Cadmore media player playing video PS26: Welcome
Video Player
Accessibility
Using forms mode, you can utilize keyboard shortcuts to make usability more efficient.
Space bar will play and pause the video. Pressing escape will close all displays and pressing it again will jump to the end of the player.
You can press A to open the Keyboard Shortcut Dialog which also provides an audio description of the player and all of its features.
Language
1x
Play Speed
Adjust Volume
Transcript
Search
Result Count
of
Click to jump to result
Search
Re-center
Segments
Result Count
of
Click to jump to result
Search
Name:
PS26: Welcome
Description:
PS26: Welcome
Thumbnail URL:
https://cadmoremediastorage.blob.core.windows.net/d2a28e2c-3182-4d9d-801e-612007f13e0c/thumbnails/d2a28e2c-3182-4d9d-801e-612007f13e0c.png
Duration:
T00H25M24S
Embed URL:
https://stream.cadmore.media/player/d2a28e2c-3182-4d9d-801e-612007f13e0c
Content URL:
https://cadmoreoriginalmedia.blob.core.windows.net/d2a28e2c-3182-4d9d-801e-612007f13e0c/PS26_WelcomeAddress.mp4?sv=2019-02-02&sr=c&sig=%2Bmvqq5Q4U8SkYoae4NSzeE3kuudrS33UwSRD46ufdSw%3D&st=2026-10-08T22%3A32%3A07Z&se=2026-10-09T00%3A37%3A07Z&sp=r
Upload Date:
2026-10-05T00:00:00.0000000
Transcript:
Language: EN.
Segment:0 .
STEPHANIE LOVEGROVE
HANSEN: Welcome, everyone,
HANSEN: to Platform Strategies 2026. Or as my colleague, Sam Green, called it when she saw the agenda, Platform Strategies, the Reckoning. So we're in for some good discussions today. My name is Stephanie Lovegrove Hansen. I'm the VP of Marketing at Silverchair, and it's delightful to have you all with us today. I'll see a lot of returning faces. So you know that we're in for a very exciting day.
HANSEN: And for those that are joining for the first time, welcome. Pro tip, don't miss the swag table. So throughout the day, your primary source of information is going to be the event app, which hopefully you've downloaded. There you can see the agenda. You can ask questions if you don't want to do it IRL. And you can also connect with each other and see who else is here.
HANSEN: So if you have not yet downloaded that, you may do that. There's some instructions at the registration desk. So today's sessions are being recorded, and will be made publicly available to you after the event. And returning attendees know that we always like to do something a little bit fun. So we've had live sketch artists.
HANSEN: We had a very creepy AI photo booth at some point. And this year, however, we're really stacking the deck. We are not holding our cards close to our chest, but instead dealing you in with custom playing cards that represent different stages of the research life cycle. So please take a minute to flip through those. But that's not all. The ace up our sleeve, as they also come with custom scholarly publishing games that are within each deck.
HANSEN: And you can also, if you play your cards right, get your photo taken as the wild card of your choice. So I have now run out of card puns. So I will take a minute to then thank our sponsors for this year's event. Digital Science, our charging station sponsor, which you can find in the back of the room if you need that throughout the day. Impelsys, our beverage station sponsor.
HANSEN: ADIV, our app sponsor. DCL, our morning break sponsor. Access Innovations, our afternoon break sponsor. And finally, Hum, our breakfast sponsor. So thank you for supporting this meeting. And now I will hand it over to Will Schweitzer. [APPLAUSE]
WILL SCHWEITZER: I didn't write nearly enough dad jokes. Very hard to follow. I hope after four years, I've learned how to use the microphone. If you can't hear me in the back wave, yell, scream. I'm Will Schweitzer, I'm Silverchair's CEO. I'm really glad that you've joined us. Over the next 15 minutes, I'm going to try to preview all of today's sessions by zooming out, and providing some commercial market and technology context.
WILL SCHWEITZER: As always, when I do these talks, there will be a few iCharts. There will be some salty remarks, trying to set the tone for the day. And I want to start with this one. The market has repriced publishing, and our readers stopped being human. So I hope you've had your coffee. Nothing really happened on February 3.
WILL SCHWEITZER: In fact, I woke up thinking about how Bad Bunny couldn't perform at the Grammys. And now Trevor Noah lives rent-free in my head as he was trying to sing "DtMF." Anthropic shipped a legal plugin for Claude. And what happens then was an information company and SaaS-pocalypse, something you've probably seen in the news every day since. But when we think about the publishing companies in our space and the value that they lost, none of them lost a customer.
WILL SCHWEITZER: Not one product failed, not one journal folded. No one lost essentially shelf space in the market. And ever since then, ever since there was the largest decline in Elsevier's stock price, in RELX stock price in over a decade, or a substantial drop in Thomson Reuters stock price have I been able to make any sense of the market? And to the credit of our friends at Clarke and Esposito for their ongoing analysis about this, the companies with the most AI announcements in our space, the Elsevier, the Thomson Reuters, the Wolters Kluwer have lost between a fifth and 2/5 of their market value over the past 12 months, despite the fact that they've all posted strong revenue growth.
