There is basically one reason to build a tower, and that’s access: There is something that is available higher up that is not available on the ground. Most commonly, that thing is either “empty space” or “information transmission”. A skyscraper can access more space than a shorter building, and so it fits more real estate in a smaller footprint. A watchtower can access strategic information from the land below; a church steeple is tall enough for the sound of the bells to travel far through the air. Many of the tallest structures in the world maximize line-of-sight for radio transmissions.
Imagine a stone or brick tower of your favorite type - one whose purpose is to access something high up. Or, if you prefer a specific suggestion, imagine the brick steeple of Old North Church in Boston, which held the signaling lanterns for Paul Revere.
Consider some unremarkable block in the tower: On either side of this block are other blocks. Sitting on top of this block are more blocks. Below, holding our block’s weight: more blocks.
One can ask, about the block we’ve chosen, the question “how?”: How is it that this block is at this spot? And the answer to that question is, locally, that it is held up by other blocks immediately below. Those blocks depend on others below them, and so indirectly our block is here by the grace of the whole set of blocks transitively beneath it.
One can ask, about the block we’ve chosen, the question “why?”: For what purpose is this block at this spot? And the answer to that question is, locally, that it is holding up other blocks immediately above. Those blocks are holding up others above them, and so indirectly our block is here in order to help hold up the whole set of blocks transitively above it.
“How are you going to put a stone there in that empty space?” asks Alice. “By first putting stones in the empty space beneath it.” replies Bob. “Why are you laying that brick?” asks Alice. “I want to rest this other brick on top of it.” replies Bob. For Bob, the bricks lower down in the tower are more “instrumental”; the bricks higher in the tower are more “terminal”.
This is sort of what plans are like - or at least, some “well-behaved” parts are structured like this. “How are you going to get a job as an engineer?” asks Alice. “I’m going to apply with an impressive resume and all the required certifications.” replies Bob. “Why are you going to get a job as an engineer?” asks Alice. “I’ll have a good steady income, and respectability in the community.” replies Bob. Bob sounds like he has a plan. Plans are kind of tower-y, and so are strategies and even agents themselves.
But as you follow the trail of blocks down to the bottom of a tower, you eventually meet the ground upon which the tower rests. And as you follow the trail of blocks up to the top of a tower, you eventually meet empty space above.
Plans are stranger. As you follow the trail of goals “down” a plan toward the more-instrumental, it’s not clear where the “ground” is, or if there even is a ground. “How are you going to actuate your finger to apply the force to press the key to write the sixtieth character of this essay for your engineering class for your engineering degree for your engineering job?”
And as you follow the goals “up” a plan toward the more-terminal, it’s not clear where the “sky” is, or if there even is a sky. “Why do you like to eat ice cream? Why do you want to start a family? Why do you want good things to happen to you?”
We see, and embody, lots of examples of “locally well-behaved” agents. Sometimes we literally place blocks one on top of the other to build towers. Some of us have written software, typically by assembling a hierarchy of procedures that the computer executes to accomplish a task. We execute simple or complex strategies to win at games and do our jobs.
But as we ask “why?”, moving upward from the well-behaved regions of human behavior - the tower wall segments we can see through the clouds - the answers stop sounding like “because I want X”. Instead we have to start giving answers that depart from the frame of seeking more-terminal values: “because that’s the kind of person I am”, or “because my neurons fire in such-and-such a pattern”. Answers to “why” questions that mysteriously sound more like answers to “how” questions.
In principle a person could be a single coherent tower with a single apex, the tip of a spire that’s the fruit of all the stonework below. Many explicit models of agency, like those used by economists and computer scientists, fit well into a tower structure. The top of the steeple becomes a utility function, or a reward function, or a goal. Lower blocks might be preferences, or subgoals, or resources, or actions.
But the top of a real human tower appears to dissolve into something not-so-hierarchical and not-so-neat. Somehow my body is inhabited at times by a relatively coherent writer of software, and at other times by a relatively coherent writer of English sentences, but whatever is selecting between the software-writer and sentence-writer modes does not look like a simple and coherent mode-chooser. My pizza-seeking mode does not seem to neatly descend from a food-optimization goal. When I was a child and my mother would call me for dinner while I was in the middle of a book, I’d sometimes be quite frustrated at the interruption, even if I was very hungry. From within a flow state, even a life-saving interruption could carry a twinge of that pain of interruption.
These tower-like hierarchical models of agency do not seem to explain human behavior very well, or at least when they describe behavior well they don’t do so very simply. Whatever we humans are, we seem to “grow” or “host” pockets of tower-like agency, rather than fully embodying it.
