Renaud Joannes-Boyau: It's a pleasure to be with you today, in person and   online. I think we have quite a few people online, so that's really good, for this year's Peer Review Week panel discussion.  

So, if you don't know me, I'm Renaud Joannes-Boyau. I was recently put in the position of acting Deputy Vice-Chancellor Research. I'll be guiding us through today’s conversation on a topic that is rapidly reshaping the research landscape, especially in a peer review, with artificial intelligence, coming in. 

Today's session is an opportunity to explore how AI is already influencing peer review, and this is a reality. It has started already. We need to have this discussion. We need to consider the risks, the opportunities, and the ethical dimension of this transformation. 

I was asked to reflect on that and I am no expert. I am not really different to any of you that are trying to embrace AI as it's growing, finding it sometimes very difficult to keep track with this very rapid evolution with the number of tools that are popping up everywhere, trying also to understand how it works because this is very complicated.   

It requires a lot of personal investment in terms of time, sometimes even financial. and so I'm no different. I'm very passionate about AI and I have a tendency to be quite optimistic about things.  

So, while I understand that there's a lot of risk, I also see a lot of opportunities with AI.  

I think for the discussion about today, I want to set kind of the scene. The importance for us today is that it's a very rapid evolution. AI has been around for a very long time, actually since the sixties, even perhaps before, but it was extremely unhelpful at the time. I don't know if you ever played with it.  

I was able to be in a bootcamp AI, where we use that large language model (LLM) to play around and to create your own, language model. And I can say that it was very inefficient. Constantly glitching, giving me wrong answer. And if you were to picture that, I would say the first model would've been the size of let's say a dot on a piece of  paper.   

And if you were to compare in terms of the size of the ability of that AI to the new Chat GPT that is coming out, you'll be looking at something that is several metres large as a sphere.  

It's something that is growing so exponentially that it's hard to understand even the speed at which it grows, which is perhaps frightening, but also very exciting in terms of what kind of new discovery and other aspects in our life that we can use AI with. 

Why we talk today about peer review and AI is because peer review is a cornerstone. It's a cornerstone of science. This is what we use to make sure that what we do, we keep it, we keep the trust, we keep the credibility of science. So, if we remove that aspect of peer review, we remove the efficiency of science to produce things that we can trust that can be used in society.   

So that's why today's discussion is very important. It might be very technical in a way that we're talking about peer review for academics, but it has a profound impact on society. If we lose that aspect of credibility, of peer review, we lose that aspect for science. And in the world of today, we know that credibility and trust is a very important thing. 

So, I see opportunities for AI efficiency, that's for sure. Streamlining, we can use it for plagiarism, we can use it for statistical error. we can also even run it sometimes to see if there's methodological flaws compared to other techniques and so on. 

I can see this. I also see equity to access tools. For example, if you look at the world today, science is dominated by English speaking, papers and discoveries and so on. And we have extremely talented scientists that perhaps do not have access to English language. AI could completely change that. Giving access to people that do not speak English, the ability to publish their amazing work, but also to participate in the peer review process. 

So, I see that as an opportunity. I see also the ability to provide some consistency, with conflict of interest of finding hidden bias. With AI that can run a very large amount of data very quickly that human can't do.

So I see a lot of opportunities. There's obviously risk and we're going talk a lot about it.   

If I had to cite a few, of course there's a bias amplification, with, with AI and we know that historical data have bias. We know that there's some bias also with AI in terms of gender equity and so on. And I think we're going to talk about that.   

Today, there's overrepresentation of certain aspects in the large language model, and these can be amplified and this could lead to even more bias in the future.  

So that's certainly a risk. There is a responsibility in terms of, AI making mistakes. Who is there, accountable if AI does mistakes in reviewing something, who will be accountable for those things. So that's an important risk that I think we need to have in the discussion for peer review. There is also obviously the erosion of trust, with AI, the fact that if we start to have AI doing most of the work or all of the work of the peer review, then we remove the human having the ability to have this critical thinking being completely removed from the science, removed from that discussion, that dialogue that is between the scientist and the reviewer.  

This constant back and forth where we try to make the paper better rather than AI saying everything has to be done that way. And it's very formatted and not very constructive sometimes. 

And then there is the ethical dimension, probably the most difficult to discuss because.  

It's a very complex aspects to discuss.  I think one of the importance is disclosure, and I think we see that more and more everywhere, that we do not police the use of AI per se in a way. 

We just ask people to disclose, and then we let reviewers and other people judge this importance of the use of AI data privacy. 

This is something that pops up constantly when we upload the data. What happened to the data? 

We hear a lot of things, and we were just mentioning that with Jo, regarding, is it true that this data is deleted after a certain time? 

Is it true that no one can access it? Right now?  

It looks like no one really knows. And then there's the human judgment that we remove completely from it. So, I think this is important and we should discuss that. Is it important to remove the human judgment? After all, human judgments are also full of mistakes. Full of bias. The imperfect. AI can offer some aspects to that. I think the idea for today is a call for discussion for dialogue.  

