We should incentivize lower p values, or -log(p), with necessary penalization for p-hacking. This would still allow for people to publish moderate p-value studies, but generally should increase the concentration of higher quality studies. Also, I definitely agree that the sheer volume of scientific studies is not necessarily a good thing. It can really make things opaque. This is not to disparage researchers, but a lot of research does seem largely redundant until someone comes along and does a meta-study. The time-scale for these meta-studies to come out is the issue then, so I see room for a good machine learning mining tool for research, but instead, the meta-heuristic should be to continue to push p-values rather than the current paradigm of increasing quantity of studies.
The human temperature analogy is so apt. Haha... Yes, researchers are smart people but caught up in a machine. This is such a necessary lecture for anyone desiring to enter the academia.