How can AI-powered language models (like ChatGPT) be useful for researchers in the environmental sciences?

This is work in progress....but I think there are exciting opportunities here (updated).

How can AI-language tools for researchers in the environmental sciences? (Image created by DALL-E 2)

(updated 4.2.2023)

Unless you have been off-grid on a tropical island sipping margaritas, you will likely have heard of ChatGPT, OpenAI’s chatbot that everybody is excited about, worried about, of both. It is truly an impressive tool, without a doubt, so if you haven’t played around with it, give it a whirl. There are other tools like this as well, and it’s also pretty clear that these products will improve rapidly. In fact, check futurepedia for a directory of AI tools. These large language model tools will also become increasingly integrated with other apps, making their use more seamless.

Beyond the worries that these apps will eliminate homework or essay writing in colleges, or that it will flood the internet with automatically generated text, can such tools also be useful for researchers, for example in the environmental sciences? So in an initial self-trial during the last few weeks, I have tried to figure this out for myself, and this is clearly work in progress.

If you are a researcher in the environmental sciences, what tasks could large language model tools such as ChatGPT help you with?

1. Editor and reviewer. Without a doubt, these tools has excellent language skills (and you can prompt it to produce output in a certain style, for example academic language).

   1. Line editing. So especially if English is not your first language, you can use this app to help you edit specific sentences (i.e., line editing) for grammar, clarity and style. I have tried it and was quite pleased with some of the results. Since a lot of our tasks as researchers involve writing, this is probably the most important and most obvious use of these apps. And I don’t mean: ‘write this next paper for me’ type of help, but just when you are stuck on how to say something clearly.

   2. Text appraisal. You can also ask the language model to appraise your text style: about this bullet point it said that it is written in an accessible manner, suitable for an academic audience. Good.

   3. Routine tasks: You can ask the model to suggest keywords, titles and produce a draft of a summary for a paper (I received this on twitter, haven’t tried it myself yet). Might also be useful for adjusting word counts of abstracts, where requirements differ among journals (tip from: Françoise Cardou), and perhaps other such routine tasks.

   4. A word of warning: The key word in all these points is ‘suggest’. It still needs to be you that checks these text proposals for accuracy. As anybody who has played around with this app can tell you, some responses are just inaccurate. Human verification is key!

2. Idea creation/ brainstorming. I am always on the hunt for tools or situations that help me ask more creative questions, and admittedly this is why I was initially becoming interested in ChatGPT. Now, if you ask it ‘generate 10 new questions in community ecology‘, it will not come up with anything beyond boilerplate stuff, at least when I tried it. But if you ask it to combine some concepts and topics, then I have found that the replies contained some statements that I found quite interesting and worth pursuing. This is great, I think. New ideas typically arise from combining seemingly unrelated topics, and so making the chatbot produce language using and combining the topics in your prompt can give rise to new thoughts. It’s then of course up to you to check which prompts are worthwhile, but that is no different from a brainstorming session.

   1. You can fiddle with the prompt text quite a bit by changing specifications about the output; for example, I liked asking it ‘give me a list of 20’, and then screen those for useful content. In general, you can tell it what you want the algorithm to consider in generating the output. You can also use the ‘recreate’ button or slightly change the prompt text (by going up and clicking the edit button) to obtain different results. You can use the prompt “act as a ….”, for example, “act as an engineer” to get answers from a particular vantage point.

   2. You will need to start a new chat when you are starting a session about a new topic, since evidently it remembers what you talked about before in any given conversation, so start with a clean slate by reloading the page.

   3. One interesting use is to ask the language model to summarize concepts from other fields (as pertaining to an issue you are working on); this I found very useful, and it’s hardly surprising that this can be interesting, since ideas are often generated by importing ideas from one field to another. I prompted the app to produce structured outputs according to certain criteria, like hierarchies or precise relationships among terms.

   4. I found that this works best with more general requests, like linking some themes in the environmental sciences and ecology, or linking concepts in ecology with those in other fields. The app tends to not work very well with very specific technical topics; because for the latter, maybe ChatGPT will not know the exact definitions or background, which might limit the usefulness of responses, and typically you will know more about this yourself anyway.

3. Outreach and communication text material. It can be used quite readily for creating first drafts of texts for use in communication, like tweets, summaries, newsletter posts or other materials. You can specific the type of output (including number of words) and text style in the prompt. Perhaps also useful for some routine emails.

4. Coding. Not really my department, but people tell me on twitter and tiktok that the chatbot can be very helpful in providing, checking or improving code, for example using R or python, but also other programming environments. Given the paramount and ever-increasing importance of statistical analyses and coding in ecology & environmental research, this is likely going to be a big one. Perhaps this app can play the role of a tutor.

5. Routine text screening tasks. As discussed in this blog post by Shinichi Nakagawa, these tools may have a use in the screening of literature for inclusion in systematic reviews, after being properly ‘trained’. In initial trials, it seemed to perform quite well, checking title and abstract for potential inclusion in a systematic review (thanks to Phil Martin for alerting me to this).

6. Teaching. I would be quite interested in including this tool in teaching. Students will know this tool anyway, and it could be a good opportunity to have students play with this app in a productive way. I found this interesting post from Jackie Gerstein on a collection of ideas to use ChatGPT in teaching.

7. Note taking. This seems to have quite a few people excited, and there are already integrations with popular note taking apps. I can see how this can be useful in summarizing notes, and in terms of linking them.

This is what I came up with so far (also with the help of the fine people on twitter, mastodon and tiktok). I do not believe tools like ChatGPT are currently useful for producing scientific text, like for use in a manuscript (for example, there are no real references). But this doesn’t mean it can’t be helpful in other ways.

I think that in the near future, or even already now, we as scientists will all be integrating in some way this interaction with an AI-tool like ChatGPT in our workflow. Van Dis et al., writing in Nature (3.2.2023) even go further:

This technology has far-reaching consequences for science and society. Researchers and others have already used ChatGPT and other large language models to write essays and talks, summarize literature, draft and improve papers, as well as identify research gaps and write computer code, including statistical analyses. Soon this technology will evolve to the point that it can design experiments, write and complete manuscripts, conduct peer review and support editorial decisions to accept or reject manuscripts.

Maybe this will happen. Would also love to hear from you if you found other uses for researchers, please share in the comments!

Liked this post? Please consider subscribing...thanks!