If you don't like videos, you can also look up his papers from STEPS circa 2010.
I think the 3 body problem is indicative of what Kay says. One simple rule produces very rich behavior. Why not have something like that for your network stack?
> Why not have something like that for your network stack?
In a way, it is like that already. After all, all computation can be described as a repeated application of, say, the very simple reduction rules of the lambda calculus. How does that help? Does the fact we know what the gravity equation help us predict the system's behavior (e.g. to ensure that some bad state is never reached)?
> Computers have been invented to surprise us
> If we knew what computers do, we would not use them, and we would not have built any.
This means that computers are most useful precisely where the complexity of the problem is beyond our efficient reasoning powers. Like Brooks, I think that our accidental complexity is already quite low (no more than 50%), and while it's always good to decrease it, what we're left with will still be very complex -- that's why it's useful.
[1]: https://shemesh.larc.nasa.gov/NFM2018/talks/Dowek.pdf