Ep. 26 - AI, Authorship, and the Editorial Process

(knocking) (knocking) Hey, Brett. How are you doing today? Hey, Karen. I'm good. How about you? I'm all right. You know, getting ready for some more podcast recordings. What are you working on right now? I was going to start working on that, but just give me a second. I'm just finishing typing something into a chat chat B.T. Oh, what are you asking for help with this time? Well, if you must know, I'm asking it for help on how to introduce this episode. Are you really doing that? Okay, no, but that seems like a good excuse to bring up the topic of AI. Oh, yes. It's true. AI, it comes up in almost every conversation I have these days. It does. I feel the same way. But I mean, you can use it for everything, right? Like, drafting an email, coding data. Done, done. And you could even use it to make the best race crispy treats ever, which yes, I may have done last week for my son's birthday. It is getting better and better, right? We can do more and more with it. We can do more and more. And, you know, I used it just yesterday, true story to help deal with a car issue. And if I had used it a few minutes earlier to not do the thing to my car that I did, I would have been better for it. But all of this would not have been possible before it. So it's really quite amazing. And I think that what folks maybe don't realize is as the editorial team at JMR, we get a lot of questions from people out there about what they can or can't do with AI as either an author or as a reviewer. And I felt sometimes a little bit nervous about responding because we don't always have a clear answer ourselves about kind of where the boundary is. And these questions have prompted us to seek those answers. And so we thought we would bring people kind of into those discussions with all of you. So today on this episode, we're going to go behind the scenes of using AI in this special episode of how I wrote this. Ooh, sounds exciting. Let's meet the team. We are all here together today. I've got all of the co-editor team. It's just like we're having one of our biweekly meetings. And this is a special opportunity for our how I wrote this listeners. We do all five of us come together on a regular basis. And one of these meetings we were chatting about all the questions we get on AI. So many questions. So we're really excited today that we're going to just kind of dive in and talk about some of the most common questions that we get. But I guess we'll go around maybe briefly and we want to introduce ourselves. Obviously I'm Karen Winnerick. And you might have heard me frequently when some of our how I wrote this podcast episodes. Hi, I'm Brett Gordon. I'm the other co-host on the how I wrote this podcast and was very excited for us to be able to pull together this special episode, which as Karen referenced, we do actually meet biweekly. And we have met biweekly for now two and a half or so years discussing a variety of fascinating topics relating to the editorial process and management of JMR. So we have been led throughout this entire time by the esteemed Rebecca Hamilton Rebecca want you say hello. Hi everyone, Rebecca Hamilton. I have really enjoyed these last two and a half years with my co editor team and we have of course gotten many questions from both reviewers and authors. But the one that seems to come up most frequently these days is how is it appropriate to use AI both as an author and as a reviewer. But we're going to jump into that very soon. But let's first make sure we say hello to our two other co editors. Rago. Great. No, thank you. August company. Wonderful to be here seeing a new side of Brett and Karen as broadcasters. Wonderful to be lovely to join. I'm going to go for it. Good morning. Good evening from the past, I guess. 12 hours, 30 hours ahead of you guys. My name is couple to Lee and it's been an absolute joy. Two and a half years working with everyone. And I am looking forward to being a part of this podcast. I'm not sure if we have all the answers but if we discuss all the AI questions, we do want to make a clear disclaimer that we want to emphasize these are just our personal views. And I'm going to go for it. How do you think it could be most useful as an author? I have really appreciated ChatGPT for its help with coding and R. I have used SPSS for years and years and years and the ability to do more sophisticated analyses like a multi-level meta regression in R by creating code that is called from SPSS and then returns back to SPSS has been terrific. I would not have been able to figure that out very easily by myself, but I really have appreciated just being able to tell ChatGPT what I would like to do and what kind of data I have and then the code is something I can plug right into SPSS. So that's been great. Awesome. I think I've been using it in more basic ways, but a little bit, even just in some stimuli development. So a mock article that I want to look like a real article, but it's very specific in controls for various factors or even some images where I want to just modify an image for this stimuli, one particular factor to add more people or change something around. I've used it a bit in those. So I think all of these are totally appropriate ways to use AI. And this is why we do ask that authors disclose