Okay, it's about finding a balance where we can harness the power of AI, and retain the control of the peer review, retain the trust that we have with the peer review. And I think we need to avoid that question. Should we have a peer review that is human or machine?   

I think what we need to do with this discussion today is really how can we create the best partnership between AI and humans to get to the best peer review and to improve science in the future. 

As you know, the university is developing a lot of guidelines, research guidelines, but guidelines for everything regarding AI. 

I think we will have discussion across the university to improve those guidelines constantly as AI evolves. But we're going to start today. 

We are very lucky to have as a keynote professor Mark Hughes, editor-in-chief of the Australian Asian Journal of Aging. Mark is a former editor of the journal, Australian Social Work. His research relates to aging and age care service delivery with a particular focus on LGBTQ Aging and is currently leading a large medical research future fund project.   

Mark is also a member of the Australian Research Consult College of Experts. And, he will share a case study on how the journal has addressed the use of AI   in peer review. 

We will then continue and I will ask, the panel members, that will be with us today, professor Nigel Andrew. He's the chair of Science in the Faculty of Science and Engineering. 

Nigel is an editor-in-chief of the Austral Ecology, section editor of Austral Ecology and the interim editor-in-chief of the Journal Ecological Management and Restoration. Nigel is an active reviewer and assessor of Australian Research Council grant application, and has been a member of the College of Experts.   

Thank you, Nigel, for joining us today. Then we have Dr Jo Bradbury, she's the deputy chair of the Human Research Ethics Committee at the University. Jo is also a clinical and health service researcher with interest in nutrition and mental health and is based in the Faculty of Health. Jo has methodological expertise and biostatistics including clinical trials analysis.  

Thank you so much for being here. And we have Clare Thorpe, who's the director of the Southern Cross University Library. Clare has worked in academic and state libraries since 2001. And she's co-creator of a maturity model for evidence-based practice in library. H

We have prepared some question for panel discussion and we will wrap up with audience questions at the end.  

To begin, I'd like to welcome a keynote speaker, professor Mark Hughes.

I think one of the themes that might come through today is that none of us are experts on AI, or at least can I say I'm not an expert on AI. I think we're all on a big learning curve in relation to thinking about how artificial intelligence impacts academic work and research, and I'm definitely in that situation myself.  

What I was hoping to do today was to share with you a case example of our journal, the Australasian Journal on Aging just to share with you some of the challenges, that we are facing in the journal in relation to peer review in particular. and to just reflect on some of the opportunities and challenges that AI poses for us in working through some of these, challenges that we face.  

To start off, I thought I'd just contextualise a little bit in relation to the journal and give you a bit of background info on us. Our mission is to be an international multidisciplinary journal on aging, communicating the best evidence that influences policy and practice to improve aging in the lives of older people. 

So there's a very strong focus on translation and transferability implications for policy, practice and, and clinical practice. we publish original articles in any area of gerontology, geriatric medicine, and related fields. We are a multidisciplinary journal. We publish poetry, we publish, RCTs, we publish meta-analyses. We cover the whole gamut, basically of gerontology and geriatric medicine. we are published by Wiley, and we're owned by an, organisation called AJA Inc, which represents our three society members, the Australian Association of Gerontology, the Council on the Aging Australia, and the Australian and New Zealand Society for Geriatric Medicine.   

AJA was first published in 1982. It's currently in its 44th volume. My first article was in AJA in 1993, and I've published many, many articles in that journal since. I'm very committed to it, and have had a long history with it.  

We're now fully online, and we transitioned to continuous publication, this year.  

We are currently migrating from Scholar One to Wiley's, research exchange platform, a member of cope and we have three deputy editors and 19 associate editors, of those 19, 14 are handling editors, and that's what I mean by that is they handle articles that go out for review and make decisions on articles.  

Just a little bit background in terms of some of our key metrics. We're strongly committed to the Declaration on Research assessment, and very committed to equity in the reporting of research metrics. We do aim to report a wide range of metrics in relation to the journal. 

So in 2024, our, impact factors 1.8, site score 2.9. We are listed as Q1 in community and home care in SciVal. Look, to be honest, we're a middle range journal. We are not at the top, we're not at the bottom, but we do have a very strong commitment to supporting ECRs, to supporting clinician practitioners to get their work out and published.

We do have, very high standards though. And our acceptance rate is actually under 30%, and that's been pretty consistent for the last few years in terms of, speed to publication, which is a really important issue for us and for authors. 21 days to first decision, just under 300,000 full text views last year. just under 400 submissions.  

We published around 120 articles and we're a hybrid open access journal. Some articles are behind a paywall. We have individual subscriptions, but we're also obviously included in the call agreement. And, just over 50% of our articles that we published last year were open access.  