WILL SCHWEITZER: So the market is no longer scoring product announcements. The market is scoring what they see as the durability or the lasting value of what all of us in this room do. So fortunately for us, we have Wendy Queen, the chief transformation officer of Johns Hopkins University press. Reflecting on what we've collectively learned in the past 12 months, and then maybe she can help us make sense of stock prices.
WILL SCHWEITZER: There's another analog out there, and I am really hoping that we are not sad Pepsi. Three years ago, in one Bloomberg interview, there was a Walmart executive who said because of GLP-1 drugs, like Ozempic, customers were starting to put less snack foods in their carts. And within one week, every single company that has something in the snack aisle was repriced. They collectively lost about a quarter of their value in one week alone.
WILL SCHWEITZER: A majority of those companies have continued to grow, just like Elsevier, just like Wiley, just like companies in our space, their stock value has never recovered. And there's an interesting contrast here in Coca-Cola and Pepsi. And I really do hope that we are not sad Pepsi in our market. In that Coca-Cola, their stock prices come back up. They have a slightly different business strategy, but fundamentally, the market believes in Coke and has more confidence in their long term strategy than they do Pepsi.
WILL SCHWEITZER: And a lot of that is just about sentiment and mood. This is a bit of a tangent, and I want to try to bring this back with the confusing headlines mean for all of us. The global AI build out. When we read about the hyperscalers, the companies like Amazon or Oracle, building data centers here in Northern Virginia or elsewhere across the country.
WILL SCHWEITZER: All of it is being done with borrowed money, and is being done with borrowed money at an incredible scale. So last year, we talked about the amount of corporate R&D being spent on AI, and how that was going to transform our daily existence. It was going to transform our products and our market. This is particularly troubling. The amount of debt raised this year is 121 billion. That's just on the books.
WILL SCHWEITZER: Off the books, there are $662 billion of financing that isn't shown on balance sheets. That isn't transparent to the market. And what is even more troubling is the bets against this are now being done with a credit swap options, which are the same things that led to the mortgage collapse in the early aughts. That is never a good sign for the market. So what does that mean for all of us in the room?
WILL SCHWEITZER: It means our cost of borrowing money are going to be higher. It means they're going to continue to be confusing headlines. And my best suggestion is this, tune out the day-to-day noise, and focusing on the long-term pragmatic products, pragmatic developments. Leveraging these tools in your company is about all that you can do, because the volatility is here to stay.
WILL SCHWEITZER: That was a slight tangent. So thank you for indulging me. Following Wendy's keynote, my good friend and spiritual counterpart, Dustin Smith, the CEO of Hum will moderate a panel with Dawn Melley from IEEE, Miriam Maus from IOP, Jonathan Woahn from Cashmere on leveraging AI systems to ensure research integrity and quality from submission all the way to content discovery and synthesis.
WILL SCHWEITZER: And here's some data from ScholarOne to help set the stage. Desk rejects and ScholarOne grew 72% from '22 to '25, against 43% growth in decisions overall. Our publishers now perform 2 and 1/2 desk rejects for every acceptance. The cost of us collectively saying no is the fastest growing expense line in scholarly publishing, and not one of us has a way to recover that expense.
WILL SCHWEITZER: And on the reading and synthesis end, last week, the OECD released data from their PISA learning assessment, a global learning assessment, and it paints a really interesting picture. The headline is this, there is more text produced faster, read more hastily by readers who evaluative capacity is measurably weakening. And I say that without trying to be alarmist. If you haven't discovered Tina Austin and her unbloomed Substack, please go read it.
WILL SCHWEITZER: She is the director of AI education, licensing and public interest technology at UCLA. While she often writes about AI in the classroom, the findings she's drawling out can help us inform our product strategy, or help our colleagues leverage AI tools at work, or help us with all of our editorial stakeholders in the world. I think the kids will be OK.
WILL SCHWEITZER: Her headline coming out of those kind of dire PISA data snippets was that students who repeatedly engage with AI tooling, who are in advantaged classrooms where their instructors can provide the right framework for using AI tools do better than they would otherwise. But that says a lot about relative equity or advantage in the classroom, and how we are setting students up to succeed. Following Dustin's panel, we'll move on to table topics-- apologies to the introverts in the room in advance-- and we'll discuss how peer review is under pressure from every direction.
WILL SCHWEITZER: So here's a bit more context from ScholarOne. And I'm purposely using business terms here to see if it can help shift our thinking. The aggregate reviewer pool, the number of reviewers captured in the ScholarOne system has grown 54% since 2018. Invitations to review have more than doubled. Acceptances to review have halved. It takes 4 and 1/2 invitations to land one review, and per 100 invitations, an editor spends a whopping 407 days waiting for a response.
WILL SCHWEITZER: And that response, more often than is no. That isn't a shortage of reviewers. That is a dead inventory problem, and inventory problems tend to have business solutions. What's more important here is as publishers, we are not passive recipients. We can actually change our systems and processes. After the table topics, we'll break for lunch. Introverts like me can go recharge.
WILL SCHWEITZER: And then we'll come back to a panel discussion that I'm really excited about. My colleague Ryan Ross will talk with David Sampson from NEJM, Meagan Phelan from Science, Suze Kundu from Defend Research on the role of technology in the misinformation era. And that truth may really be a platform problem. I was sleeping really well at night until I read the preview for this session.