when there's somebody in manuscript the way in which they've used AI. And so hopefully you're hearing all these different ways that examples and gives you a little more confidence that it's totally okay to disclose the fact and if you did use it, in fact, should disclose that you used it. But the use cases out there are pretty broad. Sometimes, especially in the beginning, we got other questions along the lines of, you know, what was kind of an inappropriate, was there some inappropriate level of using AI to the point where AI would be sending a kin to a co-author? And where we landed on that mostly is that in the end, you're as the author or authors responsible for all the errors, good or bad, that AI might have produced and you're responsible for all its output. And because the AI can't take responsibility for itself, it has unable to be a co-author and should not be a co-author, a name-bizic author. And so this is kind of created sometimes a blurry boundary between what kind of feels like sort of minor help on something like re-wording a sentence or creating a better plot versus someone, you know, using AI to kind of help implement fully a model or even extend elements of a model. In the end, you take responsibility for the whole thing. And so that's important to kind of point to highlight. And Sage does distinguish between the levels of AI assistance you receive in their new policy AI assistance where no disclosure is required, actually applies to things like improving your grammar. So you can use it as a copy editor and that's, you know, perceived by Sage to be something that you don't even need to disclose. But as Brett said, as you go into, you know, more depth with AI, that becomes something you should disclose. So if you're, you know, really, really using it from more than more than the grammar and structure, you'll need to disclose that. And then there are also prohibited uses of AI. So they use a model where they distinguish between three different levels. You know, that brings up a good example, though, I think that I think all of us have kind of seen in this space now, which is okay. So it's, oh, you don't have to disclose it when it's when AI is a copy editor. And so that would be obvious things like, oh, I forgot a period here there. Something wasn't capitalized properly, you know, you're very, very basic errors of that sort. Copyators also help you with a little bit of grammar, you know, something was past tense. It should have been present tense. Maybe use passive voice. But as you can kind of see where I'm going, okay, so at some point, someone is slightly more than a copy editor and kind of helping you write more clearly. And I think, I guess for myself, I had to kind of think of what the test would be, is that the text would have to contain sort of the exact same ideas originally that I wouldn't expect a copy editor to kind of add or substantially remove kind of concepts and, you know, ideas and thoughts and points and all that. But if it did, then that seems to move beyond copy editing to real writing help. And I can imagine that for some people, that line is hard to identify. Maybe for myself included. So I'm curious if others have reactions on that. And I think if you're in doubt, disclose. That's always the right way to go. Yeah, because again, it's not a bad thing. We're not like, oh, you say, I, automatic negative. We're actually like, well, that's common. That's kind of normal. If you didn't use it, are you really, is that an accurate response? If you're saying you didn't use it, regular, were you going to comment on something? Yeah, no, I think I really like the way the backup put it, which is be conservative. You know, if you've used it in any form, I think it's just better to reveal it as opposed to, you know, later on kind of trying to figure out as an author. Where was that blurry line? Did I just use it as a copy editor? Will it kind of for some places or some paragraphs? Maybe did it add something that I hadn't thought about? Just be conservative. Just say you've used it. And I think that covers a lot of these issues that perhaps are coming up. Yeah. And have there been, you know, we're saying like, these are good spots to use it, but there is this blurry line. And, you know, of course, using it to the extent where it's more than just kind of a copy editor or assistant doing these basic level tasks, like, you know, plots and things. There are some areas where it's not so good, right? So, Raghu, I really liked how you were saying, you know, it gives you this basic kind of overview of an area, right? But it's actually not good at writing a literature review because it's not really, doesn't really get the nitty gritty details of research papers, right? I don't know. Cap, what do they think? Now, I think we just got to remember just a little perspective. You're still in the dial-up era of this technology, you know? And people are actually started to actually train models for more and more specialized purposes. Right? If you look at it, there's a company in California that is actually paying people almost three, four hundred dollars an hour to train LLM's for specific professions like your ideology. So, you know, I mean, don't be surprised in three years' time. It becomes incredible at doing literature reviews. Just to go back