It’s difficult for me to say because this sort of introspection is difficult, but I believe that I generally reject values when I find that they don’t promote happiness.
You can probably think of a happiness-based justification for any value someone throws at you.
But some justifications are legitimate and some are rationalizations. With the examples of discovery and creativity, I think it’s obvious that they increase happiness by a lot. It’s not like I came up with some ad hoc justification for why they maybe provide a little bit of happiness. It’s like discovery is responsible for almost all of the increases in quality of life that have taken place over the past several thousand years.
I suspect an AI designed only to maximise happiness would probably find a different way that would produce more happiness while disregarding almost all values we think we have.
I think a lot of our values do a very good job of increasing happiness, and I welcome an AI that can point out which values don’t.
With the examples of discovery and creativity, I think it’s obvious that they increase happiness by a lot.
The point is that’s not sufficient. Like saying “all good is complexity, because for example a mother’s love for her child is really complex”. Yes, it’s complex compared to some boring things like carving identical chair legs out of wood over and over for eternity, but compared to, say, tiling the universe with the digits of chaitin’s omega or something, it’s nothing. And tiling the universe with chaitin’s omega would be a very boring and stupid thing to do.
You need to show that the value in question is the best way of generating happiness. Not just that it results in more than the status quo. It has to generate more happiness, than, say, putting everyone on heroine forever. Because otherwise someone who really cared about happiness would just do that.
I think a lot of our values do a very good job of increasing happiness, and I welcome an AI that can point out which values don’t.
And they other point is that values aren’t supposed to do a job. They’re meant to describe what job you would like done! If you care about something that doesn’t increase happiness, then self-modifying to lose that so as to make more happiness would be a mistake.
You need to show that the value in question is the best way of generating happiness.
You’re absolutely correct. Discovery may not always be the best way of generating happiness; and if it’s not, you should do something else.
And the other point is that values aren’t supposed to do a job.
Not all values are terminal values. Some people value coffee because it wakes them up; they don’t value coffee in itself. If they discover that coffee in fact doesn’t wake them up, they should stop valuing coffee.
With the examples of discovery and creativity, I think it’s obvious that they increase happiness by a lot.
The point is that’s not sufficient.
What is sufficient is demonstrating that if discovery does not promote happiness then it is not valuable. As I explained in my sorting sand example, discovery that does not in any way promote happiness is not worthwhile.
It’s difficult for me to say because this sort of introspection is difficult, but I believe that I generally reject values when I find that they don’t promote happiness.
But some justifications are legitimate and some are rationalizations. With the examples of discovery and creativity, I think it’s obvious that they increase happiness by a lot. It’s not like I came up with some ad hoc justification for why they maybe provide a little bit of happiness. It’s like discovery is responsible for almost all of the increases in quality of life that have taken place over the past several thousand years.
I think a lot of our values do a very good job of increasing happiness, and I welcome an AI that can point out which values don’t.
The point is that’s not sufficient. Like saying “all good is complexity, because for example a mother’s love for her child is really complex”. Yes, it’s complex compared to some boring things like carving identical chair legs out of wood over and over for eternity, but compared to, say, tiling the universe with the digits of chaitin’s omega or something, it’s nothing. And tiling the universe with chaitin’s omega would be a very boring and stupid thing to do.
You need to show that the value in question is the best way of generating happiness. Not just that it results in more than the status quo. It has to generate more happiness, than, say, putting everyone on heroine forever. Because otherwise someone who really cared about happiness would just do that.
And they other point is that values aren’t supposed to do a job. They’re meant to describe what job you would like done! If you care about something that doesn’t increase happiness, then self-modifying to lose that so as to make more happiness would be a mistake.
You’re absolutely correct. Discovery may not always be the best way of generating happiness; and if it’s not, you should do something else.
Not all values are terminal values. Some people value coffee because it wakes them up; they don’t value coffee in itself. If they discover that coffee in fact doesn’t wake them up, they should stop valuing coffee.
What is sufficient is demonstrating that if discovery does not promote happiness then it is not valuable. As I explained in my sorting sand example, discovery that does not in any way promote happiness is not worthwhile.