WILL SCHWEITZER: [LAUGHTER] Really well. So zooming out just a little bit. Pew Research watched 69,000 people perform Google searches. When an AI summary appears, the click through rate onto the underlying or the resulting article dropped from 15% to 8%. 1% of people click on a link that is in that AI summary. The replacement channel isn't arriving either.
WILL SCHWEITZER: Those are eyeballs that are lost. And outside of our industry, when we look at media web properties in general, ChatGPT, which is the most common consumer agent, is only 2/100 of a percent of publishers referring traffic. We are starting to adapt in our industry. Counter was consultant in February. They released an AI best practice in April, and it is a really good start.
WILL SCHWEITZER: The problem is there are a lot of things that counter, and this best practice still aren't counting. Counter is the proposed best practice is only kind of measuring agents that come in through the front door, and that kind of resolves to an article itself. What we haven't solved for yet is what happens when there is federated authentication? When there's easy proxy in the mix.
WILL SCHWEITZER: When you are using a tool that is working with your browser to then come and navigate to an article and perhaps scrape it or extract it. That is a lot of traffic that we can't see. So this is a good start, but there's a lot of work for us to do. I'm really concerned about the training corpus, and the worldview that is behind leading models, whether that is Claude, or ChatGPT, or Chinese open-weight models like Kimi.
WILL SCHWEITZER: When ChatGPT was new, researchers in Beijing tested it, and it thought Yao Ming was a woman. Their conclusion wasn't that the model didn't work. Their conclusion was that it had been trained on the wrong data. So there is now a large scale government-funded initiative within the Chinese government to fix that by the Chinese research apparatus becoming the largest supplier of training data. And that includes scientific research. A Princeton sociologist who studies this put it better than I ever can.
WILL SCHWEITZER: AI disconnects the messenger from the message. Keeping those two things attached. Provenance, attribution is the job of everyone in this room. So I don't know what your geopolitics are. I don't know your relative sense of ethics. But I think this gives us some urgency to the thought that our knowledge belongs in this broader and connected AI ecosystem. Following that discussion of truth, relative or otherwise, I'll be back on stage with some really good colleagues and friends to redo a panel discussion that happened on this stage two years ago on what is your strategy if there is no platform?
WILL SCHWEITZER: So I'll be joined by Alison Mudditt from PLOS. Andy McGregor from Sage, Ann Michael, David Crotty from Cold Spring Harbor Press. And just a little bit of data from the Silverchair platform. Between June and August. One publisher on our platform, only 41% of content requests scored as likely human. And the rate of likely human readers continues to fall every month.
WILL SCHWEITZER: When we look at providers who can provide a larger sampling of the internet, Cloudflare estimated that 52% of caller requests were for training. 53%, Imperva would say that 53% of internet usage is non-human or robots. The shape of the reader on the internet is changing fast. For 20 years, our policy of asking non-human readers, robots, was essentially a polite text file on our site.
WILL SCHWEITZER: There have been some recent kind of judicial law developments here. In December, a federal court described robot.txt accurately as a mere request not to scrape. OpenAI's own documentation says that when a user initiates an actions robot.txt may not apply. The opt-out era is over, and what replaces that is cryptography. And here's the part that interests us.
WILL SCHWEITZER: We may be able to prove who is asking, who is coming to our sites using a chat agent before we can tell them what they are allowed to do? So we may know their identity, but we may not be able to pass back to that user to their agents whether or not they can scrape content, whether or not they can access full text, whether or not a subscription is valid. So it is a interesting time for evolving technology here.
WILL SCHWEITZER: When we think about the products in our space and those chat agents, this agentic distribution channel exists. It is getting a lot of use. The default pathways, the ways end users using these agents can connect to corpuses in our market. They're being established now. And nearly none of you in the room are part of it. Two publishers in our space have shipped connectors into assistants like Claude and ChatGPT.
WILL SCHWEITZER: One of them is Wiley. None of them are independent or society publishers. And when Anthropic turns on Claude science for an end user, the entire scholarly literature layer is drawn on OpenAlex and arXiv, on open metadata and preprints. That means for all of the publishers in the room, your content sets are not part of the default answers. Reachability into these AI tools is now the new indexing, and the index is being built right now without the participation of a lot of us.
WILL SCHWEITZER: Last Thursday, there essentially was plaintiff evidence from the OpenAI New York Times lawsuit that was unsealed. Nothing's been decided yet. Replies are due in November. The US Justice Department has just weighed in, but the line that sticks with me is here on the left-hand side, which was ahead of ChatGPT within OpenAI saying their products are largely substitutive. And on the right-hand side, Satya Nadella, the CEO of Microsoft, was being disposed, and he was asked if he thought ChatGPT or other model builders should have to pay to train their models on paywalled content, and the New York Times in particular.
WILL SCHWEITZER: And his answer was yes. Two things for the room. This was largely a case concerning news content, and news content is a summary of something that happened. So if a summary is summarized, that's it. That's a very different construct from research where a summary of research may be enough for a fraction of use cases, but not for all of them. And access to the underlying data, the discussion around the paper.
