But the stuff we do day-to-day (excepting graphics) is appallingly inefficient. And most of the look-at-this stuff (speech recognition, and you mention VR and eye-tracking) is not actually used for anything day-to-day; we don't routinely talk to our computers ("Hello Computer" https://youtu.be/v9kTVZiJ3Uc?t=10 ).
That's perhaps because doing so in an office or cafe environment isn't great.
Kids are using text-to-speech more while doing their homework. Word has been beefing up its transcription feature for a while now[1].
I got to see this happen during lockdown, where teachers I know recommended transcription for younger kids who don't have touch typing skills and were still expected to turn in work on a computer. Talking to your computer is much more natural while doing your homework in your room.
[1] https://www.theverge.com/2020/8/25/21400623/microsoft-transc...
You've giving the gubment too much credit. It's easy to throw up your hands and say "aww geez I guess everything is backdoored anyway, why bother?" -- this is exactly what they want. The truth is a lot more complex, and as a lot of leaks have revealed, their capabilities are far from the supernatural omnipotence you seem to be implying.
But ultimately this would all fail. Aside from the network effects - nobody's going to use it - the fast processors and tons of memory are necessary. We just don't think about it unless we're running out of it.
- high resolution photo and video - photo editing programs like GIMP and Darktable can spend some time processing photos; these days even with complex effects you usually never experience lag more than a few hundred ms on many megapixel photos, because our hardware is fast. Same for video, the memory and storage space and bandwith is a hard requirement, and going back to 360p is not really acceptable.
- high resolution monitors - no point in having great looking 4k 10 bit color video without a 4k monitor, and now you're stuck having to push 20gbps through your displayport cable on a 500 MHz processor. And text also looks much better at 4k. Also you could say this is a bit unnecessary but compositing window managers are pretty great and I would say a core requirement of modern GUIs, and having eye candy like wobbly windows, window shadows (those are actually very helpful, try turning them off), etc. is expensive. My Dell XPS laptop from 2017 couldn't handle wobbly windows without visible stutter at 4k60; my desktop with a $400 GPU from 2019 can keep up at 4k@144hz.
- new technologies - VR/AR, fast voice recognition and neural networks - this is all cutting edge stuff but the use cases are obvious and they have started to be applied more commonly. Also the failures like eye tracking, Kinect - they may have failed commercially but they were good ideas and a valid use of fast computers. Also online meetings with many participants, each with their own video streams that require decompression.
- obviously, video games - not that you can't have fun games with shitty graphics, but good graphics are nice.
I thought of a few more but can't remember them at the moment. Oh, also, Bluetooth audio - 192kbps audio, but you need to decode it since it takes compression to get it to that level, and then you have the additional overhead of sending it over a digital protocol instead of just having your ADC do the work. You would need expansion cards or a dedicated audio core in your CPU to accomplish this if your computer wasn't fast.
And of course, science needs big data and big hardware to process all of that data.
If you're only working with text then slow computers are fine, but as soon as media gets involved they are not practical.