Since closed discussion venues and forums are the default for the vast majority of academic discussion, I am not super worried about this. The forum’s adoption seems to me to have overall been a pretty substantial step towards making the field’s discussion public. And the field overall also has vastly more of its discussion public than almost any academic field I can think of and can easily be responded to by researchers from a broad variety of fields, so I feel confused about the standard you are applying here. Which other fields or groups even have similar forums like this without very high standards for membership? And which one of the ones with membership display the comments publicly at all?
While I agree with you on the object-level claims, I don’t think this is how it would be perceived. By default academics work on projects that (eventually) get published in conferences / journals (and in CS these are often free to access). If you see AIAF as part of the same reference class as conferences / journals, then AIAF is certainly much more closed in terms of its explicit policies.
(The object-level counterarguments are (1) AIAF is more like the conversations that researchers have with each other when attending a conference, which are typically private and (2) while conferences in theory have double blind reviews, it is usually the case that in practice you can’t contribute to the conversation without following a set of norms of the field, many of which are only tangentially related to quality and only known to existing academics. But (1) is not obvious to people who don’t follow AIAF much, and is in fact false of OP’s post in particular (which really is quite close to an academic literature review / survey article that could be published) and (2) is not commonly believed / accepted.)
If you see AIAF as part of the same reference class as conferences / journals, then AIAF is certainly much more closed in terms of its explicit policies.
I am a bit confused by this. It seems that submitting a post to the AIAF is much easier than submitting a post to a journal, and is much less of a hassle. So in what sense is it much more closed?
Only members can submit a post to the AIAF, whereas anyone can submit to a conference / journal?
(Maybe you count “writing a post on LessWrong” as a “submission” of a post to the AIAF, that’s not how I would expect it to be perceived by default, and not how I perceived it prior to this post.)
Yeah, I would count writing a LW post plus maybe messaging an admin about it as the equivalent of “the barrier to participating in the discourse”. We don’t have a submission system as such, so of course it isn’t exactly equivalent to submission, but the overall barrier to participation strikes me as substantially lower.
Both before and after reading the post, I think that AIAF caused AI alignment discussion to be much more publicly readable (relative to academia).
After reading the post / comments, I think that the AIAF is more publicly writable than academia. Before reading the post / comments, I did not think this—I wouldn’t have said that writing a post on LW was “submitting” to the AIAF, since it didn’t seem to me like there were people making a considered decision on whether to promote LW posts to AIAF.
Both before and after reading the post, I think that the AIAF will not be perceived (at least by academics) to be more publicly writable than academia.
It seems to me like the main problem is that AIAF currently does a bad job at giving a naive visitor an idea about how it’s setup works. Do you think that a better explanation on the side of AIAF would solve the issue or do you believe it to be deeper?
I’m still confused by half the comments on this post. How can people be confused by a setting explained in detail in the only post always pinned in the AF, which is a FAQ?
I want to push back on that. I agree that most people don’t read the manual, but I think that if you’re confused about something and then don’t read the manual, it’s on you. I also don’t think they could make it much more obvious than being always on the front page.
Maybe the main criticism is that this FAQ/intro post has a bunch of info about the first AF sequences that is probably irrelevant to most newcomers.
It would for example be possible to have a notice at the bottom of alignment forum pages to user that aren’t locked in that says: “If you aren’t a member of the alignment forum and want to comment on this post, you can do so at [link to LessWrong post]. Learn more [link to FAQ]”
Such a link would strengthen the association between LessWrong and AIAF for a naive user that reads a AIAF posts. There might be drawbacks for strengthen that association but it would help the naive user to get the idea that the way to interact with AIAF posts for non-AIAF members is through LessWrong.
I agree that most people don’t read the manual, but I think that if you’re confused about something and then don’t read the manual, it’s on you.
The goal of AIAF is to be well accepted by the AI field. If people from that field come to AIAF and have a lesser opinion of AIAF because they don’t really understand how it works, you can say that’s on them but it’s still bad for AIAF.
I agree that most people don’t read the manual, but I think that if you’re confused about something and then don’t read the manual, it’s on you.
I think responsibility is the wrong framing here? There are empirical questions of ‘what proportion of users will try engaging with the software?’, ‘how many users will feel confused?’, ‘how many users will be frustrated and quit/leave with a bad impression?‘. I think the Alignment Forum should be (in part) designed with these questions in mind. If there’s a post on the front page that people ‘could’ think to read, but in practice don’t, then I think this matters.
I also don’t think they could make it much more obvious than being always on the front page.
I disagree. I think the right way to do user interfaces is to present the relevant information to the user at the appropriate time. Eg, when they try to sign-up, give a pop-up explaining how that process works (or linking to the relevant part of the FAQ). Ditto when they try making a comment, or making a post. I expect this would exposure many more users to the right information at the right time, rather than needing them to think to look at the stickied post, and filter through for the information they want
I think part of the problem is that it’s not always obvious that you’re confused about something.
If you don’t know that the UI has led you to make wrong assumptions about the way it works, you won’t even know to go look at the manual.
(Also, as someone who has designed lots of UI’s...for many types of UI’s, if the user has to go look at the manual it means I’ve got something to improve in the UI.)