In terms of this year, applications acceptance rate is holding pretty steady around 27%. It's dropped a little bit. days to first decision is pretty steady. Full text views we're seeing an increase and submissions we're seeing a significant increase. So compared to this time last year, we've had a 40% increase in submissions to the journal. S

That's probably enough on that. In terms of peer reviewing for the journal, we operate a double anonymous peer review system for most article types. There are a few article types. our reflections articles led us to the editor and editorials and so on that aren't peer reviewed, but most are double anonymous peer reviewed.  

As Renaud highlighted, this is really important for us in terms of our reputation.  We are very conscious that we operate in a very wild west world of predatory publishing, and the reputation of our journal. And the society members and the publisher very much rests on a rigorous and, well executed peer review process.  

It's especially important for our constituent communities. And what I mean by that is the members of our societies, so the members of the AAG and the ANZSGM, that's around 3000 members, they take a very keen interest in the journal and how it's operating and how it's performing. And the peer review process is really important for them to have confidence that we're a reputable publication. 

We rely on reviewers as specialist experts. So, we are not just going through the motions and getting these reviews. We really need specialist expertise feedback on these articles because as I said, it's multidisciplinary. We can get very sophisticated medical articles, we can get very complex cultural studies articles, and we need specialist experts in those fields.  

And neither myself nor many of our associate editors are experts in those fields. So, we really do rely on peer reviewers very strongly. But the challenge, as all editors would know, and I'm sure Nigel will agree with me, is it's so difficult to recruit reviewers and an enormous amount of energy and work actually goes into that process. So last year we issued,  

1800, review invitations and we had 387 reviews completed. So that's a 35% success rate, and the median review duration last year was, 55 days. 

We do tend to have a bit more success with Australian reviewers. So the success with Australian reviewers was 35% last year, with us. I'm not sure if that's the case with every Wiley Journal, but reviewers get two weeks to complete. they get reminders, they’re auto declined after two weeks, and that's just to keep the whole process moving because every reviewer non-response is a delay to publication and slows down the whole publication process.   

Some papers we approach 30 plus people, to do reviews, and it can be really, really difficult for us. Some of the challenges that we are facing in terms of the peer review process is that it is a slow and resource intensive activity for us. We have incredible difficulties sourcing appropriately qualified reviewers.  

We get articles submitted from all over the world. nearly half of our submissions this year are from China and we need to have experts in Chinese Research to make sure that what we are publishing is actually legitimate research and meaningful research in China. That it's not fringe stuff, that it's actually meaningful transformative research. Some people feel there may be minimal benefits for peer reviewing, although obviously we do try to promote the benefits of peer reviewing in terms of collegiality and the other benefits that come with peer reviewing. 

But we're very conscious, of course, that all peer reviewers get bombarded or many of them get bombarded with requests to peer review.  

And I do, and I'm sure most of the people in the audience today do too. So why should you choose to review that article with us as opposed to the other 20 requests for review that you've had this year? 

I mentioned editor and reviewer expertise. We do rely on the expertise of peer reviewers very strongly. Often, it's hard to recruit the people with the right expertise for the right paper and as simply getting the right match between the reviewer, their expertise, the focus of the article and the methods that have been used can be very challenging at times. 

Reviewer delays and implications for time to publication.  

Obviously, that's a big concern for us. We want to get the research out as quickly as possible. Authors want to get their research out, they want to get their work out into the field to get it cited, to get it being used, and the delays in the peer review process are just a barrier to that. There can be an issue around quality of peer reviews, so they can be very variable.  

Sometimes we get fantastic reviews, sometimes they're very, marginal and I was talking to colleagues the other day. Just remember people see these reviews. They may be private in the sense that they're not released to the public, at least not in our journal, but the editors see these reviews and your reputation goes along with the reviews that you complete. 

And so it's just about important to build that quality and academic integrity into the reviews that people are completing. there's a big issue for us in the journal around identifying, supporting and engaging new reviewers, including ECRs. we have a bit of a systems gap at the moment between people who say they'd like to do reviews for us and then being invited to actually do reviews.   

The two systems are not really talking to each other, and we do have a plan in place to develop some more boutique resources for new peer reviewers for the journal. the other thing I'd just say is the systems can be very clunky. The emails they go out can be very messy. we are still in scholar one and I just feel that some of the communications are pretty unprofessional. Sorry, but you know, I just think we need to be communicating with our colleagues in professional ways, using their titles appropriately, and not having too much junk in emails and so on. 

Another key challenge is identifying AI. With the journal at the moment, we do obviously subscribe to Wiley's, guidelines around peer review. And, authors need to be able to, are required to disclose the use of AI in any form. but for peer reviewers, it can be challenging for them to actually identify whether or not AI's been used.  

I'm not an expert, but I did write an editorial about AI and the AGA the other day. So this is that. You can see it online. It wasn't, part of it was focused around peer review, but not all of it. Some of the general key messages, I think in that editorial are basically the ubiquity of  

AI at the moment. That we cannot escape it. It's a reality. We have to work with it. 

We set out some acceptable uses for authors, for example, in relation to grammar, citation management, data analysis, modelling. But of course, this has to be disclosed and often it needs to be explained in the methods section. Author voice needs to be maintained, accuracy has to be checked, copyright must be protected, and we recommend people keeping a record of their search queries as evidence.  