to Rebecca's point, I think that's such a nice simple rule of thumb when in doubt disclosed because we can't see the line after a few years. The lines will become very blurry. And there's that's true. That's true. That's true. Disclosure. Yeah. Now, Brett, I know you had a very nice, I think a few months ago, LinkedIn article, would love to talk, you know, if you can describe it, like how did you think about what would be a good way to put large language models and kind of the way you should think about them? Well, thanks for team that up. Yeah, you know, that was all based on, that was based on our discussions as you all know. So, you know, when we have these bi-weekly meetings and our conversations range from issues that we're seeing in different papers and just questions that we're getting, I guess we all had that same question of what was the boundary between kind of appropriate AI use and something that seemed more like a co-author. And so, yeah, so the view I took there, which was informed by just others, I wouldn't claim any, you know, ownership on that, was that to think of AI as a research assistant and that you might ask an RA to help you write a research, how we try to literature review, help you with some code, create plots, run subjects, and you might think that research assistant on, in the acknowledgments, but you wouldn't necessarily give them co-authorship unless they had created something more substantial, some kind of clear intellectual contribution. The distinction then is that even if somehow you thought that Nell-Alem had created some true intellectual contribution, which, you know, some does happen, perhaps, the thinking right now is still that it doesn't deserve co-authorship because in the end, it's not accountable. It's not responsible for its errors. If something was wrong, knowing can kind of go to Cheshire BT and say, "Oh, you were wrong about this paper and you know, that's bad and you know, here's what we're going to do." They won't care. It's all about you, the authors. And so in the end, the authors have total control, total ownership, and AI kind of doesn't rise to the level of a co-author. Even if it does perhaps create a more significant contribution. So I think just authors should just kind of keep this in mind as they continue to develop and explore new use cases for it in their own work streams. Just like discovery through AI, you know, I mean, if I spend five hours going through, say, 50 prompts and AI takes me in a direction and I discover something, I'm still the author. You know, the idea might have come to digging into the AI, but still I discovered it. So I, if I derive the benefit, I also have the responsibility to stress test it because at the end of the day, the author is the one who takes the responsibility. And I think that's another great example where you could imagine that at the same end being achieved using an RA. Right? Perhaps a very good RA where you, you know, directed them to read this literature and then they found those other 10 papers and dug into them and all that back and forth in many meetings, etc, etc. But in the end, you can imagine, we can imagine so many tasks that kind of would be completeable by an excellent RA. And I feel like I've heard someone out there describe Chaggit and similar as like having an infinite army of RA is essentially. And that is I think a really, a really apt kind of analogy. So it is important to check your RA. I think that's true. where we really have to encourage authors to do their due diligence. There's always been the temptation, maybe, to cut things short, where maybe you want to cite a paper and you haven't read that paper. Yep. You have to read and you have to be sure what you're citing to make sure that you characterize correctly the prior work. And this is no different from other tools and technologies where you really have to be sure what you're putting in your paper as an author. And if something goes wrong, you can't say the audio muted mistake. It's your paper. Yeah. Yeah. Yeah. That's in the news out there. So I've heard. Okay. Shall we move on to the next big question, Karen? Yeah. I think so because it is another one we get a lot. Rebecca, I guess I'll direct this to you because the common question is, how can I use AI in the review process? As a reviewer. As a reviewer. I agree. You too, not even think about using AI as part of the review process. So it's very, very, very specific about this. And the core is that when we receive manuscripts from authors, they are submitting to us confidentially. And we cannot upload their work to the cloud to. So AI because that eliminates their confidentiality. So you might ask about various corpus that you might have at your institution or on your own computer. But sage is drawing a very careful line where we're not as reviewers or editors allowed to upload any material submitted by authors to a large language model. So that is the logistical hurdle. Also, we really don't want a summary of a paper when we submit, when we ask you to review it. What we would like is your expert judgment on what kind of contribution is this paper making? And what could the authors do to make their claims more credible? And so far, those are not things that AI has been able to do successfully. So we really rely on the expert