Must intelligence be single-threaded?

How would multitasking on a conscious level really look like?

Imagine for a second that LLMs actually will lead to AGI (instead of being just an input layer to it). Imagine that this is done by something like a continually updated and executing prompt that integrates the sensorium and adds a dash of “true will” or at least the sense of purpose. This kind of process sounds awfully like what we are experiencing in our daily existence.

But this process is limited in at least one way - it’s linear and single-threaded. AIs of science fiction tend to get a large boost of intellectual power by being able to somehow monitor or interact with many entities at the same time. Iain Banks’ Culture “minds” are known to be able to hold simultaneous meaningful conversations with billions of people - literally with everyone on the planet - while at the same time doing other useful and complicated tasks.

Bot how would that work? How would it feel?

Is it just doing the prompt-context thing really, really fast so it looks like multitasking? Or is there really a way to make a “multithreaded” intelligence?

In computer science, we have developed multiple ways of making sure multithreaded code running on different CPU cores, CPUs, or even completely unrelated systems doesn’t step on each-others toes (memory). Lockless algorithms are just about the pinnacle of that, especially with memory access paradigms like the one introduced by Rust, and this kind of hints that it could, one day, be a truly multitasking intelligence. Other paradigms like message passing with “channels” could be an alternative.

Oracle from The Matrix said “We can never see past the choices we don't understand” and it just might be that we can’t really envision a mind whose workings are foreign to our own.