I think giving people better beliefs about how the AIAF works would probably solve the issue, though that doesn’t necessarily come from better explanations, e.g. I much prefer things like your suggestion here, where you’re providing some info at exactly the time it is relevant, so that people actually read it. (Perhaps that’s what you mean by “better explanations”.)
While I agree with you on the object-level claims, I don’t think this is how it would be perceived. By default academics work on projects that (eventually) get published in conferences / journals (and in CS these are often free to access). If you see AIAF as part of the same reference class as conferences / journals, then AIAF is certainly much more closed in terms of its explicit policies.
(The object-level counterarguments are (1) AIAF is more like the conversations that researchers have with each other when attending a conference, which are typically private and (2) while conferences in theory have double blind reviews, it is usually the case that in practice you can’t contribute to the conversation without following a set of norms of the field, many of which are only tangentially related to quality and only known to existing academics. But (1) is not obvious to people who don’t follow AIAF much, and is in fact false of OP’s post in particular (which really is quite close to an academic literature review / survey article that could be published) and (2) is not commonly believed / accepted.)
I am a bit confused by this. It seems that submitting a post to the AIAF is much easier than submitting a post to a journal, and is much less of a hassle. So in what sense is it much more closed?
Only members can submit a post to the AIAF, whereas anyone can submit to a conference / journal?
(Maybe you count “writing a post on LessWrong” as a “submission” of a post to the AIAF, that’s not how I would expect it to be perceived by default, and not how I perceived it prior to this post.)
Yeah, I would count writing a LW post plus maybe messaging an admin about it as the equivalent of “the barrier to participating in the discourse”. We don’t have a submission system as such, so of course it isn’t exactly equivalent to submission, but the overall barrier to participation strikes me as substantially lower.
To be clear...now having read the post and comments you do not consider it more closed?
I feel like we should taboo “closed”.
Both before and after reading the post, I think that AIAF caused AI alignment discussion to be much more publicly readable (relative to academia).
After reading the post / comments, I think that the AIAF is more publicly writable than academia. Before reading the post / comments, I did not think this—I wouldn’t have said that writing a post on LW was “submitting” to the AIAF, since it didn’t seem to me like there were people making a considered decision on whether to promote LW posts to AIAF.
Both before and after reading the post, I think that the AIAF will not be perceived (at least by academics) to be more publicly writable than academia.
It seems to me like the main problem is that AIAF currently does a bad job at giving a naive visitor an idea about how it’s setup works. Do you think that a better explanation on the side of AIAF would solve the issue or do you believe it to be deeper?
I’m still confused by half the comments on this post. How can people be confused by a setting explained in detail in the only post always pinned in the AF, which is a FAQ?
I think most people just don’t read the manual? And I think good user interfaces don’t assume they do
Speaking personally, I’m an alignment forum member, read a bunch of posts on there, but never even noticed that post existed
I want to push back on that. I agree that most people don’t read the manual, but I think that if you’re confused about something and then don’t read the manual, it’s on you. I also don’t think they could make it much more obvious than being always on the front page.
Maybe the main criticism is that this FAQ/intro post has a bunch of info about the first AF sequences that is probably irrelevant to most newcomers.
It would for example be possible to have a notice at the bottom of alignment forum pages to user that aren’t locked in that says: “If you aren’t a member of the alignment forum and want to comment on this post, you can do so at [link to LessWrong post]. Learn more [link to FAQ]”
Such a link would strengthen the association between LessWrong and AIAF for a naive user that reads a AIAF posts. There might be drawbacks for strengthen that association but it would help the naive user to get the idea that the way to interact with AIAF posts for non-AIAF members is through LessWrong.
The goal of AIAF is to be well accepted by the AI field. If people from that field come to AIAF and have a lesser opinion of AIAF because they don’t really understand how it works, you can say that’s on them but it’s still bad for AIAF.
Yeah, I was proposing something like this in this comment response to Peter.
I think responsibility is the wrong framing here? There are empirical questions of ‘what proportion of users will try engaging with the software?’, ‘how many users will feel confused?’, ‘how many users will be frustrated and quit/leave with a bad impression?‘. I think the Alignment Forum should be (in part) designed with these questions in mind. If there’s a post on the front page that people ‘could’ think to read, but in practice don’t, then I think this matters.
I disagree. I think the right way to do user interfaces is to present the relevant information to the user at the appropriate time. Eg, when they try to sign-up, give a pop-up explaining how that process works (or linking to the relevant part of the FAQ). Ditto when they try making a comment, or making a post. I expect this would exposure many more users to the right information at the right time, rather than needing them to think to look at the stickied post, and filter through for the information they want
I think part of the problem is that it’s not always obvious that you’re confused about something.
If you don’t know that the UI has led you to make wrong assumptions about the way it works, you won’t even know to go look at the manual.
(Also, as someone who has designed lots of UI’s...for many types of UI’s, if the user has to go look at the manual it means I’ve got something to improve in the UI.)
I think giving people better beliefs about how the AIAF works would probably solve the issue, though that doesn’t necessarily come from better explanations, e.g. I much prefer things like your suggestion here, where you’re providing some info at exactly the time it is relevant, so that people actually read it. (Perhaps that’s what you mean by “better explanations”.)