There is basically one reason to build a tower, and that’s access: There is something that is available higher up that is not available on the ground. Most commonly, that thing is either “empty space” or “information transmission”. A skyscraper can access more space than a shorter building, and so it fits more real estate in a smaller footprint. A watchtower can access strategic information from the land below; a church steeple is tall enough for the sound of the bells to travel far through the air. Many of the tallest structures in the world maximize line-of-sight for radio transmissions.
Imagine a stone or brick tower of your favorite type - one whose purpose is to access something high up. Or, if you prefer a specific suggestion, imagine the brick steeple of Old North Church in Boston, which held the signaling lanterns for Paul Revere.
Consider some unremarkable block in the tower: On either side of this block are other blocks. Sitting on top of this block are more blocks. Below, holding our block’s weight: more blocks.
One can ask, about the block we’ve chosen, the question “how?”: How is it that this block is at this spot? And the answer to that question is, locally, that it is held up by other blocks immediately below. Those blocks depend on others below them, and so indirectly our block is here by the grace of the whole set of blocks transitively beneath it.
One can ask, about the block we’ve chosen, the question “why?”: For what purpose is this block at this spot? And the answer to that question is, locally, that it is holding up other blocks immediately above. Those blocks are holding up others above them, and so indirectly our block is here in order to help hold up the whole set of blocks transitively above it.
“How are you going to put a stone there in that empty space?” asks Alice. “By first putting stones in the empty space beneath it.” replies Bob. “Why are you laying that brick?” asks Alice. “I want to rest this other brick on top of it.” replies Bob. For Bob, the bricks lower down in the tower are more “instrumental”; the bricks higher in the tower are more “terminal”.
This is sort of what plans are like - or at least, some “well-behaved” parts are structured like this. “How are you going to get a job as an engineer?” asks Alice. “I’m going to apply with an impressive resume and all the required certifications.” replies Bob. “Why are you going to get a job as an engineer?” asks Alice. “I’ll have a good steady income, and respectability in the community.” replies Bob. Bob sounds like he has a plan. Plans are kind of tower-y, and so are strategies and even agents themselves.
But as you follow the trail of blocks down to the bottom of a tower, you eventually meet the ground upon which the tower rests. And as you follow the trail of blocks up to the top of a tower, you eventually meet empty space above.
Plans are stranger. As you follow the trail of goals “down” a plan toward the more-instrumental, it’s not clear where the “ground” is, or if there even is a ground. “How are you going to actuate your finger to apply the force to press the key to write the sixtieth character of this essay for your engineering class for your engineering degree for your engineering job?”
And as you follow the goals “up” a plan toward the more-terminal, it’s not clear where the “sky” is, or if there even is a sky. “Why do you like to eat ice cream? Why do you want to start a family? Why do you want good things to happen to you?”
We see, and embody, lots of examples of “locally well-behaved” agents. Sometimes we literally place blocks one on top of the other to build towers. Some of us have written software, typically by assembling a hierarchy of procedures that the computer executes to accomplish a task. We execute simple or complex strategies to win at games and do our jobs.
But as we ask “why?”, moving upward from the well-behaved regions of human behavior - the tower wall segments we can see through the clouds - the answers stop sounding like “because I want X”. Instead we have to start giving answers that depart from the frame of seeking more-terminal values: “because that’s the kind of person I am”, or “because my neurons fire in such-and-such a pattern”. Answers to “why” questions that mysteriously sound more like answers to “how” questions.
In principle a person could be a single coherent tower with a single apex, the tip of a spire that’s the fruit of all the stonework below. Many explicit models of agency, like those used by economists and computer scientists, fit well into a tower structure. The top of the steeple becomes a utility function, or a reward function, or a goal. Lower blocks might be preferences, or subgoals, or resources, or actions.
But the top of a real human tower appears to dissolve into something not-so-hierarchical and not-so-neat. Somehow my body is inhabited at times by a relatively coherent writer of software, and at other times by a relatively coherent writer of English sentences, but whatever is selecting between the software-writer and sentence-writer modes does not look like a simple and coherent mode-chooser. My pizza-seeking mode does not seem to neatly descend from a food-optimization goal. When I was a child and my mother would call me for dinner while I was in the middle of a book, I’d sometimes be quite frustrated at the interruption, even if I was very hungry. From within a flow state, even a life-saving interruption could carry a twinge of that pain of interruption.
These tower-like hierarchical models of agency do not seem to explain human behavior very well, or at least when they describe behavior well they don’t do so very simply. Whatever we humans are, we seem to “grow” or “host” pockets of tower-like agency, rather than fully embodying it.