So in terms of AI and peer reviewing for us, I think the core principles in terms of the use of AI and academic work generally apply in peer reviewing as well. We expect AI tools to be used responsibly and transparently. They should be used in a way that ensures accuracy or accuracy is checked and that the author voice and expertise is protected, that intellectual property is protected and that ethical and professional integrity is upheld.  

We are, as I mentioned, transitioning to research, exchange, screening and review.  

And there are a number of ways in which Rex uses AI that'll hopefully, Nigel's going to disabuse me of this. I think his journals already transitioned to Rex, but hopefully actually relieve some of the pressure points that we're experiencing at the moment. 

For authors, hopefully Rex will simplify manuscript submission. AI extracts key information from material that authors submit and then populates it in a way that's acceptable for the journal saving all the time. It assesses scope of the article and provides suggestions. for alternative journals that might be more suitable for an article that's been submitted.   It undertakes some integrity screening.  

For example, it has its own AI check. It checks author verification across with orchid checks for fake references, plagiarism and paper mill screening. and it also operates a publication knowledge graph, which collects data from public repositories such as web of science, PubMed, and orchid, and builds an author keyword profile based upon the person's entire publishing history and maps the relationship between authors, publications, and keywords. 

And that's supposed to be used to get a more perfect match between an article and a reviewer. And it also takes into consideration, how many reviews people have conducted, how many reviews they've currently got with them, time to reviews, rejections, and so on, it can also flag conflicts of interests between the authors of a paper and potential reviewers.  

Sorry, I'm not going to show this. It was a little, demonstration of research exchange, but I won't go into that. So just some open questions that I have in relation to peer review and AI.  

The first question I think is for peer reviewers themselves, so do peer reviewers and indeed editors. Do we understand what is acceptable use of AI by authors? 

And I'm not sure we do, to be honest, and I'm not sure our peer reviewers know what's an acceptable and appropriate use of AI. 

And what the range of AI tools might be available to authors, for example, in conducting data analysis or modelling or whatever. 

Understanding what's allowable, understanding what's appropriate? 

Thanks Julia.  

Big question. Can reviewers submit manuscripts to Gen AI? 

No. Basically Wiley policy? No. ARC policy? No. NH and MRC policy? No. 

You cannot release manuscripts into the public domain to help you write a peer review.  

People have asked us can they do that to check if AI has been used in an article and to generate advice or notes or a written review in itself?

But can review reviewers actually use AI to assist them in completing their review? 

Well. Yes, potentially. Following the same principles in terms of author integrity, reviewer integrity, you might use AI to run a general search query on a topic you don’t know much about, to help you better understand it. 

You might use AI to assist with language, grammar, written expression in the framing of your peer review. There might be some acceptable ways in which peer reviewers might be using AI in their work. but definitely not submitting articles into chat GPT. 

And just finally. Yeah. I put this into chat, GPT and asked chat GPT what do you think AI could be doing in the future to assist peer review? 

And what it said was quite interesting. So it said, helping format articles and look, I think that would be great.   

Formatting articles, it's a bug bear for authors to format articles and to get them into the way a journal wants, especially if you've just had your article rejected by another journal and don't want to submit it somewhere else. 

It could help ensure compliance with guidelines, so potentially, helping the editorial and review process by making sure articles are compliant with our requirements.  

It could potentially compare original and revise manuscripts. People submit an article, they get reviews, they have to go and revise it, and sometimes it's not always clear what the difference is between the two articles, so potentially could be used for that. 

Checking if the reviewer's comments have all been addressed. it suggested an AI bot as a peer coach for reviewers. If you trained a bot really well, feeding it the right information, the right knowledge, it might be a good sounding board to assist peer reviewers completing their work. 

But three areas, which is pretty challenging. checking the novelty of the submission with the literature, that's going to mean submitting it into the public domain. So probably not from my perspective at this point in time. Methodological checks, we struggle analysing to  

determine whether or not the appropriate statistics have been used. People all have their learning curve in terms of stats, and that can be a big challenge for us. 

So there might be some potential there, but again, it's problematic if it's releasing it into the public domain. Summarising papers for reviewers, we've talked about that. 

That's definitely a no-no at the moment. predicting reviewer timelines based upon past performance. Yeah, I mean, maybe that's something useful that could be done. And maybe, it's suggested a transparent peer review audit across the whole journal.   

To actually determine the efficiency of peer review and so on.  So I thought, yeah, maybe there's some potential in that. Anyway, I'm at time, so thank you for letting me take a little deep dive into the Australasian Journal and Aging. I don't have the answers, but hopefully, highlighted some of the practical, things that we're dealing with in the journal.   

Renaud Joannes-Boyau: Thank you so much, Mark, for this great presentation. I'm going to ask the panel to come.  

Okay. Thank you. thank you very much. Okay. I've got some question for you. Easy  

question. Very easy question. we're going to start with the broad one, and we're going to start with you, Mark. 