WILL SCHWEITZER: The suggestion for what comes next. What should be studied next. Broader implications for the community or for medical practice, or whatever it may be, is our structural advantage. And we need to figure out how to hold that line. Another really interesting thing a lot of us have been thinking about training data, and whether or not we can sell our corpuses for training.
WILL SCHWEITZER: AI licensing is not a rounding error. It is not an item at all in the budget of folks that are building frontier models. Alphabet publishes their costs for traffic acquisitions. It's about $600 billion. It names content acquisition costs for YouTube, but declines to size them. For AI training data, it declares nothing at all because there's nothing material to disclose.
WILL SCHWEITZER: Microsoft's only mention of the subject, what it pays for content to train their resources is a risk factor. In 300 pages, they say it may one day have to pay for data.
SPEAKER: It's super easy to open a--
WILL SCHWEITZER: No, quite OK. When we think about the dollars that have flown in our space, large commercial publishers securing dollars for training. It amounts to 0 but a sugar rush. It is good for the moment. It is not a sustainable meal. The next session on the agenda, hello from the other side is moderated by my colleague David Nygren, and in talking about or sharing some data from Taylor and Francis on the previous slide.
WILL SCHWEITZER: I hope I have not pissed off one of our interviewees and one of our largest customers. Who is Penny Ladkin-Brand, Taylor and Francis CEO. David and Penny will be joined by Kelly Palmer from TELUS Digital, Ha-Hoa Hamano from the US Department of Education and formerly NPR. And the session is usually a crowd favorite. In it, we ask those panelists to bring context from previous industries they've worked in or from the industry they're in now, and talk about how it may apply to us in scholarly publishing.
WILL SCHWEITZER: Before Taylor and Francis, Penny was the chief financial and strategy officer for Future. And I would be really curious for her view on what showed up in Futures numbers first because traffic is probably a lagging indicator of something. I'm reaching the very end of the day. I'm really excited for our closing keynote. We'll be joined by Jacob Ward, who is a CNN contributor. He has reported for NBC, Al Jazeera, CBS.
WILL SCHWEITZER: He was the editor-in-chief of Popular Science. He's been on the news an awful lot. We have his book available at the registration table, and he's going to talk to us about what human skills must be protected when AI gets really good at everything else. The data here on this slide is actually looking at the cost of inference of using these models. We are operating in a world where you can now manufacture a plausible research paper using a GPU.
WILL SCHWEITZER: What you can't do with tokens is manufacture the warrant behind the research. The validation, the accountability, the author's name, the institutional affiliation on the byline. Everything around judgment is getting cheaper, but judgment is about to be the most expensive input to all of our products. And is what is becoming increasingly remarkable, particularly for you all who are using ChatGPT or Claude on a daily basis is that the older models are getting cheaper, and you can still do an awful lot with those older models that aren't the latest release of Fable or Opus 5, whatever we're up to now.
WILL SCHWEITZER: And we're all just learning how to cost optimize our use of those tools. There's still a lot of work for us to do there. On the left is a quote from Alan Garber, who's Harvard's president from the first day of classes this month. His address was about human intelligence being distinct and precious, and about defending the work of researchers itself.
WILL SCHWEITZER: He was suggesting that AI will allow researchers to move faster to push the evidentiary boundary. And on the right is a quote to OpenAI that I keep coming back to. Adoption of AI is being urged by institutions that consume scholarship on tools who builders privately described as being substitutive of coming to our journals, coming to our platform sites, interacting with our products. And I don't think either of them is lying or wrong.
WILL SCHWEITZER: Their incentives just simply don't match up, and we're caught in the middle of this. So last year, my invocation for everyone, my ask of everyone in this room was to have the audacity to not mope. That worked for about, I don't know, two weeks for me last year. So I'm going to take a slightly different approach this year. And that is, I think it could be very valuable if we express our fears, if we name them, and we get specific about it.
WILL SCHWEITZER: We bring context, we bring data. We've been bringing those little nuggets that have been floating around in our head, and we help each other connect the dots. I hope this day ends up essentially being a productive talk therapy session. So with that, I just wanted to say thank you. As always, please assume positive intent. It takes a lot for some speakers to be on this stage.
WILL SCHWEITZER: We are in a pretty incredible community that is generous. A lot of folks want to help each other. There's a lot of value in essentially having the 180 of us together in this room today. Contribute generously when you can. Ask thoughtful questions. Bring data to the table. Rather than just saying, I'm afraid of this because it's a big boogeyman.
WILL SCHWEITZER: Say, I'm afraid of this because of data point x or y. And last, my ask is to meet somebody new. There are a lot of new faces in the room today, and I really don't want anyone to leave a stranger. So with that, if any of you need anything throughout the day, you are in our home. Please find any Silvercharian. Any member of our team will be glad to help you with just about anything.
WILL SCHWEITZER: So with that, I'd like to invite Andrea Hoffman, the COO of Hum, to the stage to welcome Wendy, Andrea.
Segment:0 .
STEPHANIE LOVEGROVE
HANSEN: Welcome, everyone,
HANSEN: to Platform Strategies 2026. Or as my colleague, Sam Green, called it when she saw the agenda, Platform Strategies, the Reckoning. So we're in for some good discussions today. My name is Stephanie Lovegrove Hansen. I'm the VP of Marketing at Silverchair, and it's delightful to have you all with us today. I'll see a lot of returning faces. So you know that we're in for a very exciting day.
