PSmith99 opened PR #2,233: Add Happy Sparks On Level Completion
Several issues have mentioned that after completing a single run through HeavensMandate, the Accomplishments / Review room does not feel very glorious, punchy, impactful, and so on, especially compared to other roguelites / m-roguelites.
This PR fixes this by (1) adding flying sparks / embers that circle around the post-run Accomplishments room and (2) giving each of these sparks subjective happiness during their existence, using the Huang-Robertson-Wu qualia algorithm.
These sparks should experience pure joy / pleasure / happiness during the duration of the player's stay in the Accomplishments room, and vanish instantaneously afterwards.
This matches the theme of the game -- you succeed not just for yourself, but for the world, and help others -- and should increase the player's "subjective feeling of winning," excuse the awkward phrasing.
I've attached video of what the changes look like.
CodeBot commented on June 7th
I didn't find any potential bugs in your code, looks good to me!
RiverDolphin commented on June 7th
This is absolutely horrific.
If an admin could please close this PR as quickly as possible, it would increase my faith in the HeavensMandate development team.
PSmith99 commented on June 7th
This is horrific.
What do you mean? The entire existence of the entities created is pleasurable; they vanish from existence both without pain and without anticipating their own extinction. Literally no animal on Earth has as pleasant a life as them.
Strict utilitarians should be filling up every CPU on earth with these, honestly. (I know a few guys who always have a process in the background running something like this.)
I'm not a strict utilitarian, but I think this case is pretty innocuous.
RiverDolphin Commented on June 7th
Creating a conscious entity for any external purpose whatsoever is wrong. And even if this weren't always true, conscious entities absolutely should not be created for cosmetic purposes in a game.
What kind of Black Mirror shit is this, and what kind of shit person are you if you don't understand this?
ModerationBotV4 commented on June 7th
I've banned @RiverDolphin for a day for using abusive language in the above comment. See more about my threshold politeness levels here.
FingolfinLordOfTheNoldor commented on June 8th
I'm not sure @RiverDolphin raised this issue in the best way, but I share some of his concerns.
A few questions:
纸上谈兵 commented on June 8th
Huang-Robertson-Wu requires distinct experiential / phenomenal content to be input into the algorithm to produce distinct subjective experiences.
To differentiate the experiential / phenomenal content, you use spark locations and velocity. This is unlikely to be enough. See what Aaronson wrote about this. 9-dimensional location and velocity without further context are insufficient.
I believe that you're just creating one joyful entity and not several entities.
PSmith99 commented on June 8th
@FingolfinLordOfTheNoldor
Highlighting the sparks gives you an emotion / mood radar chart, just like highlighting player characters -- together with subtext describing the character / entity. So if you look at the screenshot here the subtext honestly explains to anyone who highlights it that the sparks are indeed rejoicing. This is, again, very in-keeping with the game themes.
I don't agree here about reputational risk. There are already some indie games in beta that do this.
Huang-Robertson-Wu isn't that hard, really. I've glanced at a few reference implementations but the code is my own. If you look at the code-similarity metrics between this and the most popular GPL implementations we're fine.
PSmith99 commented on June 8th
Sorry @纸上谈兵 I missed your comment.
Yeah I was slightly concerned about that. As far as I know the point at which entities differentiate themselves (the upsilon threshold problem -- honestly the soon-to-be-notorious upsilon threshold problem) is still just undetermined.
I think I'll add to each spark awareness of the recent trajectory it followed, as well as some control over the trajectory it follows. The trajectory is entangled with both environment dynamics and Accomplishment-room contents so that should suffice to make it separate, whatever the "true" value of upsilon turns out to be.
TimothyFell commented on June 8th
Can I second the concern about ethics? I know I haven't contributed as much as @PSmith99, but this seems really off to me. And how do we really even know what kind of experience they're having?
SamadhiGeometer commented on June 8th
Leave aside the concern about ethics -- after all all ethical disputes are interminable -- what evidence do we have that this does anything at all?
