Research is slow, and most of the slowness isn't the thinking.
It's the work around it, like finding the papers, setting up the experiment, cleaning the data, and then checking whether the result still holds when you change one thing. A good researcher can spend weeks on that before they learn anything new, and most of those weeks go on steps they've done many times before.
We're building systems that take that work off people's hands. Give one a question and it reads what's already known, plans the experiments, runs them and comes back with what it found and how sure it is. People still choose the questions and decide what the answers mean, while the system does the legwork, day and night, without getting bored on the fortieth run.
Automating research only works if the method can be written down, and most method lives in people's heads. A senior chemist knows which results to distrust and when a reading looks off, and a good analyst knows which shortcut is safe to take, but none of that is written anywhere a machine could follow it.
So a lot of our work is formalizing method. We take the steps an expert follows, along with the reasons behind each one, and turn them into rules and checks a system can run. This is where philosophy earns its place. Philosophers have spent centuries asking what counts as evidence and what makes an argument sound, and their answers turn out to be very practical once you have to write them as code.
A written method can also be checked, which is something an expert's gut feeling never allows. When the system reaches a conclusion, you can see every step it took and the rule that allowed it. If a step was wrong, you fix the rule once and every later run gets the fix.
Even so, a method isn't finished until it has been deployed and has held up outside a demo. The same system can run an automatic factory, working out what to build next and keeping the lines fed, or a research lab, planning the next round of experiments while the last one is still running. We put it to work on the problems people actually have and watch where it breaks, and every failure goes back into the method.
Over time that becomes a loop. The system runs the research, the method gets sharper, and each deployment starts from a better place than the last, until more and more of the work runs itself, the way an AE2 base does once everything is wired into the network.