HANSEN: And for those that are joining for the first time, welcome. Pro tip, don't miss the swag table. So throughout the day, your primary source of information is going to be the event app, which hopefully you've downloaded. There you can see the agenda. You can ask questions if you don't want to do it IRL. And you can also connect with each other and see who else is here.
HANSEN: So if you have not yet downloaded that, you may do that. There's some instructions at the registration desk. So today's sessions are being recorded, and will be made publicly available to you after the event. And returning attendees know that we always like to do something a little bit fun. So we've had live sketch artists.
HANSEN: We had a very creepy AI photo booth at some point. And this year, however, we're really stacking the deck. We are not holding our cards close to our chest, but instead dealing you in with custom playing cards that represent different stages of the research life cycle. So please take a minute to flip through those. But that's not all. The ace up our sleeve, as they also come with custom scholarly publishing games that are within each deck.
HANSEN: And you can also, if you play your cards right, get your photo taken as the wild card of your choice. So I have now run out of card puns. So I will take a minute to then thank our sponsors for this year's event. Digital Science, our charging station sponsor, which you can find in the back of the room if you need that throughout the day. Impelsys, our beverage station sponsor.
HANSEN: ADIV, our app sponsor. DCL, our morning break sponsor. Access Innovations, our afternoon break sponsor. And finally, Hum, our breakfast sponsor. So thank you for supporting this meeting. And now I will hand it over to Will Schweitzer. [APPLAUSE]
WILL SCHWEITZER: I didn't write nearly enough dad jokes. Very hard to follow. I hope after four years, I've learned how to use the microphone. If you can't hear me in the back wave, yell, scream. I'm Will Schweitzer, I'm Silverchair's CEO. I'm really glad that you've joined us. Over the next 15 minutes, I'm going to try to preview all of today's sessions by zooming out, and providing some commercial market and technology context.
WILL SCHWEITZER: As always, when I do these talks, there will be a few iCharts. There will be some salty remarks, trying to set the tone for the day. And I want to start with this one. The market has repriced publishing, and our readers stopped being human. So I hope you've had your coffee. Nothing really happened on February 3.
WILL SCHWEITZER: In fact, I woke up thinking about how Bad Bunny couldn't perform at the Grammys. And now Trevor Noah lives rent-free in my head as he was trying to sing "DtMF." Anthropic shipped a legal plugin for Claude. And what happens then was an information company and SaaS-pocalypse, something you've probably seen in the news every day since. But when we think about the publishing companies in our space and the value that they lost, none of them lost a customer.
WILL SCHWEITZER: Not one product failed, not one journal folded. No one lost essentially shelf space in the market. And ever since then, ever since there was the largest decline in Elsevier's stock price, in RELX stock price in over a decade, or a substantial drop in Thomson Reuters stock price have I been able to make any sense of the market? And to the credit of our friends at Clarke and Esposito for their ongoing analysis about this, the companies with the most AI announcements in our space, the Elsevier, the Thomson Reuters, the Wolters Kluwer have lost between a fifth and 2/5 of their market value over the past 12 months, despite the fact that they've all posted strong revenue growth.
WILL SCHWEITZER: So the market is no longer scoring product announcements. The market is scoring what they see as the durability or the lasting value of what all of us in this room do. So fortunately for us, we have Wendy Queen, the chief transformation officer of Johns Hopkins University press. Reflecting on what we've collectively learned in the past 12 months, and then maybe she can help us make sense of stock prices.
WILL SCHWEITZER: There's another analog out there, and I am really hoping that we are not sad Pepsi. Three years ago, in one Bloomberg interview, there was a Walmart executive who said because of GLP-1 drugs, like Ozempic, customers were starting to put less snack foods in their carts. And within one week, every single company that has something in the snack aisle was repriced. They collectively lost about a quarter of their value in one week alone.
WILL SCHWEITZER: A majority of those companies have continued to grow, just like Elsevier, just like Wiley, just like companies in our space, their stock value has never recovered. And there's an interesting contrast here in Coca-Cola and Pepsi. And I really do hope that we are not sad Pepsi in our market. In that Coca-Cola, their stock prices come back up. They have a slightly different business strategy, but fundamentally, the market believes in Coke and has more confidence in their long term strategy than they do Pepsi.
WILL SCHWEITZER: And a lot of that is just about sentiment and mood. This is a bit of a tangent, and I want to try to bring this back with the confusing headlines mean for all of us. The global AI build out. When we read about the hyperscalers, the companies like Amazon or Oracle, building data centers here in Northern Virginia or elsewhere across the country.
WILL SCHWEITZER: All of it is being done with borrowed money, and is being done with borrowed money at an incredible scale. So last year, we talked about the amount of corporate R&D being spent on AI, and how that was going to transform our daily existence. It was going to transform our products and our market. This is particularly troubling. The amount of debt raised this year is 121 billion. That's just on the books.
WILL SCHWEITZER: Off the books, there are $662 billion of financing that isn't shown on balance sheets. That isn't transparent to the market. And what is even more troubling is the bets against this are now being done with a credit swap options, which are the same things that led to the mortgage collapse in the early aughts. That is never a good sign for the market. So what does that mean for all of us in the room?
