If you want to learn how computers work now is easy. Learn C and a bit of assembly, read about hardware and try to do low level stuff. Play with memory, play with system calls. Learn a bit about GPUs, some OpenGL, DirectX, shading language or Vulkan and tinker with the GPU.
I mean, all of that is grand but you remain for, for from the metal in all these instances and even in assembly you cannot go under several layers of abstraction with modern architectures
in the days of 6502 and Z80 compatible microprocessors, you could go down to the NAND gate level, to the transistor level - it was doable, not that it made much sense to go there for most people, but it was nearly a requirement to know how many cycles did each instruction take and lock conditions like bus access when dealing with different parts of the memory, etc
there were essentially no hard black boxes then, and now there are many at the hardware level and even the OS level is too impenetrable; not even the CTOs of major OS companies and initiatives have a complete understanding of the OS they produce, perhaps a few select ones have a functional understanding with only 2 or 3 black boxes involved (specific driver magic, hardware detail that goes under C and assembly optimisation etc - which typically outsourced specialists deal with)
GPUs alone these days are more complex than the entirety of the systems back then, and the implications of this are two-fold:
- the culture on the producer end is that software, machines and accessories are sold as black boxes, both by expectations and by enforcement on the manufacturer's end that gets no incentive whatsoever to expose internals of their product, and many reasons to the contrary as the competition might use it; these dynamics are extending to repair-ability as well
- the culture on the consumer end is that internals are not possible to deal with, are best taken for granted and the curiosity is just not there anymore; also if things break you just get new things, which disincentives also producing anything meant to last, as the assumption is that it will be obsolete soon anyway, or the system will just stop working, reinforcing that dynamic and cheapening the value of work