AGI will, over the course of the next few years, be put to work at national intelligence agencies and in military devices. Some countries will be more careful than others in ensuring that their AGI is safe and moral, but they all need all the help they can get in getting this right (or as close to that as possible). For that reason, there needs to be a place they can look up to find out what is the current state of the art in machine ethics. Such a place should maintain a league table of proposed solutions listed in order of how well they perform, and any issues with them should be listed under each entry to invite people to try to resolve them, thereby enabling any advances to lead to that proposal moving up the league table, or down if a new fault with it is uncovered.
I have been unable to find a league table if that kind thus far. All I see is a mass of different approaches scattered about with no clear answers. I set out my own proposed solution in the opening part of this series ( https://www.lesswrong.com/posts/Lug4n6RyG7nJSH2k9/computational-morality ), and I want to see it placed into a proper league table with all the other proposed solutions. People would then be invited to provide objections to show where they think it fails, and these would be listed in order with the most serious objections at the top. People would also be free to try to show that those objections aren't valid (so any that are shown to be wrong would be moved to a list of failed objections), and they'd also be free to add new objections. The same would happen with all the other proposals, and over the course of time we would see them all migrate slowly to their proper places.
The biggest problem with such a league table might be the issue of who gets to decide when a claimed fault has been invalidated (or whether it applies in the first place), so it would all depend on fair judging, but any kind of bias that's introduced would soon lead to a rival site offering a better alternative table which would gain greater credibility and take over as the go-to place for everyone involved in this business, so it would be in the interests of the people in charge to make sure that they are being fully fair.
But would it really be worth the bother?
Well, judging from my experience as an AGI system builder, yes: I'm trying to find out what the latest thinking is on machine ethics, and what happens is that people just point to a large number of old articles which don't shed much light on the issue and say "go and read those". You can be sure that other builders of AGI are in the same boat, and most of them won't take as much time as I have to read through large amounts of what is for the most part irrelevant material. It needs to be made much easier. So, what I'm going to do for now is extend this series of posts to set out my search for answers in the hope that it will help others to find their way into the subject without costing them so much time. I also think my proposed solution works, so I'm going to continue to use that as a way to coax information out of anyone who is able to help, and I will compare it with other proposals in future parts of this series of posts (which you can find by clicking on my name at the top, and don't be put off by the negative score attached to my name by people who haven't yet seen the light - their field needs to be tidied up, and I'm going to do it for them).
AGI will, over the course of the next few years, be put to work at national intelligence agencies and in military devices. Some countries will be more careful than others in ensuring that their AGI is safe and moral, but they all need all the help they can get in getting this right (or as close to that as possible). For that reason, there needs to be a place they can look up to find out what is the current state of the art in machine ethics. Such a place should maintain a league table of proposed solutions listed in order of how well they perform, and any issues with them should be listed under each entry to invite people to try to resolve them, thereby enabling any advances to lead to that proposal moving up the league table, or down if a new fault with it is uncovered.
I have been unable to find a league table if that kind thus far. All I see is a mass of different approaches scattered about with no clear answers. I set out my own proposed solution in the opening part of this series ( https://www.lesswrong.com/posts/Lug4n6RyG7nJSH2k9/computational-morality ), and I want to see it placed into a proper league table with all the other proposed solutions. People would then be invited to provide objections to show where they think it fails, and these would be listed in order with the most serious objections at the top. People would also be free to try to show that those objections aren't valid (so any that are shown to be wrong would be moved to a list of failed objections), and they'd also be free to add new objections. The same would happen with all the other proposals, and over the course of time we would see them all migrate slowly to their proper places.
The biggest problem with such a league table might be the issue of who gets to decide when a claimed fault has been invalidated (or whether it applies in the first place), so it would all depend on fair judging, but any kind of bias that's introduced would soon lead to a rival site offering a better alternative table which would gain greater credibility and take over as the go-to place for everyone involved in this business, so it would be in the interests of the people in charge to make sure that they are being fully fair.
But would it really be worth the bother?
Well, judging from my experience as an AGI system builder, yes: I'm trying to find out what the latest thinking is on machine ethics, and what happens is that people just point to a large number of old articles which don't shed much light on the issue and say "go and read those". You can be sure that other builders of AGI are in the same boat, and most of them won't take as much time as I have to read through large amounts of what is for the most part irrelevant material. It needs to be made much easier. So, what I'm going to do for now is extend this series of posts to set out my search for answers in the hope that it will help others to find their way into the subject without costing them so much time. I also think my proposed solution works, so I'm going to continue to use that as a way to coax information out of anyone who is able to help, and I will compare it with other proposals in future parts of this series of posts (which you can find by clicking on my name at the top, and don't be put off by the negative score attached to my name by people who haven't yet seen the light - their field needs to be tidied up, and I'm going to do it for them).