WILL SCHWEITZER: It means our cost of borrowing money are going to be higher. It means they're going to continue to be confusing headlines. And my best suggestion is this, tune out the day-to-day noise, and focusing on the long-term pragmatic products, pragmatic developments. Leveraging these tools in your company is about all that you can do, because the volatility is here to stay.
WILL SCHWEITZER: That was a slight tangent. So thank you for indulging me. Following Wendy's keynote, my good friend and spiritual counterpart, Dustin Smith, the CEO of Hum will moderate a panel with Dawn Melley from IEEE, Miriam Maus from IOP, Jonathan Woahn from Cashmere on leveraging AI systems to ensure research integrity and quality from submission all the way to content discovery and synthesis.
WILL SCHWEITZER: And here's some data from ScholarOne to help set the stage. Desk rejects and ScholarOne grew 72% from '22 to '25, against 43% growth in decisions overall. Our publishers now perform 2 and 1/2 desk rejects for every acceptance. The cost of us collectively saying no is the fastest growing expense line in scholarly publishing, and not one of us has a way to recover that expense.
WILL SCHWEITZER: And on the reading and synthesis end, last week, the OECD released data from their PISA learning assessment, a global learning assessment, and it paints a really interesting picture. The headline is this, there is more text produced faster, read more hastily by readers who evaluative capacity is measurably weakening. And I say that without trying to be alarmist. If you haven't discovered Tina Austin and her unbloomed Substack, please go read it.
WILL SCHWEITZER: She is the director of AI education, licensing and public interest technology at UCLA. While she often writes about AI in the classroom, the findings she's drawling out can help us inform our product strategy, or help our colleagues leverage AI tools at work, or help us with all of our editorial stakeholders in the world. I think the kids will be OK.
WILL SCHWEITZER: Her headline coming out of those kind of dire PISA data snippets was that students who repeatedly engage with AI tooling, who are in advantaged classrooms where their instructors can provide the right framework for using AI tools do better than they would otherwise. But that says a lot about relative equity or advantage in the classroom, and how we are setting students up to succeed. Following Dustin's panel, we'll move on to table topics-- apologies to the introverts in the room in advance-- and we'll discuss how peer review is under pressure from every direction.
WILL SCHWEITZER: So here's a bit more context from ScholarOne. And I'm purposely using business terms here to see if it can help shift our thinking. The aggregate reviewer pool, the number of reviewers captured in the ScholarOne system has grown 54% since 2018. Invitations to review have more than doubled. Acceptances to review have halved. It takes 4 and 1/2 invitations to land one review, and per 100 invitations, an editor spends a whopping 407 days waiting for a response.
WILL SCHWEITZER: And that response, more often than is no. That isn't a shortage of reviewers. That is a dead inventory problem, and inventory problems tend to have business solutions. What's more important here is as publishers, we are not passive recipients. We can actually change our systems and processes. After the table topics, we'll break for lunch. Introverts like me can go recharge.
WILL SCHWEITZER: And then we'll come back to a panel discussion that I'm really excited about. My colleague Ryan Ross will talk with David Sampson from NEJM, Meagan Phelan from Science, Suze Kundu from Defend Research on the role of technology in the misinformation era. And that truth may really be a platform problem. I was sleeping really well at night until I read the preview for this session.
WILL SCHWEITZER: [LAUGHTER] Really well. So zooming out just a little bit. Pew Research watched 69,000 people perform Google searches. When an AI summary appears, the click through rate onto the underlying or the resulting article dropped from 15% to 8%. 1% of people click on a link that is in that AI summary. The replacement channel isn't arriving either.
WILL SCHWEITZER: Those are eyeballs that are lost. And outside of our industry, when we look at media web properties in general, ChatGPT, which is the most common consumer agent, is only 2/100 of a percent of publishers referring traffic. We are starting to adapt in our industry. Counter was consultant in February. They released an AI best practice in April, and it is a really good start.
WILL SCHWEITZER: The problem is there are a lot of things that counter, and this best practice still aren't counting. Counter is the proposed best practice is only kind of measuring agents that come in through the front door, and that kind of resolves to an article itself. What we haven't solved for yet is what happens when there is federated authentication? When there's easy proxy in the mix.
WILL SCHWEITZER: When you are using a tool that is working with your browser to then come and navigate to an article and perhaps scrape it or extract it. That is a lot of traffic that we can't see. So this is a good start, but there's a lot of work for us to do. I'm really concerned about the training corpus, and the worldview that is behind leading models, whether that is Claude, or ChatGPT, or Chinese open-weight models like Kimi.
WILL SCHWEITZER: When ChatGPT was new, researchers in Beijing tested it, and it thought Yao Ming was a woman. Their conclusion wasn't that the model didn't work. Their conclusion was that it had been trained on the wrong data. So there is now a large scale government-funded initiative within the Chinese government to fix that by the Chinese research apparatus becoming the largest supplier of training data. And that includes scientific research. A Princeton sociologist who studies this put it better than I ever can.
