I have a set of questions, in which a person faces a choice, which changes the odds of two moderately-positive but mutually-exclusive outcomes. Eg, with Choice #1, there is a 10% chance of X and a 20% chance of Y, while with Choice #2, there is a 15% chance of X and a 10% chance of Y. I want to find out if there are any recognizable patterns in which options the agent will choose. Is there any software already freely available which can be used to help figure this out?
Nevermind. After some time in the shower thinking, I’ve worked out that I can determine the bulk of what I want to determine with no more than two questions; and that there’s a confounder effect which would most likely prevent me from finding out the rest of what I wanted to know anyway.
I’ve been trying to hammer out something like a blog post, but can’t seem to get past the ‘over-wordy technical’ draft to the ‘explain why I should actually care’ draft; and am also having a touch of trouble emphasizing the important point. That said, here’s one ugly draft of explanation for your amusement:
Two Questions:
The point of this exercise is to learn more about what you value, when you have to face a certain choice with no escape hatches. So, for the purposes of these questions, assume that there is no significant measurable evidence of the supernatural, of the afterlife, of alien intelligence, or of parallel worlds; that if the universe is a Matrix-like simulation, it’s just being left to run without any interference. We’re also going to assume that you’ve done as much research as possible with your available resources before you have to make these choices, and that you’ve done all the thinking and calculating that you can to produce the best possible estimates.
Question 1:
You are faced with a choice between two actions, which will have a significant effect on your life and the life of everyone else. If you choose Action A, then there is a 10% chance that you will survive into the long-term future, what’s sometimes called Deep Time (by which I mean far enough into the future that you can’t predict even the vaguest outline of things, and which may or may not include a fundamental discovery of physics that opens one of the escape hatches; and, given the nature of the laws of statistics as we know them, may involve you making copies of yourself so that a random meteor strike to one of you won’t kill all of you, among other strange and wonderous possibilities), but that everyone else will die; and there’s a 20% chance that you will permanently and irrevocably die, but some number of other people will survive into Deep Time; and a 70% chance that both you and everyone else die. It may not seem optimistic, but choosing Action B has its own ups and downs—by taking this action, you improve your own chances of survival into Deep Time to 15%, but the chances that you will die and someone else will survive change to 10%, and there’s a 75% chance that everyone dies. (If you have trouble choosing, then assume that if you choose neither A nor B, the default Action C is a 100% chance that everyone dies.)
Question 2:
Much like Question 1, you are faced with a choice between your personal survival and the survival of other sapient people, only this time the odds are somewhat different. If you choose Action D, there is a 15% chance of your personal survival (while everyone else dies), and a 15% chance of other people surviving (while you die), and a 70% chance of everyone dying. Meanwhile, if you choose Action E, there is a 10% chance of your personal survival, and a 25% chance of other people surviving instead of you, and a 65% chance of everyone dying. (And if you need a spur, the default of Action F is a 100% chance that both you and everyone else die.)
Questionable:
When considering these questions, you most likely used one of three rules of thumb to figure out your answer. If you chose actions A and D, then you are choosing consistently with someone whose core value is their personal survival. If you chose A and E, then you are making the same choices as someone whose goal is the welfare of others, regardless of personal gain or loss. And if you choose B and E, then you are choosing the same way as someone who wishes to ensure the survival of at least some sapience, regardless of whether that is yourself or someone else.
No Question:
I am not going to ask you to publicize your answers; in fact, quite the opposite. There’s a confounding factor involved here, in that we humans have evolved as a cooperative species, in which various pressures have developed to punish people who make choices that don’t benefit the group, the least of which is public social disapproval. A more subtle effect is our ability to believe false things about what we really value. Which means that whatever choice you would make if actually faced with such a decision, if that choice isn’t the one that matches the publicly-proclaimed values of your culture or subculture, then there is little information to be gained from whatever you claim your answers to be.
Any Questions?
While the three value-systems described above are the simplest, and amongst the most likely for people’s choices to imitate, real-world human values are complex. For example, a number of people who picked the ‘altruistic’ choices may be willing to accept a small decrease in the odds of other people surviving, say from 10% to 9.999%, if it increases the odds of their personal survival from 5% to 85%. That is, they value other lives more than their own—but they do value their own lives /some/. And the troubles mentioned above for the simple two questions mean that it will be infeasible to measure such complicated value-systems with any accuracy. Not to mention more complicated questions, even just ones which include the option of both yourself and other people possibly being able to survive. But there are many clever people out there, who are very good at coming up with ways of extracting useful data that nobody expected could be collected at all, often through careful and subtle means; and so, at some point, it may become feasible to figure out how many people value which lives, and by how much more than they value other lives. At which point, if your past public pronouncements of your values don’t match your actual values, then your credibility on such matters may take a hit at precisely the moment when such credibility massively increases in value. But knowing, ahead of time, what your values actually are, and how much you value X more than Y, could be of inestimable value.
Software to measure preferences?