judgments of our reviewers and really hope that you take that seriously, as well as the confidentiality with which we send you papers. So just like you're never out allowed to share papers you receive for review with your students, with your colleagues, you're also not allowed to share them with large language models. Yeah. And I think about it as, you know, would I write a paper? I'm writing it. AI is not writing it. And I'm putting a lot of time and effort into that paper. And so I kind of think of like a do unto others, right? As you would have them do and do you. So I put a lot of effort in. I want the reviewer to put some efforts into reviewing my paper and actually giving me constructive feedback. Right. That's kind of part of the serve. Service to be a reviewer. That's how the field functions. But that requires that you put a little time in an effort into actually thinking what are what is constructive feedback. Right. How could this be improved? I think that's that's pretty key. Do you think, Brian? To just be a little bit contrarian. Yeah. Of course. I expect another one. Of course. And I'm curious what others think. And I'll just repeat that disclaimer that this is not does not represent the official views of the journal, sage or any other official organization. But yeah, I put a lot of work into a paper and I want others to give me their honest, you know, feedback on it. But if the AI was really good, maybe I'd be happy with an AI's feedback because not all reviewers, unfortunately, provide great feedback for a variety of reasons. Time constrained lack of poor match with the journal, except with the paper, etc, etc. So if a journal could offer you, hey, we have this excellent AI that we've actually trained somehow, you know, if the gallery was allowed it. And it's going to be great feedback and give a great recommendation to me, the editor, like would anyone here be interested in using that as a reviewer for themselves? It might be something we could eventually provide to authors. So on our JMR platform, maybe we provide a service for you to pre review your own paper and get that feedback. I think that would be a good use of these are for files. But when people submit their work to the journal, they expect expert judgment. And we are very lucky as a top journal that our reviewers mostly say yes when we're fighting them to review. And they, you know, really almost all the time do an amazing job. So I'm very frequently just, you know, taking really taking a very seriously because it is, yeah, our reviewers are terrific. Yeah, I think reiterating kind of one of those things where, you know, what is most important, like when is the human judgment most central. And that is like identifying the path forward. But maybe if the AI is like Brett, you're saying, you know, perhaps in the future, they're trained so well that they can give us those path forward. But I do think one thing this kind of brings me back to is we've talked in earlier podcasts like being a great reviewer, how to be a great reviewer. And also like what are the benefits besides just like doing a good deed and contributing to the field via service. One of the good things about reviewing is I'm learning in the process, right. And so if AI is doing the review, I'm not actually learning, you know, what the existing research is doing or the kind of most current research is being submitted. I've got a reader in my self and be like, oh, that's really interesting. Really, I'm learning from that. Oh, I haven't read that paper. Let me go back and check that one out before I complete my review. So we don't want to forget those benefits to the self of learning in the process. I don't know. What do you guys think? I just want to go back, you know, like to the point that I said, you know, I want human judgment because to be very honest, that is the biggest contribution we make as a species. But there is also an element of fact checking in the review process, right. And maybe that is where AI can play an important role. Maybe a minor nuance of a model, a very complicated model, web appendices are best friends, right, especially by the time you reach the 75th page of a web appendix. Tiredness can kick in, right. Maybe that is where AI can play a limited, clearly defined role. That, hey, I want you to check this model and identify any potential not to replace humans, but to complement human reviewers, not today, maybe in the near future. Verifying proofs in a theory paper, which, you know, I already know of people who are using them to create proofs, so verifying proof seems like a fairly clear use case. Yeah, right. Yeah, Brett, your point was very well taken actually. You were, I think not being contrary in some sense. I think I like it. Maybe I'm a contrary, I'm not sure. But I think this idea that you can use and very similar to what Rebecca said, perhaps both as an author, if you want to submit to a journal right now, for example, we have these latex style files. Maybe you can have the AI reviewer for that journal, where before you submit, you run it through that to make sure that you as an author are doing all you can to make your sort of manuscript as good for that journal as it can be. That could be offered by the journal itself as Rebecca was mentioning like a pre reviews you want to call it that