WILL SCHWEITZER: AI disconnects the messenger from the message. Keeping those two things attached. Provenance, attribution is the job of everyone in this room. So I don't know what your geopolitics are. I don't know your relative sense of ethics. But I think this gives us some urgency to the thought that our knowledge belongs in this broader and connected AI ecosystem. Following that discussion of truth, relative or otherwise, I'll be back on stage with some really good colleagues and friends to redo a panel discussion that happened on this stage two years ago on what is your strategy if there is no platform?
WILL SCHWEITZER: So I'll be joined by Alison Mudditt from PLOS. Andy McGregor from Sage, Ann Michael, David Crotty from Cold Spring Harbor Press. And just a little bit of data from the Silverchair platform. Between June and August. One publisher on our platform, only 41% of content requests scored as likely human. And the rate of likely human readers continues to fall every month.
WILL SCHWEITZER: When we look at providers who can provide a larger sampling of the internet, Cloudflare estimated that 52% of caller requests were for training. 53%, Imperva would say that 53% of internet usage is non-human or robots. The shape of the reader on the internet is changing fast. For 20 years, our policy of asking non-human readers, robots, was essentially a polite text file on our site.
WILL SCHWEITZER: There have been some recent kind of judicial law developments here. In December, a federal court described robot.txt accurately as a mere request not to scrape. OpenAI's own documentation says that when a user initiates an actions robot.txt may not apply. The opt-out era is over, and what replaces that is cryptography. And here's the part that interests us.
WILL SCHWEITZER: We may be able to prove who is asking, who is coming to our sites using a chat agent before we can tell them what they are allowed to do? So we may know their identity, but we may not be able to pass back to that user to their agents whether or not they can scrape content, whether or not they can access full text, whether or not a subscription is valid. So it is a interesting time for evolving technology here.
WILL SCHWEITZER: When we think about the products in our space and those chat agents, this agentic distribution channel exists. It is getting a lot of use. The default pathways, the ways end users using these agents can connect to corpuses in our market. They're being established now. And nearly none of you in the room are part of it. Two publishers in our space have shipped connectors into assistants like Claude and ChatGPT.
WILL SCHWEITZER: One of them is Wiley. None of them are independent or society publishers. And when Anthropic turns on Claude science for an end user, the entire scholarly literature layer is drawn on OpenAlex and arXiv, on open metadata and preprints. That means for all of the publishers in the room, your content sets are not part of the default answers. Reachability into these AI tools is now the new indexing, and the index is being built right now without the participation of a lot of us.
WILL SCHWEITZER: Last Thursday, there essentially was plaintiff evidence from the OpenAI New York Times lawsuit that was unsealed. Nothing's been decided yet. Replies are due in November. The US Justice Department has just weighed in, but the line that sticks with me is here on the left-hand side, which was ahead of ChatGPT within OpenAI saying their products are largely substitutive. And on the right-hand side, Satya Nadella, the CEO of Microsoft, was being disposed, and he was asked if he thought ChatGPT or other model builders should have to pay to train their models on paywalled content, and the New York Times in particular.
WILL SCHWEITZER: And his answer was yes. Two things for the room. This was largely a case concerning news content, and news content is a summary of something that happened. So if a summary is summarized, that's it. That's a very different construct from research where a summary of research may be enough for a fraction of use cases, but not for all of them. And access to the underlying data, the discussion around the paper.
WILL SCHWEITZER: The suggestion for what comes next. What should be studied next. Broader implications for the community or for medical practice, or whatever it may be, is our structural advantage. And we need to figure out how to hold that line. Another really interesting thing a lot of us have been thinking about training data, and whether or not we can sell our corpuses for training.
WILL SCHWEITZER: AI licensing is not a rounding error. It is not an item at all in the budget of folks that are building frontier models. Alphabet publishes their costs for traffic acquisitions. It's about $600 billion. It names content acquisition costs for YouTube, but declines to size them. For AI training data, it declares nothing at all because there's nothing material to disclose.
WILL SCHWEITZER: Microsoft's only mention of the subject, what it pays for content to train their resources is a risk factor. In 300 pages, they say it may one day have to pay for data.
SPEAKER: It's super easy to open a--
WILL SCHWEITZER: No, quite OK. When we think about the dollars that have flown in our space, large commercial publishers securing dollars for training. It amounts to 0 but a sugar rush. It is good for the moment. It is not a sustainable meal. The next session on the agenda, hello from the other side is moderated by my colleague David Nygren, and in talking about or sharing some data from Taylor and Francis on the previous slide.
WILL SCHWEITZER: I hope I have not pissed off one of our interviewees and one of our largest customers. Who is Penny Ladkin-Brand, Taylor and Francis CEO. David and Penny will be joined by Kelly Palmer from TELUS Digital, Ha-Hoa Hamano from the US Department of Education and formerly NPR. And the session is usually a crowd favorite. In it, we ask those panelists to bring context from previous industries they've worked in or from the industry they're in now, and talk about how it may apply to us in scholarly publishing.
WILL SCHWEITZER: Before Taylor and Francis, Penny was the chief financial and strategy officer for Future. And I would be really curious for her view on what showed up in Futures numbers first because traffic is probably a lagging indicator of something. I'm reaching the very end of the day. I'm really excited for our closing keynote. We'll be joined by Jacob Ward, who is a CNN contributor. He has reported for NBC, Al Jazeera, CBS.