What are the main risks of using AI in peer review? 

Right? We, you've talked about it, but can you just go a little bit deeper on that? 

Mark Hughes: I think, the key risk, I think is the temptation for people to submit an article into chat GPT and to generate, or some other, large language learning model thingy. and to generate a review based upon that. 

I think that is very tempting for people. When you want to do your best, you want to help the journal out, but you're feeling pressured.  

It's a complex paper, you feel you need help, it would be so easy to do that. but I think, Clare, you can speak more to this, but from my perspective, that is a big risk.

Clare Thorpe: What Mark's referring to there is, intellectual property. And the fact that really you shouldn't be putting anything into a large language model that you do not own the intellectual property or the copyright for.   

So as a reviewer, I'm receiving a submission from someone. They retain the intellectual property and copyright of that work. I don't have permission to put that into a large language model. I think, you know, that is a huge risk. 

When I was thinking about this particular question, I was thinking about how these tools, are seen as automation and augmentation tools. I think there's opportunities and risks around automating processes and Mark's spoken to a number of those editorial processes that could be, automated and sped up, in order to reduce the time from submission to publication, which is what every author wants. 

But there's also thinking about the tools in terms of how they augment the work that we do. 

And I think that's where the grey area comes in. And I think the whole thing for me about peer review is remembering the human. And so one of the risks I think for people in the process is around stigmatise of AI use and people feeling that they can't be transparent about the use of it and moving into this kind of shadow use space. 

And if we don't de-stigmatise it, if we don't have conversations like this and put those guardrails in place and, and coach people around when you can use it and when you can't and how you can, then I think we're actually going to create this environment where people will use it but won't declare it.  

And that erodes the whole trust of peer review. 

Renaud Joannes-Boyau: Yeah, that's a very good point and that's a really good, segue to the next question actually and that's going to be for Jo, so do you see any aspects of the peer review processes that must remain uniquely human?  

Joanne Bradbury: Thanks for that question.  Great. It's a great question but I think it's ultimately human judgment.  You know, we have to retain, human judgment, especially around ethical evaluations and also discipline specific expertise. These are things that are unique to humans and that the current crop of gen AI, it's beyond the scope of their practice.   

They can't do critical reviews. They're not, trained to do that. They're probability models that are looking on the basis of probability, what's the next likely outcome. Given these prompts and this background information, they're not concerned if they're right or wrong. 

They're not concerned about just or unjust or good or bad. They're just concerned about being plausible in their outcomes. 

And it's something that's still thankfully unique to humans to be able to make that human ethical judgment. And it is really important in ethics as well, because these bots can't feel, they can simulate emotions and empathy, compassion, but they can't, they haven't had the lived experience of being a human in the context of the world.   

They don't know what it's like to experience suffering. They can't really apply cultural capabilities and nuances that are just so unique to humans.  

And so that those roles of making critical evaluations, especially around ethical decisions, should always remain with the humans. 

Renaud Joannes-Boyau: Yeah, that's very interesting. And yes, I would say not yet, but maybe we'll have artificial emotion, and, perhaps that's the next stage of AI, so if we go back to the journal and I'll go for you. I'll ask you Nigel on that. 

So how do you see journals and institutions safeguarding this research integrity given AI's limitation in detecting AI generated content?  

Nigel Andrew: I guess a lot of the time it's difficult because we're in a space where the systems are going through exponential change. At the moment, and in a scheme of things where from an editor's point of view, a lot of the time the peer review system is seen as broken. Like we are talking about a peer review system where there are so many problems with it because of the, the sheer amount of information being put out there for peer review. 

As we saw before, there's not necessarily enough people to do the peer review. And when you get peer reviews back, they're not all of a wonderful quality. There are qualities of peer review that go from being fantastic to one's being fairly mediocre, to one's being less than ideal.   

And you also have the experience where we have, it's southern Hemisphere based, so we actually have English as second language author. We also have English and second language peer reviewers. And the quality of the peer review system that comes in and goes out is variable. 

There's no perfect system. And it is difficult to identify it as a broad suite. What a lot of times we do is essentially, having authors at least identify how they use it and actually have a clear and open statement. 

Because AI is a huge beast. It's a bit like saying I use a computer for my research. Which part are you using? You use it for doing literature reviews, you're doing it for analysis, you're doing it for interpretation. AI has very explicit parts to it now, and we are slowly maturing in the conversation as well. And it makes it difficult to identify what's happening because it's been in the process of changing our systems really only in the last few years. 

There are major changes happening. We are trying to catch up with it. We're in this very much transition phase where it is difficult to make interpretations. 

Renaud Joannes-Boyau: I think you're right, disclosure, transparency, that's the key aspects. And every line is very blurry when it comes to AI and sometimes it looks like it's unacceptable to use it in a certain way, but actually if you look at the way it's used, if it's disclosed, then it becomes acceptable.  

There's a complexity and I think the transparency discussion around it is very important in that way. Clare, I've got one actually for you. 

