Hi Glenn,
Great questions.  As far as specific licensing goes, Copilot and such do not care.  The details of the various open licenses that mean something to us are irrelevant to the LLMs.  It will take in anything on the public web.  Now, I do wonder whether it's capable of "reading" PDFs or if the content has to be in HTML for it to be intelligible.  Might see if I can figure out a way to test that.
As for identification of sources, yes, that's one of its great strengths.  I'll provide a specific example.  Just now, I asked Copilot, "How did multilevel society evolve in the primates?"  (I researched ecology/evolution/behavior before I became a librarian.  I find when judging the performance of these tools it helps to ask questions where I already have in-depth knowledge.)  The answer was an OK but not great summary, but more importantly, it linked directly to this paper https://link.springer.com/article/10.1007/s10764-012-9618-z which is a key source in this area of research.  So, one would presumably read and cite that paper.  By the way, this can also be used in reverse to judge how much stock you ought to put in the quality of Copilot's answers...if your sources are a string of random blog posts, you probably shouldn't take the response too seriously.
In any event, I hope the AI's application as a discovery tool is clear now.
Best,
Ali K.

From: Glenn Hampson <[log in to unmask]>
Sent: Tuesday, June 4, 2024 11:33 AM
To: Ali Krzton <[log in to unmask]>; [log in to unmask] <[log in to unmask]>
Subject: [EXT] RE: OA papers more visible to AI search
 
CAUTION: Email Originated Outside of Auburn.

Hi Ali,

 

Interesting. Can you elaborate? Specifically, you say that Copilot “can only surface peer-reviewed sources that are open access.” Does this mean it doesn’t intake information from PubMedCentral (most of which is copyrighted VOR past embargo but not CC-BY licensed), or from preprint servers (generally not peer reviewed), or from institutional repositories (like PMC, free to read but with a variety of copyrights)? And does it only include CC-BY licensed material or does it also include material with CC-BY-NC and CC-BY-NC-ND licenses (these latter two are more popular but carry restrictions on derivative use)? And finally, are sources identified, because no author (even when publishing in CC-BY) wants their work to be reused without some acknowledgement of provenance, especially in academia (unless we’re talking about data being released CC-0)?

 

Thanks,

 

Glenn

 

Glenn Hampson
Executive Director
Science Communication Institute (SCI)

 

 

 

From: OpenCafe-l <[log in to unmask]> On Behalf Of Ali Krzton
Sent: Tuesday, June 4, 2024 8:57 AM
To: [log in to unmask]
Subject: [OPENCAFE-L] OA papers more visible to AI search

 

Hi all,

Given recent discussions around the impact of AI on research generally and libraries in particular, I thought I would mention another possible implication of open access.  In recent months, I've been working with Microsoft Copilot, as our university has rolled out a protected instance of it.  A colleague and I poked and prodded it extensively in the service of developing a workshop around responsible use of AI as a research tool, and one thing we both noticed was that it can only surface peer-reviewed sources that are open access.  For those unfamiliar with Copilot and similar systems, these LLMs can interface with content on the open web, unlike ChatGPT.  However, they can't get around a paywall, regardless of whether your institution has the relevant subscriptions.

Now, more advanced AI-powered research tools that allow users to build their own curated libraries (you can probably think of some) can access paywalled research so long as the user has uploaded it (which may create licensing friction, but I digress).  But Copilot is something anyone on any device can pull up and use for free to ask natural language questions and get answers with associated references.  So, an additional incentive authors might have to make their work open access is knowing it can be highlighted as a source by these public LLMs.  Before anyone says, "but serious researchers at academic institutions are more likely to use proprietary databases for their searching/discovery", please bear in mind the popularity of Google Scholar.  In my experience, given how rapidly Google has degraded their search experience of late, Copilot now works better as a general search tool, so I don't see any reason it couldn't challenge Google Scholar for academic search.

I submit, then, that AI creates an additional incentive for savvy authors who want to broaden the impact of their research to make it open access.  And librarians like us that already struggle to get our users to search comprehensively face yet another barrier to connecting them to the content we actually subscribe to.  Curious to hear others' thoughts on this.

Ali Krzton, Research Data Management Librarian

Auburn University

 

Access the OPENCAFE-L Home Page and Archives

To unsubscribe from OPENCAFE-L send an email to: [log in to unmask]


Access the OPENCAFE-L Home Page and Archives

To unsubscribe from OPENCAFE-L send an email to: [log in to unmask]