Can ChatGPT support scholarly communications? Interview with Christopher Leonard
Since its release in November last year, ChatGPT has led to a flurry of activity, with users testing its capabilities and limitations in all different ways imaginable. It has been two months now, and it certainly does not seem like this is just a temporary buzz surrounding a new AI tool that will die down soon. Its range of potential applications seems vast—from mundane to highly complex tasks, and for both honest and deceptive purposes.
It is able to generate texts mimicking human-created content in response to even complex queries. And this means that it is very likely to impact scholarly communications in a big way, because scholarly progress depends on knowledge shared through the written word.
In this interview, I ask Christopher Leonard (Director of Strategy & Innovation at Cactus Communications) his views on how this tool may find use in research and academic publishing and the broad implications its use may have.
Chris, to begin with, could you tell readers a little about yourself?
For 25 years, I’ve held a variety of roles in publishing, technology, and product development. My belief is that NLP, AI, machine learning, quantum computing, and blockchain will power the next age of industry and society, but my fear is that they will be used inappropriately. Therefore, I have a strong interest in both the ethics of implementation and the environmental cost of these technologies.
AI tools and applications have been around for a while and have been gaining a lot of attention in scholarly communications. But none has generated as much excitement, worry, and debate as ChatGPT. Could you share a brief view on what is different this time?
The accessibility of ChatGPT and the ability to get something meaningful and surprising out of it on your first go is probably what sets it apart. Then, when you get better at prompting and interrogating it, you realize that this could change a lot of things from education to copywriting—and it’s here now, not in some distant future. This sense, that the future has arrived, is probably why we’re all talking about it.
Do you see ChatGPT playing a positive role in facilitating research? For instance, can it support researchers in their tasks of knowledge collation, creation, and synthesis? What do researchers need to keep in mind when using this tool?
I think it will be most useful for proposing new research projects or exploring new avenues (especially if you are “stuck” on something), but its suggestions will need to be sanity-checked and filtered through human wisdom before being acted on.
We saw with the brief access to Galactica, a large language model based on academic text, that generative technologies tend to speak authoritatively, but without substance. “Facts” can be just probabilistic sequences of words, and citations can be entirely fabricated. There is a whole bunch of issues to work on for a large language model trained on the entire corpus of academic output—not least of which is the fact that many of the historical parts of academic output have since been proved wrong. I hold out more hope in this regard for directed knowledge graphs, rather than generative text.
How can ChatGPT help scholarly publishers? For instance, can it help them communicate more effectively with authors or gain insights about research segments they wish to engage with?
One thing that ChatGPT really excels at is the whole “explain-this-to-me-like-I’m-an-8-year-old” angle. If we want to engage a wider public with research, then summaries at different levels of accessibility would be a great way to allow non-experts to get into the world of academic research. This could be a service the publisher or author avails themselves of, although authors have the advantage of being able to double-check the output, tweak it, and promote it on social media.
For publishers specifically, ChatGPT could be used to reduce the length of article titles, or to polish (and usually shorten) the abstracts supplied by the author. It may be that ChatGPT can be trained to rewrite abstracts so that they are more “keyword-heavy” as well as being more readable. This will help in their discovery in A&I databases.
A simple use case, but one with a potentially large upside, would be to use ChatGPT to write personalized e-mail invitation letters to peer reviewers. These would be more enticing for a potential reviewer and hopefully explain exactly how they are qualified to review a certain manuscript.
The misuse of advanced AI tools in scholarly communications has already been recognized as a problem, for example, with some being used to generate fake images. A major concern around ChatGPT is that it may make it easier to plagiarize content or to create fake papers—problems already plaguing academic publishing. What are your views on this? How can this sort of misuse of the tool be prevented or detected?
It’s going to exacerbate a real problem that we already face, with paper mills generating nonsense papers and fake images. Add to this the ability to paraphrase text to avoid simple plagiarism detection, and we have a set of problems we are likely to see get worse before we get a handle on them.
My response to this is one based on best practices of open science:
- We need to have “open” peer review. We want to see who reviewed this paper and what they said about it. It’s harder to get nonsense papers published when the reviews have to be published alongside the paper itself.
- We need to see the data accompanying any paper to be published and downloadable with the manuscript itself. If there is no data to accompany an experimental paper, alarm bells should be ringing. If there is data and it has been “edited,” that will become obvious when people start to replicate the experiments.
Other than the views you’ve already shared, what broad implications do you see tools like ChatGPT having on scholarly communications in the long run?
Some of the advantages we’ve already mentioned above, but the ethics of generative text are something we are going to learn as we go along in 2023. Most text is generated from, or at least mimics, existing published material. How much of that is under copyright, and what are the implications for the generated text (who owns the copyright to a derivative work with many unidentifiable inputs)? Similarly with plagiarism—if we can easily paraphrase some existing work, who is going to know?
At some point in the future, the unit of academic communication may not be the article at all. We may become more reliant and trusting of datasets without narratives. Since we can generate text and images so easily now, maybe the real value lies in the data and not the researcher’s interpretation of their own data.
On a very broad and more positive level, we’re going to have to reconsider how we test students so that we are not so reliant on written assignments. What could be viewed as a threat to education is an opportunity to change how we teach and test students.
Comments
You're looking to give wings to your academic career and publication journey. We like that!
Why don't we give you complete access! Create a free account and get unlimited access to all resources & a vibrant researcher community.
Subscribe to Conducting Research