Can AI enhance reviewer integrity and transparency?  

Clare Thorpe: I'm going to sit on the fence and say it depends. 

Renaud Joannes-Boyau: I think it's going to be a comment, right? 

Clare Thorpe: Yeah, I think it depends. I actually asked one of my mentors this question, or a similar question around the use of this, and what they said to me is, reviewing a paper is entering into a type of relationship or partnership with the author.  

And this is where I, I keep coming back to this idea that it's in the name peer review. We haven't reached a stage yet where I have seen any journal or any reviewing body say that they will accept an AI tool as an author, and therefore if they can't be an author, how can they appear in order to review?  

I think as you've alluded to, Rena, we might get there one day, but in 2025 we're not there yet. I think where AI could potentially support research integrity would be in automating the processes and the verification of reviews by humans. If we think about how technology issues say in non-fungible tokens as a piece of metadata, in order to build trust in things like NFE artworks and crypto cryptocurrencies and that sort of thing.   

We're using the technology in order to verify and track and audit and have these trails. 

I can see a world where AI tools are used in order to provide verification and to really underpin trust and integrity in order to have a robust review process. However, I think that's a little way off yet.   

Renaud Joannes-Boyau: That's fair enough. I think that's a very good answer. I've got another one. 

We have other places, all the funding bodies and publishers. You have the, Australian Research Council and NHMRC. They're prohibiting the AI use in application and peer review.

They're very strict about it. I'm not sure how they enforce that, to be honest. Obviously, we talked about it, we don't want to upload data. There is a problem with privacy ethics to the AI, but should we look into that? A standardised kind of guideline for all those kind of things in terms of application for grants, application paper, peer review, and so on. 

Nigel Andrew: I think again, it's really difficult because at the one element, the ARC and, and NHMRC are giving federal government money to researchers and to universities to carry out research, whereas at the other end, and so they might give it out to 10 to 12, 13% of researchers. 

There's another 80, 87% also of academics who don't fit into that pool. And on the other end, there is research being published. And I think having a one size fits all approach is problematic.  I mean, for example, if you are working in information technology doing gen AI work using agent development. That obviously won't work. 

There's a great example where, you know, restricting AI usage would not work when you're working with AI. And in the same sense where you are working in a publication. one of the examples where we are really trying to understand the impacts of AI is when we do have English as a second language authors from Brazil and Argentina, they're actually writing their manuscripts in their native language.   

They're getting that to a perfect level, and then they're putting that work into chat GTP to actually translate it into English. Now to my mind they're putting their work into that to make sure it is at the level where it is in the language of science, which at the moment is in English. 

If we didn't allow them to do that, they would be completely ostracised from being in high quality research work. We are actually in the middle of trying to make sure if they do that, because when the publication goes into our research systems, it gets flagged as being Gen AI-written. We have to take that into account.  

Renaud Joannes-Boyau: I touched on that when I was reflecting on it. The equity also, the, the fact that AI opens to a lot more people because it's been very English focused, in terms of all the literature published outside. And I think it's good that we have the ability to translate very quickly and this is an efficiency, this is a great tool with AI. 

I agree. Jo, if you could change one aspect of the current peer review process or system, using AI responsibly, what would it be? 

Joanne Bradbury: Well, if I had a wish list of what I could ask for, and if we could do it responsibly, I'd think that we could help improve our ethics, review processes. If we had a tool that was privacy assured and that wasn't sitting somewhere, we knew what was happening to the privacy that we could make available to the authors, that they  could get a quick check of all of their documents prior to submission to ethics,  and it could just check that all the documents that are required are there, and that there's consistency between the documents that, and then they just get a quick checklist of what they have to do to improve and then they go and submit. 

And that's just an optional thing for those authors that are on a speedy timeline, for example, that could save weeks of the peer review process that goes on in ethics because we end up as, as reviewers for ethics, picking up a lot of those inconsistencies across documents that are not really approvable in the current state. And that could be automated.   So that would be my wish list. 

Renaud Joannes-Boyau: Technically this is possible. You just have to have these completely isolated, like a sandbox. AI and that way the information that goes into the sandbox, stay in the sandbox. 

There's a possibility to do that. It's just isolation.  It means also creating your own model, which is not that easy. but in the future AI model will be probably able to help you create your own model.   

Joanne Bradbury: Oh, that's great, isn't it? 

Renaud Joannes-Boyau: I think it's a very near future for that. I've got one last question is for all of you. 

What is one principle in your mind that must never be compromised no matter how advanced AI becomes when it comes to peer review and the submission process?  

Who wants to take that one? 

Mark Hughes: I think that point about human judgment is needed in essence. However, we are using AI to assist us and to improve things, it needs to be submitted to scrutiny in terms of accuracy, integrity, in terms of the authorship, inclusivity, in terms of non-discriminatory crap that it generates.  

And just ensuring that there's human judgment in ensuring the integrity of the content. 