or you know find an LLM that will do it as an author. You can of course do whatever you want with the paper, but as a reviewer, you won't be obviously as we mentioned. Don't upload anything. I think one thing that may be interesting is as a reviewer, for example, I have seen many times that certain reviewers can be a little bit more aggressive than they should be. So, you know, I might go back to that reviewer and sort of say, you know, you need to kind of make sure that you gave you credit to the authors and so just you know, tone down the review in terms of just you look at the content, not the person. Is that something that AI can help with in terms of just, you know, it's like poppy editing your review. So it's not about the paper, but it's the tone of voice. I don't know whether people have any thoughts on that. I'm going to throw that a little bit with that because once the reviewer writes their review, it's their work product. However, that review often contains very detailed descriptions of what's in the paper, you know, maybe even excerpts. So that's where it becomes, you know, a breach of confidentiality again when you upload your own review to an LLM to improve your tone or improve something about it. So we really have to ask you as a reviewer to be very careful about only using LLM's when you said in very abstract terms, not disclosing the authors intellectual property when you upload those those review notes. It would almost be like it's probably not a very good review if there's not enough detail that's not sharing some of the confidential ideas in the paper, right? Then your reviews are probably not detailed enough. That's the challenge. Yeah, that's a great point. Yeah, there's some was a kind of a really huge contrast here of that we've said that you have greatly ways authors to use LLM as an author and basically very, very little. leeway to use it as a reviewer and Bordering on the line of probably just don't use it because just or Karen's point It's hard to kind of think of what text you'd want to put in there that wouldn't somehow Be related to the paper that was submitted under confidentiality and so What doesn't seem like it's very easy Stragi seems like the safest bet is just don't do it Similarly as editors. We're not using LLMs. We're You know potentially gonna do our own literature search on the side to see who might be a good reviewer for the paper but it's not gonna be a Possibility for us to upload a paper to engage in our editorial duties So with that let's move on to that topic of we have to make handle a lot of manuscripts There's various you know tasks in terms of finding reviewers finding a ease Making the decisions etc. You know what is the possible role of AI in this entire process and Just as Rebecca noted as reviewers are sent more or less barred from using LLMs And they're that part of the process so are we as as editors Even though one could imagine ways in which you might want to use it author out there post this question and be of you know should for example AI be used to help with desergex for example that you know a lot we get in great volume to some degree And it was this kind of efficient use of editors or a's time because we do often rely on our a's to give us some opinion on those papers Without the unreasonable use case No We give each paper attention You know we as an editorial team read each paper that submitted to JMR and we are not uploading those papers to AI basically the same the same for the same reasons that reviewers are not allowed to do it We're not allowed to do it and shouldn't do it But there is one use case I've kind of wondered about which is just helping us find reviewers in the assignment process It's obviously we do we have gotten questions from people about kind of the manuscript central interface and What you see as an author is just a portion of the interface what we see is a bit more it is not the most intuitive interface or Slick interface perhaps and if everyone listening could see the reactions right now on zoom of my other co-editors I think what their inference would be is that I'm trying to be a little sensitive on this But yeah, we can all kind of imagine ways we'd like to make it better and having some kind of slick AI kind of suggestion tool would be nice Other's reactions on that or related. I mean I heard you like I think we were chatting about it You know one of our biweekly calls and I thought okay, I'm gonna I'm gonna try this like I'm I could use some more ideas for reviewers and I just you know maybe it's on me because my prompts aren't as good as they should be or something But and I would be very clear. I was not uploading the paper to LLM right? I'm just asking very you know who are experts in this particular, you know area right both like kind of the topic You know as well as methodology, you know who are some experts in this area and marketing and you know And it was giving me people that hey, I'd already thought of right? It's giving me the experts but the top of my experts that I also already knew I wanted like other You know and then I was like give me some more junior people that are publishing in this area right? And I saw I was trying to kind of go through the prompts but One thing it doesn't know and scholar one may not make this super you know helpful