The easiest and safest thing is usually still to let people control more of what happens because there's a lot of edge cases that might add up, or have really bad consequences for when they rarely occur.
The easiest and safest thing is usually still to let people control more of what happens because there's a lot of edge cases that might add up, or have really bad consequences for when they rarely occur.
Even though it often feels like you do, humans generally don't know their own intents. It's not that high of a bar for a predictive system to clear.
Unless you're going to be repeatedly walking too and from your car all day, starting a computer needlessly isn't going to be the end of the world.
Yes, it may mean doing more work but that's the trade off here for "instant on". It's either always on, you wait, or it guesses when you're going to use it and you accept there are a few false positives.
Or engineers reduce startup time to a point where it feels instant.
Compare e.g. startup times of DSLRs and point-n-shoot cameras. With a DSLR you switch it on and it is ready to shoot by the time your finger gets to shutter button.
It's also very hard to engineer a boot time down. TFA is a great example of that. Hard means expertise and time and (often) more expensive hardware. So to boil it right down with my car computer example, it's far easier and cheaper to detect the key externally and boot earlier than it is to strip a 10 second boot time down to 0.1 second.
If you can also use that getting-in-the-car time to do other stuff (lookup up recent searches or upcoming appointments and suggesting navigation, setting fuel waypoints if low, getting traffic advice, etc), that's all the better.