancer) or downright deleterious effects is higher than it needs to be.
It’s certainly well worth investigating along these lines, from now on.