easily accessible Was like do they accept reviews right are they a constructive reviewer right? And so you kind of have to play on Do they actually review and provide quality reviews and back and forth? But I mean, I'm sure I can get better at it. I think there's probably some potential there But it wasn't like it gave me 10 great ideas for reviewers that I'd never thought of before and it was super helpful in that way So I don't know what what are your experiences? I can I can still learn obviously Yeah, I mean, I think at least when I have seen it being used for me Manuscript center as as was mentioned bread mentioned. I get it's an I would say it's non-intuitive and One has to get good at it, but I guess that's you know the the last one a half years we've all gotten better at it And so we've been able to find but I think it's a great point Maybe this is something that scholar one or manuscript central should offer on their website So that they integrate in LLM and all it does is basically takes all the reviewers that have the hat in the past or They're I don't know the ERBs and it helps us find the long tail and because you know the more popular ones or top of mind are already there with us But perhaps that's the opportunity in some sense, but at least it's not there as far as I know on the website And since we have to do the work, but there is perhaps an opportunity And we would certainly be great if um we could create our own database of Reviews and you know submissions to the journal and use that to find reviewers that that would allow us to Better identify those in our day-amour community who would be a great fit for a manuscript reviewer Yeah, so we have a lot cut out for uh scholar one and manager of central like a whole list of things that they need to do to update And I'm sure that the new editor team is listening to this and I have every faith in their ability to tackle These tasks and more Actually, there are some low-hanging fruits think about it right right now. We have to manually search for How many reviews as somebody has you know, you can quickly just right now You can actually get a dashboard immediately right like if you want to choose an a what is the current Or ERB members what is the current load Uh If that can be developed as a dashboard that that would be that'll save you a lot of all of us a lot of time Great idea instead of the reports Perfect. Yeah, I agree I'm sure we'd also all be willing to take a small pay cut to help fund that Development oh wait, we don't get paid anything Yes Yes Sage you must have some funding available for this you know make our free jobs easier just a little bit Maybe someday I You know going back going with our discussion of you know having a eyes. I like the idea of this Well, I liked and it was also nervous about this idea of a of a pre submission AI reviewer because you kind of imagine So I liked the idea because I thought that was a cool idea You could also imagine then authors sort of optimizing for the test you know And I guess if you because you'd had to limit the number of times someone could submit like once let's say For any given paper Because always you just kind of keep on submitting until the AI reviewer comes back and says this is the most amazing thing I've ever read you know knows that this is perfect automatic except exactly And then you then they'd submit and you know persons gets they get distracted because oh well, you know, sorry But in an interesting idea would be if journals to some extent come would compete with each other on having a good AI pre-reviewer To smooth out the eventual reviewer process right like you know, well, we we are the some top journal We have an amazingly well-trained reviewer for you that helps you the author is a service to you The author and feet the field that could sway some people on the margin perhaps but Managing that Not having kind of the whole process kind of go off the rails was kind of porncentives seems not obvious to me Yeah They're just like this during the the biweekly's but to everyone Couple how about you go? Sure No, I was just like I would be a little worried if JMR or JM or marketing science had an AI pre-reviewer Because it unnecessarily sets up expectations if there is an AI Available as a pre-review feedback I would not go to the viewer say okay friendly friendly review feedback with no liability to the journal Friendly R2 Well, that's a noxie moron Sorry, this is a yeah, no, no, no, this is a great point actually I'm just thinking out loud and some of you are way qualified than I am like imagine a world in which there are Multiple journals and each journal has to now decide whether to offer that AI Three reviewer. I don't know whether in the equilibrium the quality of papers will actually I mean, I don't know what'll happen to it because in some ways just like Brett said It becomes so super optimized, but then everybody else is also optimizing and so then does it sort of become more homogenous In terms of what's coming in or do some journals specialize In a certain type of AI reviewers so to speak Uh, not quite clear. I'm sure people have looked at it. Maybe in a different way, but this might be an interesting thing that might be in the future I think it'll um sort of reduce the heterogeneity we get so many papers from all over