Despite the common trope that software/hardware engineers think they understand everything, automation can make exponential development possible in bio. I know someone who works at a biotech company, and the stories I hear lead me to believe that traditional thinking in e.g. medicine and insurance, and from colleagues, really is a bottleneck. Maybe exponential growth will never apply to the number of diseases cured, but it can and should apply to the hardware and software that facilitates the next iteration of biological discovery.

It's not that software/hardware engineers think they understand everything, it's that the absolutely common idea "my way of thinking about understanding things is appropriate in this other domain" breaks down very fast.

I'm not speaking about running hospitals or dancing with insurance companies. I'm speaking about much earlier in the pipeline: fundamental, blue-sky biological research is fundamentally different from software. The reason is that we are not studying designed systems; the effect is that there is so much that we don't know that we don't know that things that sound easy are in fact basically impossible because the prior knowledge is simply not established. (Until they aren't. When does that change? First slowly, then all at once.)

As a biomedical engineer, i'm on board with hardware and software improvements: it's kind of my job. The trick is knowing what you're doing, what you aren't, and what you can expect, and to balance confidence in the value of what you _can_ tightly constrain and design versus humility in accepting that the natural world simply doesn't care.

I think what bugs me the most is the common assumption that we just need better modeling tools to make a lot of the messy lab work go away. Everyone who raises this idea seems to think it's original and revolutionary and a no-brainer, but there's at least 40 years of bitter experience in biotech and pharma development saying otherwise. Even genuinely impressive feats like AlphaFold get inflated wildly in importance (and used to retroactively bash experimentalists for all the time they wasted by not listening to software people). Actually speeding up the entire process of biomedical research requires improvements across the board in many different fields, and it's not something that is magically going to be solved by computer science wizardry alone.

I primarily do modeling, and I am first in line to tell anyone who is willing to listen that the very first thing you should do to learn to be a good modeler is spend a year at the bench first in the problem area you want to model.

Wet lab is so informationally dense in terms of personal knowledge that it's not even funny.