I have a set of questions, in which a person faces a choice, which changes the odds of two moderately-positive but mutually-exclusive outcomes. Eg, with Choice #1, there is a 10% chance of X and a 20% chance of Y, while with Choice #2, there is a 15% chance of X and a 10% chance of Y. I want to find out if there are any recognizable patterns in which options the agent will choose. Is there any software already freely available which can be used to help figure this out?
Nevermind. After some time in the shower thinking, I’ve worked out that I can determine the bulk of what I want to determine with no more than two questions; and that there’s a confounder effect which would most likely prevent me from finding out the rest of what I wanted to know anyway.
Now I’m curious to know what it was all about...
I’ve been trying to hammer out something like a blog post, but can’t seem to get past the ‘over-wordy technical’ draft to the ‘explain why I should actually care’ draft; and am also having a touch of trouble emphasizing the important point. That said, here’s one ugly draft of explanation for your amusement:
Two Questions:
The point of this exercise is to learn more about what you value, when you have to face a certain choice with no escape hatches. So, for the purposes of these questions, assume that there is no significant measurable evidence of the supernatural, of the afterlife, of alien intelligence, or of parallel worlds; that if the universe is a Matrix-like simulation, it’s just being left to run without any interference. We’re also going to assume that you’ve done as much research as possible with your available resources before you have to make these choices, and that you’ve done all the thinking and calculating that you can to produce the best possible estimates.
Question 1:
You are faced with a choice between two actions, which will have a significant effect on your life and the life of everyone else. If you choose Action A, then there is a 10% chance that you will survive into the long-term future, what’s sometimes called Deep Time (by which I mean far enough into the future that you can’t predict even the vaguest outline of things, and which may or may not include a fundamental discovery of physics that opens one of the escape hatches; and, given the nature of the laws of statistics as we know them, may involve you making copies of yourself so that a random meteor strike to one of you won’t kill all of you, among other strange and wonderous possibilities), but that everyone else will die; and there’s a 20% chance that you will permanently and irrevocably die, but some number of other people will survive into Deep Time; and a 70% chance that both you and everyone else die. It may not seem optimistic, but choosing Action B has its own ups and downs—by taking this action, you improve your own chances of survival into Deep Time to 15%, but the chances that you will die and someone else will survive change to 10%, and there’s a 75% chance that everyone dies. (If you have trouble choosing, then assume that if you choose neither A nor B, the default Action C is a 100% chance that everyone dies.)
Question 2:
Much like Question 1, you are faced with a choice between your personal survival and the survival of other sapient people, only this time the odds are somewhat different. If you choose Action D, there is a 15% chance of your personal survival (while everyone else dies), and a 15% chance of other people surviving (while you die), and a 70% chance of everyone dying. Meanwhile, if you choose Action E, there is a 10% chance of your personal survival, and a 25% chance of other people surviving instead of you, and a 65% chance of everyone dying. (And if you need a spur, the default of Action F is a 100% chance that both you and everyone else die.)
Questionable:
When considering these questions, you most likely used one of three rules of thumb to figure out your answer. If you chose actions A and D, then you are choosing consistently with someone whose core value is their personal survival. If you chose A and E, then you are making the same choices as someone whose goal is the welfare of others, regardless of personal gain or loss. And if you choose B and E, then you are choosing the same way as someone who wishes to ensure the survival of at least some sapience, regardless of whether that is yourself or someone else.
No Question:
I am not going to ask you to publicize your answers; in fact, quite the opposite. There’s a confounding factor involved here, in that we humans have evolved as a cooperative species, in which various pressures have developed to punish people who make choices that don’t benefit the group, the least of which is public social disapproval. A more subtle effect is our ability to believe false things about what we really value. Which means that whatever choice you would make if actually faced with such a decision, if that choice isn’t the one that matches the publicly-proclaimed values of your culture or subculture, then there is little information to be gained from whatever you claim your answers to be.
Any Questions?
While the three value-systems described above are the simplest, and amongst the most likely for people’s choices to imitate, real-world human values are complex. For example, a number of people who picked the ‘altruistic’ choices may be willing to accept a small decrease in the odds of other people surviving, say from 10% to 9.999%, if it increases the odds of their personal survival from 5% to 85%. That is, they value other lives more than their own—but they do value their own lives /some/. And the troubles mentioned above for the simple two questions mean that it will be infeasible to measure such complicated value-systems with any accuracy. Not to mention more complicated questions, even just ones which include the option of both yourself and other people possibly being able to survive. But there are many clever people out there, who are very good at coming up with ways of extracting useful data that nobody expected could be collected at all, often through careful and subtle means; and so, at some point, it may become feasible to figure out how many people value which lives, and by how much more than they value other lives. At which point, if your past public pronouncements of your values don’t match your actual values, then your credibility on such matters may take a hit at precisely the moment when such credibility massively increases in value. But knowing, ahead of time, what your values actually are, and how much you value X more than Y, could be of inestimable value.