the world and Some authors really need help to bring their papers to the level We're they're going to go through a full review process that jm r and I think this might be a really nice way of giving them that sort of pre review feedback to You know, it will increase homogeneity, but I think in a good way because there is a pretty long tail right now of those papers that come in from universities all over the world and those authors who haven't had the same training who could really benefit from some chips. It'll be a brave new world. Whatever it is, I'm not sure what it'll look like exactly. All right, so you know, I'm not sure how many years it's going to take until this this podcast episode is already outdated because things are changing so rapidly in this area. But nonetheless, let's let's put our little predictor hats on and say in five years as a researcher, I will now be able to do film in the blank. What? So Rebecca, what do you think? Five years from now as a researcher, what will you do? I think I'll be able to have a single interface where I ask questions of various types and I don't have to have several different dashboards open at the same time. So if I want to do literature of you on a statistical analysis, I go to different places to do those and it will be seamless for me as a researcher to ask my questions or seek input from a single source rather than these multiple sources that I still then have to integrate. Oh, nice. No, I think great point, Rebecca. Yeah, good one. I think for me, it would be in a similar spirit, but sort of helping me kind of marketing itself is going pretty large. There are so many different parts of marketing that I probably don't know anything about. So this would be a great way for me to have. Let's say an AI agent that's focused on cultural theory. There is an AI agent that's focused on perhaps eye tracking studies. There's an AI agent that's and so that way I go out there and I find out tell me what's the latest and greatest and that's optimized for that. So just increasing my own sort of knowledge base. I would also love to have that. But I just also want a specific agent that a stress checks a paper, especially for any blind spots that I might have missed because the worst thing that can happen is, oh, you did not consider this aspect because it was a paper written in say 1984. And it is, it might not be very cited nowadays, but it is fundamental to the point that you're tackling that those blind spots can happen. So if an AI agent can help you with that, that could save a lot of unnecessary trouble. I love that. Yeah, almost like kind of going back to when we were talking about the AI, like where we're like put it in AI review, kind of initially for ourselves, right to use as an author. Yep. That like checking for blind spots. Great. I think to round it off, I think in five years, I will be able to code up anything I want via vibe coding. I've been using a co pilot within VS code. I know some PhD students are using cloud code and kind of related tools or cursor. And what I've heard from some of them is something akin to, you know, in the beginning, they were sort of using it kind of half coding themselves, half using the tool. And then at some point, they sort of gave in to the tool more entirely and it just went better. And that's what I am starting to move to. And you know, it does a very, very good job. And it saves me an amazing amount of time. And I imagine that in five years, it's going to be just at a level I can't even quite be into imagine as of now. Now in five years, I may not have enough time still to actually use it. I think that the efficiency gain is just going to be amazing. And it's a little scary, but also amazing. This has been fascinating. Like I said, I, you know, I think I've mentioned in many podcasts episodes. I'm not the most tech savvy and that includes AI. So it's been educational for me. And I hope it's been educational for the listeners too. Not just think about the future because of course we don't really know what, what that will look like in five years. But how, you know, using it as an author is okay and acceptable. You don't have to check no and it's okay. It's good. It's better to err on the side of disclosure. And that's not a bad thing. But also keeping in mind, you know, kind of the trust, the confidentiality that's asked the reviewers and not breaking that with AI. Are there other closing thoughts guys? I think we'll see everyone again in two weeks that are same co editor time same co editor channel. Same zoom log in same zoom log in exactly. Some of us running from child care drop offs. Others trying to just get ready to go to sleep. Full gamut. Yes. Okay. Well, have a good day or good evening. Been fun. This episode of how I wrote this was produced by me Brett Gordon and me Karen winner it. It was ended and mixed by Andrew mayor weather. Special thanks to Regu, Capitol and Rebecca for joining us to discuss all things AI and research from writing to reviewing. And if you like this episode share it with a colleague or friend and give us a follow on your favorite podcast app. And if you have any questions, send us an email. We might just cover them in a grab back episode later this season. Thanks for listening. We'll have another episode for you next month. [BLANK_AUDIO]