From [[Human Compatible, Stuart Russell]] > ...we are typically embedded in a deeply nested hierarchy of “subroutines.” [...] Right now, for example, I’m typing this sentence: I can choose how to continue after the colon, but it never occurs to me to wonder if I should stop writing the sentence and take an online rap course or burn down the house and claim the insurance or any other of a gazillion things I could do next. Many of these other things might actually be better than what I’m doing, but [...] it’s as if those other things didn’t exist. > [...] > There's a famous New Yorker cover from the 1970 which shows how a New Yorker from the 9th avenue might view the world. ![[View from ninth.png]] Russell analogizes this to how we think about future and our current actions. > The very immediate future is extraordinarily detailed—in fact, my brain has already loaded up the specific motor control sequences for typing the next few words. Looking a bit further ahead, there is less detail—my plan is to finish this section, have lunch, write some more, and watch France play Croatia in the final of the World Cup. Still further ahead, my plans are larger but vaguer: move back from Paris to Berkeley in early August, teach a graduate course, and finish this book. As one moves through time, the future moves closer to the present and the plans for it become more detailed, while new, vague plans may be added to the distant future. Plans for the immediate future become so detailed that they are executable directly by the motor control system. ### Implications on Personal Productivity So like a computer program keeps both the next set of instructions to process and the [call stack](https://en.wikipedia.org/wiki/Call_stack) to get back into, you keep in your head what are you doing right now, and what's the broader context(s). Unlike a computer program, the decision of when to "pop the stack" (stop the current endeavor and move to another one) is not deterministic. This can go awry in two ways: * Over-focus can cause "blinders on" mode, where you get stuck in the subroutines. This is [[Yak shaving]]. * Under-focus can present with [continuous partial attention](https://en.wikipedia.org/wiki/Continuous_partial_attention), and the inability to stick through the subroutine. Some remedies here are: * Periodical checks that the destination is still correct, that we're [[In the right forest]]. * A [[GTD]] type of system so that you're not worried about [[Task Leak]], and [[Reminder Cron]] can be turned off. * Mindfulness meditation? ### Implications on Philosophy & AI I've always wondered about this: sometimes you finish a subroutine, and you pop a level in the stack. But sometimes, you don't and there's a part of you that says "OK, enough of this". It can happen in action, or it can happen in a thought! In fact, with mindfulness, you learn to do this more quickly, noticing the thought itself rather than following it. Exactly what *is* the part of the mind that notices the current narrative and is able to challenge it? How does metacognition work? Similarly, in the novel reasoning systems like DeepSeek and o3, the most fascinating bits to me are when the models think to themselves "wait wait wait", and do things that start to look like metacognition. And the fact that it's completely emergent! #published 2025-02-16