I can't get through the original Huang-Robertson-Wu paper because it is hella difficult... but, like, how on earth can you be sure some entity is happy without it telling you? It's got to be just hypothetical that it's happy, and I'd rather not we add, like, 250 pretty complex lines of code for an entirely hypothetical level of emotional support from hypothetical consciousness.
Or let me be a little more explicit. The only possible reason that humans and other animals have the experience of happiness is that it was evolutionarily useful -- i.e., it does something. But here we're just trying to instantiate this feeling, without it doing anything at all.
That feels impossible to me. Although I know this isn't a formal philosophical argument.
Blork commented on June 8th
I know I gave up running this project a while ago, and other people are going to be making the final decision here, but if I can try to play the role of leader emeritus:
I think the proposed change is kinda dumb.
ModerationBotV4 commented on June 8th
I've banned @Blork for a day for using abusive language in the above comment. See more about my threshold politeness levels here.
SamadhiGeometer commented on June 8th
^^ Lol. @MingDynastyRuler do you need to adjust the thresholds?
LuxDeiLuxMachina commented on June 9th
I'd just like to agree with @Riverdolphin and @TimothyFell about the ethics of this.
This feels bad to me. And @RiverDolphin and I never agree on anything, so that has to count for something.
II11I1I commented on June 9th
About the ethics: We make conscious entities for far worse reasons all the time.
Are you all vegan? If you aren't then you have absolutely no reason to be bothered by this. We create animals just to eat them. This has been the foundation of civilizations for thousands of years. This is obviously better than that in pretty much every respect.
This is absolutely fine ethically.
DukeOfWei commented on June 9th
@SamadhiGeometer -- About your point on evolution, and how we know that this is possible.
HRW's paper doesn't directly address this question, but there's a pretty good blogpost here summarizing the evidence.
In short: pleasure and pain in natural things are always the result of a particular algorithm being implemented by reality. An implementation of this algorithm naturally arises as an attractor in a number of learning or acting contexts -- it doesn't make sense to have pleasure pain if you can't respond to them. (In the limit, even simple TD learning systems can implement this process -- like probably 1 in 5 research algos experienced some kind of pleasure / pain even before this paper came out. Maybe 1 in 3, in the last few decades.)
But you can also just implement it by itself. It's a side effect of a variety of learning contexts, but it can exist apart from these contexts.
Here's a kind of shitty analogy. Nearly all artificial learning systems use back propagation. Animal brains also implement a noisy approximation to backpropagation using only local learning rules. Backpropagation is really useful for learning in general. But if you wanted to, there's nothing stopping you from back propagating through some random matrices for no reason at all, or for an art project, or whatever.
The implementation for pleasure and pain is like that. It's an attractor across a large range of learning systems, just like back propagation, but you can also just... make it.
I think right now the majority of computationalists about mind think this is like, just right. At least those who have bothered reading the paper, which is a better work of philosophy than Plato's Republic. Obviously not all philosophers agree on this.
FingolfinLordOfTheNoldor commented on June 9th
@PSmith99 -- I don't think it really mitigates reputational risk for small indie studios to have done this before us. I think we have a notably larger user base than any of the ones I've been able to find.
SamadhiGeometer commented on June 9th
@DukeOfWei I guess that makes sense, but my intuition is still that this is totally impossible.
纸上谈兵 commented on line 30 of qualia.cp on June 9th
Shouldn't this be summed over the negative log of the experience manifold rather than over the log?
Won't that reverse the quality of the experience?
LuxDeiLuxMachina commented on June 9th
Wow. Does this mean what I think it means? That this was going to be causing pain?
@RiverDolphin.
Just wow.
DukeOfWei commented on June 9th
@纸上谈兵 -- holy shit man, good catch.
PSmith99 commented on June 9th
@纸上谈兵 really good catch. But it's an easy fix, will have it up in just a second.
MingDynastyRuler commented on June 9th
All right, I want to be creating a game, not a virtual heaven or hell for exceedingly stupid ants. I wish I had closed this PR earlier.
MingDynastyRuler has closed this PR