Renaud Joannes-Boyau: Can I ask you to follow up on that?  Isn't human also bias? They make mistakes when they make judgments. I can hear people that are pro-using AI in that sense saying, well, they will make mistakes, but those mistakes will be kind of fair.  

They won't have any bias and so on. What do you say to that? Because humans are unreliable, they make mistakes.  

Mark Hughes: Humans are accountable as well. that's that. And we can be held to account. I'm not sure if AI bots can.

Renaud Joannes-Boyau: Not right now. No, that's correct. Yeah. 

Joanne Bradbury: I would add moral responsibility. We can't at the moment, just delegate that off. And especially in part of the peer review process as well. A peer reviewer has engaged a contract to actually take responsibility for the review. And so, they need to be accountable, but I think also in a moral sense. About fairness, and I don't know if AI will ever get to that point where they can actually make a decision about moral responsibility or take moral responsibility.   

Renaud Joannes-Boyau: They actually might make it, but we might not like it. That's a problem. 

Joanne Bradbury: Might not like it at all. No, it's clear. 

Clare Thorpe: Yeah, for me, it comes back to the human rule and that these are tools that can be used, but human oversight should always be the last step.  

You know, we've all got a review from reviewer two. Why is it always reviewer two? That's the grumpy one. Maybe AI will make reviewer two sound a little bit nicer, but AI is only a mirror. It doesn't have that ability to really engage with the critical nuances of the research expertise that as researchers you all have developed over time and peer review and publication, is very much trust-based.  

If we want to uphold the integrity and the social responsibility and respect for research and science, we need to make sure the processes are trustworthy. And the only way we do that is through having human oversight regardless of the flaws of peer review that Nigel has mentioned as well.  

Nigel Andrew: Yeah, I guess in some ways AI, as a disruptive technology, could potentially break peer review completely and it could actually break the concept of actually what we put in a publication in this idea of original ideas. I guess I always go back to the idea of every original idea is 75% recycled. 

Everything that we put out into the world anyway as an independent publication is fundamentally based on the ideas of two, 300 years of people before us. 

It could, and we've seen with COVID the rise of putting out information very quickly and being very disruptive.  

In some ways the idea of the AI broadly is enhancing, potentially breaking the peer review process and also breaking the publication process as well. 

We actually might be in a very different realm in five years’ time in terms of what's considered an original idea and original output.   

Renaud Joannes-Boyau: That's a very good point and I'll make a final comment before we open to the questions that we had received. I'm actually a firm believer in the publication itself. This is a legacy at the time where researchers were isolated and they had to write that manuscript, send it, and it was very difficult to access. 

So they would regroup into a journal that they would send to scientists. 

And this legacy is outdated in the time where access to communication and AI is completely changing the way we do science. 

And while I am not talking about peer review. But the publication itself has an article format that is grouped into a journal in a day where we can have very different aspects to communicating science, communicating our results, showing the data, more transparency is needed. 

I think this legacy format is outdated and should be changed, but that's my personal view. 

Clare Thorpe: I think it is related to peer review because the container may change, but what's contained, the body of knowledge still needs to be trustworthy and be underpinned by integrity. It really doesn't matter if we're talking about the publication process or we're talking about peer review within that, if research that is made publicly available is not seen to be trustworthy by the community at large, we're out of a job. 

Yeah. All of us. And, and that is a problem when they started this new format where the publication will go live and the peer review will be open to people to log in, and then make comments. And that was a big problem in my field. 

And you've probably seen a documentary in Netflix called The Cave of Bones, which one of the publications was based on. And that publication is extremely wrong in many, many aspects, scientifically. And everybody started to comment on that, but the publication was already out and already damaging the reputation of the discipline.   

So anyway, we are going to stop there and I think Julia has some questions for us. 

Julia McConnochie: We had quite an overwhelming response of questions that were submitted online. So, to make sure that we have time for everyone here in the room to participate in the discussion as well, we've just chosen a few of those questions that were submitted in advance for the panel. 

I'll start with a question for our two editors here today. So Nigel and Mark, as an editor, how do you think you can recognise AI generated review reports? Are there any signifiers or things that stick out to you, that help you detect where AI has been used in those reports?  

Nigel Andrew: Unfortunately, it's the M dash. I love the M dash. I'm so disappointed that I can't use the M dash anymore in our communications. If it's just done as a baseline, put through Copilot or chat GPT in a very basic format, there's a language that comes out of an AI bot that is fairly generic. 

But if someone has developed an agent and they've actually developed it in a fairly sophisticated manner, they've actually given an agent particular instructions, they've given it a voice, if you've given the agent some of your work previously and said, can you actually write a review with my voice on this publication? That once you actually teach the agent effectively to become a bit like a second brain, you actually engage it to be part of your online personality. That's much more difficult to identify. 

But fundamentally it comes down to people acknowledging it. It comes down to if you say to a peer review, a reviewer or an author, have you used any AI and what have you used? 

If they say no, then you know, it's the whole peer review process is based on openness and transparency.  