WILL SCHWEITZER: He was the editor-in-chief of Popular Science. He's been on the news an awful lot. We have his book available at the registration table, and he's going to talk to us about what human skills must be protected when AI gets really good at everything else. The data here on this slide is actually looking at the cost of inference of using these models. We are operating in a world where you can now manufacture a plausible research paper using a GPU.
WILL SCHWEITZER: What you can't do with tokens is manufacture the warrant behind the research. The validation, the accountability, the author's name, the institutional affiliation on the byline. Everything around judgment is getting cheaper, but judgment is about to be the most expensive input to all of our products. And is what is becoming increasingly remarkable, particularly for you all who are using ChatGPT or Claude on a daily basis is that the older models are getting cheaper, and you can still do an awful lot with those older models that aren't the latest release of Fable or Opus 5, whatever we're up to now.
WILL SCHWEITZER: And we're all just learning how to cost optimize our use of those tools. There's still a lot of work for us to do there. On the left is a quote from Alan Garber, who's Harvard's president from the first day of classes this month. His address was about human intelligence being distinct and precious, and about defending the work of researchers itself.
WILL SCHWEITZER: He was suggesting that AI will allow researchers to move faster to push the evidentiary boundary. And on the right is a quote to OpenAI that I keep coming back to. Adoption of AI is being urged by institutions that consume scholarship on tools who builders privately described as being substitutive of coming to our journals, coming to our platform sites, interacting with our products. And I don't think either of them is lying or wrong.
WILL SCHWEITZER: Their incentives just simply don't match up, and we're caught in the middle of this. So last year, my invocation for everyone, my ask of everyone in this room was to have the audacity to not mope. That worked for about, I don't know, two weeks for me last year. So I'm going to take a slightly different approach this year. And that is, I think it could be very valuable if we express our fears, if we name them, and we get specific about it.
WILL SCHWEITZER: We bring context, we bring data. We've been bringing those little nuggets that have been floating around in our head, and we help each other connect the dots. I hope this day ends up essentially being a productive talk therapy session. So with that, I just wanted to say thank you. As always, please assume positive intent. It takes a lot for some speakers to be on this stage.
WILL SCHWEITZER: We are in a pretty incredible community that is generous. A lot of folks want to help each other. There's a lot of value in essentially having the 180 of us together in this room today. Contribute generously when you can. Ask thoughtful questions. Bring data to the table. Rather than just saying, I'm afraid of this because it's a big boogeyman.
WILL SCHWEITZER: Say, I'm afraid of this because of data point x or y. And last, my ask is to meet somebody new. There are a lot of new faces in the room today, and I really don't want anyone to leave a stranger. So with that, if any of you need anything throughout the day, you are in our home. Please find any Silvercharian. Any member of our team will be glad to help you with just about anything.
WILL SCHWEITZER: So with that, I'd like to invite Andrea Hoffman, the COO of Hum, to the stage to welcome Wendy, Andrea.
Form Title Download
Transcript
Form Title Bookmark
Entire Media
Segment(s)
Custom Clip
Start At:
End At:
Form Title Share
Entire Media
Segment(s)
Custom Clip
Start At:
End At:
Form Title Accessibility and Keyboard Shortcuts
The cadmore media player is a professional video and audio player experience designed to deliver media as well as relevant transcript, segmentation, and metadata. On the screen is a traditional html video player with buttons such as play, pause, forward, captioning, language and play speed selection and more. Adjacent to the video player are generally tabs, one of which shows a scrolling transcript of the video and the other shows any segmentation of the media, sometimes referred to as chapters. At the top of the video is a menu bar containing an Explore button and Tools button. The Explore button allows you to dig deeper into the video by opening up an explore dialog which contains tabs for a visual navigation of the video using thumbnails, an about tab which contains rich metadata about the media, a segments tab which contains further information about the video, and a related tab which contains links to other media related to this media. The tools button opens a menu with options for sharing the media on social media, creating bookmarking links, creating embed codes for your website, generating citations, and more.
Keyboard Shortcuts
Keyboard shortcuts only work while you are in forms mode. By entering forms mode, you will be able to quickly access portions of the media player with single key commands.
Spacebar: Play/Pause Toggle
I: Info Menu,
O: Tools Menu
T: Transcript Tab,
S: Segments Tab
F: Forward 15,
R: Rewind 15
L: Languages,
P: Playspeed
M: Mute Toggle,
V: Volume Bar
N: Video Only Toggle,
B: Full Screen Toggle
C: Captions Toggle
Escape: Closes dialogs and menus / Exits Player
Z: Next Transcript Chapter
X: Previous Transcript Chapter
Q: Next Transcript Paragraph
W: Previous Transcript Paragraph
Entire Media
Segment(s)
Custom Clip
Start At:
End At:
Embed Size
-
You Size - Responsive
-
We Size - Responsive
-
Picture Overlay Popup
-
400 x 225
-
512 x 288
-
400 x 700
-
768 x 1024
-
1024 x 512
-
1280 x 608
Form Title Cite
No citation provided.
Info
Segments
Related
Thumbnails
Exit Clip Mode
You have requested to see a portion of the video outside of the clip.
Click OK to switch to the full video or click if you would like to stay inside the clip. You can always go back to the clip with Tools and Clip Mode