A typical child knows about one child with cancer. Back-of-the-envelope, the number a child knows would be (incidence rate of childhood cancer) * (typical K-8 size), but doubled since they observe all grades ahead and behind them. Incidence rate is about 20 per 100,000 and we might assume a typical K-8 is about 2000 students, so (20 / 100000 * 2000 * 2) ~ 1.
The first figure shows an order of magnitude decrease in mortality over the last few decades from childhood cancer. The average child growing up in the 70s would know a child that died from cancer, and today they would not!
Basic research has a lot in common with startups. The unicorn rate is <1%, the best new ideas sound like bad ideas, and nearly all value from the best ideas is locked in the long-term future. The ideal startup/scientific program failure rate is not 0%, and could be 95%.
The current research system has serious problems, but we need accurate criticism to build a better future. "YC is all wasteful spending; why doesn't YC just choose to only fund the hits?" is absurd, but somehow we allow this argument when discussing NSF/NIH/DOE/DARPA.
Bush's report was submitted in the months between the German surrender and the dropping of the atomic bombs on Japan. It was a remarkable context for scientific optimism, given the huge impact of science and technology on the war effort. Radar developments are obviously an unalloyed good for shooting down German bombers.
Scientific progress is arguably even more relevant today, but the vision of the future has changed for the average American. Many achievements of the ensuing eight decades (vaccines eliminating polio and measles, nuclear energy, computing and social networks, cheap solar and wind, fracking, automation and artificial intelligence, cheap spaceflight) are viewed with fear and suspicion by large, different fractions of the US. Unfortunately, ceding scientific leadership to other powers does not reduce the destabilizing force of progress -- but I think there's some explanatory power here in simultaneously shutting down science and pursuing economic isolationism.
Passenger jet engines, nitrogen fixation, silicon fabs -- these don't just magically appear. It takes careful reasoning and years of exploration in the vast ocean of ideas. Most places in this ocean are not useful, so it's helpful to sail faster.
I'm an astrophysicist. I love the idea of shaking up how we do science, but MOND models really aren't that compelling for a practicing scientist (I would love to be wrong). I periodically consider working a modified gravity project, but it never seems that interesting. Maybe I can lay out my personal impressions for you in a software analogy.
There are simple situations where physics can be reasonably distinguished from the noise. Think Newton watching an apple fall from a tree -- these are test harnesses with a debugger and careful control of the environment. There are also astrophysical situations where it's very hard to squeeze insight out of the chaos. Instead of looking at an apple to derive Newtonian gravity, imagine instead trying to figure gravity out by looking at a flock of birds. There are tricky things in the way of our understanding, like lift, turbulence, and biomechanics. We hope to understand the bad situations someday, but right now it's tough. Instead of the nice test harness, these situations are like debugging through cryptic, un-reproducible user complaints. What's their OS? Do they have the right drivers installed? Do they have all of their ports blocked for some reason?
In astrophysics, one nice system is the Universe at very large scales: echos of the big bang, the clustering of matter, and the formation of structure. We have great data these days about the large scales, and general relativity (GR) + dark matter (DM) is a very predictive model here. To our chagrin, the data always matches the theory. It takes 5 or 6 parameters, but the model has withstood huge improvements in data quality without really changing since the 90s. The other nice situation is the small scales like our own solar system: we can measure things extremely precisely at home, and again GR works remarkably well. There are a few other good situations involving objects like pulsars. Finally there are the hard situations: intermediate scales that involve the messy physics of star and galaxy formation. We don't really know how these processes work, but we can cobble together simple models with DM that sort of match the data.
Modified gravities in the literature always seem to act exactly like general relativity + dark matter in the clean, understandable situations, although it takes various contortions (screening) for these models to do this. Modified gravity models seem to only act differently in the bad arenas that are hard to understand, when there trickier issues like galaxy and star formation. That's deeply suspicious, and really weakens the value proposition of these alternatives. It's like introducing a software feature to fix something that never happens on the dev machine, based on mysterious user reports that you can only half decipher.
The first figure shows an order of magnitude decrease in mortality over the last few decades from childhood cancer. The average child growing up in the 70s would know a child that died from cancer, and today they would not!