When people fundamentally cheat the system, you know, authors do that. We've got retraction watch, which we see plenty of people trying to cheat the system. The whole system is based on trust.  

Mark Hughes: I'm pretty useless at detecting AI generated content. I know a lot of people are good at it, but I'm not really one of them. I don't think we've seen any AI generated reviews in our journal. We've definitely seen. AI generated articles and we've rejected some of them, like they were generated entirely, and we could detect that. but most of the reviews I see are pretty scrappy, to be honest. 

And if it was AI generated, I would've expected it to be more polished. We get a lot of spelling errors and busy people writing stuff really quickly. That's what we get. 

Julia McConnochie: Well thanks for that. We have one more. I think we've spoken about it a little bit throughout the conversation. but this one might be one for Jo, just around the ethical implications of using AI in peer review. 

Joanne Bradbury: Oh, thanks Julia. I think most of what we have been discussing does touch on those, ethical implications and for example, the issue of confidentiality.  

When there is ethical implications for the author when, lack of not respecting the author's, intellectual property and, and right to copyright.  

And that could even be a justice issue as well. But I think there's also a risk of harm to the reviewer, if a reviewer does upload a confidential manuscript to a public, gen AI tool, there's that very real risk of breach of confidentiality and breach of the code of responsible research conduct. And, it's a researcher integrity issue, which could have some negative consequences for the reviewer and also what I was mentioning before about using the tools out of scope. 

The tools aren't designed to do critical evaluations. But even that delegation of responsibility, once you're engaged to do a review, you have to do the review and that's part of the policy as well.   

You can't delegate that responsibility to somebody else or even ask for help without the authorisation of the journal. There's a research integrity issue, and that has ethical implications as well for practice. I think the short answer is, it does depend on how it's used but it's an ethical quagmire the whole area.  

And it's really interesting to talk and discuss them. And more and more of them come to light. 

Julia McConnochie: Thanks Jo. We've just got one final one to wrap up with the questions that were submitted in advance and then we'll open up to the floor for audience questions. 

The research environment is changing at a pace that is difficult to keep up with as a PhD student. What are your tips for going forward as a researcher in these new times? 

There a broad one there, but maybe a response from Clare on that one.  

Clare Thorpe: Look, change is constant. AI has obviously changed things in a particular way. but I think rather than focusing on change, focus on what's the same. so for people who are very early in their career or as in doing their PhD, I would be saying, do the best research possible with the tools that are available to you and be really strategic about, developing a publishing strategy to make your work accessible, discoverable, and impactful. 

I think, the research journey can feel quite isolating sometimes, but there is a community around you here at the university, through your supervisor, through your faculty, through the support networks like the folk who work with me in the library and the research office. 

It is about tapping into those people. But really, we've said the whole way through this conversation that we create new knowledge, we create research with the expectation that that research will be used and it will be trusted. 

So if you keep that as your guiding light and seek advice, as we've said, we're all learning as we go when we are talking about the AI technologies, but we are in it together. We're not, not working alone. 

Renaud Joannes-Boyau: Yeah. I'll just add a very brief thing as well. I think it's very important that we also provide our PhD students, knowledge of AI because they're going to need it for the future to be competitive. 

I think the more knowledge you have about AI, the more likely you are to use it well as a tool and also responsibly. I think the biggest risk of breach and misconduct is when people don't know and say, I was not aware of it. 

Or they don't use the tool well because they don't have the knowledge. The more knowledge we build with our PhD students and then our researchers, the less likely we are to see problems in the future. 

Clare Thorpe: Yeah, it comes to not attaching stigma to the use of these things. And where there are really good taxonomies or guidance or framework that we can use is making those known and adopting those where possible. 

Really encouraging people to have those open conversations and to learn and experiment together within the guardrails that we can provide.  

Renaud Joannes-Boyau: That's very true. 

Now we're going to open the floor. Raise your hand and someone will bring you a microphone. 

Audience: The Journal of Aging. You said something very interesting that you had received papers written by AI. Give me a feel for what the hell they did. It can't be about laboratory work or was it just social waffling. What would an AI bot write? 

Mark Hughes: Yeah. I got one last week and we rejected it and I just can't remember the topic, but the problem was, it didn't have a methodology, it didn't have finding.  

It was a very generic, descriptive piece around a topic. it was beautifully written, but I could not get to grips with what the substance was. And I didn't put it into an AI checker. We don't do that. But it was my impression if I've seen an AI generated article. I think that was it.  And I'm sorry, I can't remember the actual topic of it at the moment.  

Nigel Andrew: In Austral Ecology, I probably get about one nearly every week or two now AI generated. They're basically primarily review papers, so you can easily put in unsolicited review papers that are on really very broad topics like the impacts of climate change on the environment. 

Thank you so much everyone. 

AI is a one of the biggest revolutions that we're going to have in our lifetime, that's for sure. And, we need to, in our profession, in our everyday life, in a way we interact with other people in every way, AI is going to change a lot of things. I think we should keep on having this discussion.

Thank you. Thank you so much everyone.