Losing the root for the tree
49adamShimi
20Adam Zerner
41Dindane
3WilliamKiely
1Ryan Kidd
1bfinn
30Measure
2A1987dM
10Ruby
5Gordon Seidoh Worley
4Screwtape
4EniScien
4Michael Soareverix
3Harish Venugopal
3ATheCoder
3tula
2Roman Leventov
2romeostevensit
3Adam Zerner
2DirectedEvolution
2Adam Zerner
1newlin83
1dkdekes
1rowzzy
1bfinn
7Adam Zerner
2TLK
1WilliamKiely
1WilliamKiely
1Jarred Filmer
2Adam Zerner
New Comment
This post is amazing. Not just good, but amazing. You manage to pack exactly the lesson I needed to hear with just the right amount of memes and cheekiness to also be entertaining.
I would genuinely not be surprised if the frame in this post (and the variations I'm already adding to it) proved one of the key causal factors in me being far more productive and optimizing as an alignment researcher.
One suggestion: let's call these trees treeory of change, because that's what they are. ;)
Thanks. Really.
To make the underlying math more explicit (if still handwavy), I see the thickness as the derivative of the parent with respect to the child; this is why we can multiply them together along a path (the chain rule). This perspective helps us see a few important things:
- The thicknesses change over time and are based on marginal importance rather than absolute importance. Sometimes they change due to random external factors, but often due to your own actions - if Danslist now has a truly excellent network thanks to your efforts, improving it further may not be the most important thing anymore, even if having it is still the most important thing to the company's success.
- The thickness has units of something like [effect]/[work], where the unit of work is something like a person-hour. This means the thicknesses are based not just on importance, but on tractability; if transitioning to MySQL suddenly got 1000 times easier (resp. harder), the corresponding line is now 1000 times thicker (resp. thinner). In this example, even 1000 times thicker may not be enough to make the corresponding leaf relevant, but the general idea is important.
- It's not really a tree. Mental health is important for happiness directly. But it's also important via many subgoals, since poor mental health can make working on other things more difficult. You have to sum all paths from the root to some node when considering its importance - a lot of the time, the tree is a good approximation (the gym is a separate realm from the office), but there are some things that are so important to everything that they demand this amendment.
I think this understanding of line thickness maps onto taut/slack constraints from linear optimization (also discussed by John Wentworth here).
The thickness has units of something like [effect]/[work]
I.e. presumably benefit/cost (work being a cost, whether financial or not), = the benefit-cost ratio (BCR) used in cost-benefit analysis in economics.
Curated. I think the core idea of this post is straightforward yet crucial and powerful. I appreciate the clear explanation and examples. Feels like a more approachable version of a post I wrote a few years ago: Plans are Recursive & Why This is Important.
I like this article for having a amusing stories told with cute little diagrams that manage to explain a specific mental technique. At the end of reading it, I sat down, drew some trees, nodded, and felt like I'd learned a new tool. It's not a tool I use explicitly very often, but I use it a little, using it more wouldn't hurt, and if it happens to be a tool you'd use a lot (maybe because it covers a gap or mistake you make more) then this article is a really good explanation of how to use it and why.
It's interesting to compare this to Goal Factoring. They aren't the same, but they rhyme more than a little to me, both breaking apart what you actually want and how you might get what you want. Between the two, Goal Factoring appeals more to the creative munchkin in me, the kind that giggles at figuring out how to hack the process and get all of advantages of say, grading classes, without paying any of the drawbacks. Finding the root for the tree doesn't come with that feeling of being unorthodox and clever; every time I've used it I come away with what feels like common sense.
"A tool where ever time I use it I come away with what feels like common sense" is, when you put it in those words, actually a stronger recommendation than I originally intended to give but it's true. I wouldn't be surprised if finding the root for the tree doesn't create any noticeable improvement for half the people who use it. It's not revolutionary or groundbreaking. Nevertheless, this is the kind of article I want more of on LessWrong; a technique for making better decisions, explained clearly and straightforwardly, packaged in a neat mostly self-contained package.
I think it could have been written better, I found it a little stretched, especially in the beginning and middle (the ending looks very powerful), it could also be better with more references to already known concepts like "lost goals". But at the same time, it looks like a very important post for instrumental rationality, epistemological rationality is well solved by sequences, but instrumental seems to be what most people lack for a good achievement of goals, this is a more significant node in the tree (at least considering how bad everything is with its implementation). And this, among other things, looks like one of the significant nodes for my instrumental rationality, that, the understanding of which I lacked, both to the upper nodes and to the lower ones. Although I suspect that there are still not enough other nodes of the same importance. (I'm confused no one has written a comment like mine and that there are so few comments at all)
This post is identical to how I started thinking about life a few years ago. Every goal can be broken into subgoals.
I actually made a very simple web app a few years ago to do this: https://dynamic-goal-tree-soareverix--soareverix.repl.co/
It's not super aesthetic, but it has the same concept of infinitely expanding goals.
Amazing post, by the way. The end gave me chills and really puts it all into perspective.
Very well written article , almost reads like a short story despite the weight of its contents :)
The extra little existential dread at the end got me. Excellent mood shift to really push “realign/recall goals now” without making any arbitrary time based axioms. Thanks for the article, and also to the team for getting it curated.
The first three parts of this post can be summarized as follows: "Use state-of-the-art systems engineering methodology, even if you are a "product"! Always look at the supra system of "your system" and the function "your system" plays in its supra system."
As a person with severe bipolar disorder, I'd like to add just how little some of us can understand our own motivations. Things that make absolute sense vanish after beginning lithium. I feel #5 was somewhat touched upon when sons of wealthy men would reject wealth, become celibate vegetarians, and just sit there thinking about dying (Stylites, Desert Fathers, Buddhists, Ebionites, etc.).
This is a beautiful framework that seems to have lots of potential to identify the true root of a lot of problems. I’m surprised I’ve never seen anyone use anything like this before to solve problems, either professionally or in personal life. Thanks for this!
This post brings to mind Ludwig Wittgenstein's Tractatus Logico-Philosophicus, which aimed to explore the connection between language and reality, as well as to delineate the boundaries of scientific inquiry.
Is the moral of this really that all decisions should be made so as to maximize the ultimate goal of happiness x longevity (of you or everyone), in utilitarian fashion; whereas maximizing for subgoals is sometimes/often a poor proxy?
Or is it impractical to do utilitarian calculus all the time, but calculations/heuristics with the thin and thick lines can clarify the role of the subgoals so they can be used as adequate proxies?
(It's partly unclear in my head as I didn't grok the exact meaning of the lines & their thicknesses. And it's too late at night for me to think about this!)
Ultimately you want to optimize for the supergoal. For some the supergoal is utilitarian happiness x longevity, but not for others. The post is agnostic on this question.
The best way to optimize for whatever the supergoal is might be to lean towards calculations, or it might be to lean towards heuristics. The post is agnostic on that question as well.
I think the big point in this post is that when you lose sight of what the supergoal actually is, you often fall into some bad failure modes and do a bad job of optimizing for it. (And also that simply being alive is, err, a pretty important thing.)
This summary of your post is exactly how I experienced it. From this reader’s perspective, you accomplished the goal of expressing this.
Also, I appreciate your agnosticism and acknowledgement that others may not have the same super goal. I do know quite a few people who have had or are having your experience.
I wonder if part of your experience is because you have a choice. Some do not, because, in their circumstances, complete attention must be focused on surviving. I wonder, also, if humans are a bit behind in their evolutionary adaptation to having this level of choice.
Your post also makes clear the incredible difficulty people face in AI alignment. It is difficult to align our own selves. We fall back on heuristics to save time and our own mental resources. There are multiple “right” answers. Rewards here have costs there. It’s difficult to assign weights. The weight values seem to fluctuate depending on where we focus our attention. If we spend too many resources trying to pick a direction, the paths meanwhile change and we have to reassess.
And there is the manipulation of incentives, particularly praise. Is the praise worth the cost? Did your start state set you up to put one foot in front of the next in response to praise? Do you always have to do a good job at being a CEO, a husband, a father? Is being in your wife’s company its own reward or are you doing the job of being a husband? Or do you feel both ways in fluctuating degrees? Also, it may be that the goal is not the only thing that directs your behavior. It may be that, sometimes, the push and pull of whatever small, repeated incentives are happening are guiding less planned behaviors. These less planned behaviors over time become your life.
Anyway, I appreciate what you have said and how you have said it.
What does the thickness of the lines represent precisely? Something like "the thicker the line, the more important the thing underneath is to the thing on top"?
The context behind why I'm asking these questions is because I feel like without a clear sense of what the thickness of the lines represents and how to compare two interventions (represented by two different multi-line paths along the tree) using the thickness of the lines to inform the decisions, it's not clear to me how this framework is more helpful than a reminder to think big picture and not get lost in the details. (And I'd like it to be a useful framework for me.)
Curated and popular this week
