Why any scientist can benefit from ChatGPT and other natural language processing tools
I work as an academic scientist. As for most of my peers “doing science” (in my case developing and applying AI for research in life sciences/sustainability) is only part of my job. I also direct a research school, develop and teach courses, supervise undergraduate and PhD students, give talks to scientists, external stakeholders and the general public, listen to talks given by others, write scientific articles and the occasional popular science text or commentary, organize workshops/seminars/conferences or attend them, conduct outreach via social media and professional networks, search for and read research articles, review grants and articles, give interviews to news outlets, apply for funding, maintain websites, develop sustainability guidelines for my group, sit on the department board, recruit staff, buy equipment, write various reports and a gazillion emails and chat messages. And more.
I enjoy many of these additional tasks tremendously and see them as a great complement to my core research tasks. Discussions with fellow scientists, external stakeholders and students inspire and improve my research ideas and help me to continuously expand my competency. Reading research articles, listening to scientific talks, and writing grant applications always gives new ideas — more than I could ever follow up on. Teaching and supervising the next generation is fantastic fun and often educational for myself — and I am humble enough to realize that the numerous people I will teach over my career will have a larger cumulative impact on this world than I could ever hope to have. I also think outreach to the general public is an essential aspect of my work, my way of enabling informed and nuanced societal discourse — the backbone of any democratic country — and my chance to share our research insights and raise issues I care about.
Nevertheless, there is one big problem: there is not enough time and too much of the time I have is spent on tasks that are necessary but do not really help propel science forward or contribute much to education or outreach. Many of these tasks involve writing. I cannot even begin to count the hours I have spent on re-formatting my CV, publication list and project description for grant applications. Some foundations want a 2 page CV, some want 3, others want specific headings, some want publications from the last 5 years, some want my top 10, some want a 4 page project description, others want 10, many have specific section order and headings. I hope one day funders will standardize this but until then it is a total nightmare and the most colossal waste of collective researcher time I can think about. It’s similar with short project descriptions for websites, texts announcing seminars or conferences and email texts, which are to be polished and tailored to different contexts and recipients, even though the main content stays the same. I also need to write course plans and descriptions and various types of reports. Many of these tasks do not take much time individually but they DO ADD UP. And that is before we come to the truly massive writing tasks: drafting, rewriting and polishing research articles and project descriptions for grant applications. I truly wish I could do more popular science writing, but where should I find the time?
So, will I use ChatGPT and other large language models, Google Translate, AI-based grammar and spell checkers and any other wonderful natural language processing (NLP) tool that can save me a bit of time on these tasks? Would I let AI write an entire text for my work? A THOUSAND TIMES YES!!! And I say this as someone who loves writing (I even briefly worked at a newspaper) but also comprehends the limitations of NLP technology (I do research on this, after all). The matter is as simple as this: my time is limited, there is way too much to do and I need to prioritize. And surely, any technology, that can increase productivity and give me and my fellow scientists more time for hands-on science, education, outreach and a personal life is to be embraced wholeheartedly.
Before anyone freaks out let me be clear that I will not let AI write my research articles and grant proposals in full. These two types of text are obviously much too important and also require deep scientific insights and knowledge of my specific work that the NLP tools of today do not yet have (and please note the “yet” here). But I see absolutely no problem in using AI-based natural language processing tools, whether it is ChatGPT or others, for text polishing/rewriting/shortening of scientific articles or grant proposals. These tools can also help write short general introductions and summary paragraphs and generate ideas for how to structure the larger text. And when it comes to less essential texts I will hand off some tasks entirely to AI. I truly see no issue with letting an AI model write a generic description of my research field for my website, for example. Will I have to carefully review and potentially revise anything written by AI? Of course. But I also do that with anything written by my students.
Executive summary: Are AI tools for natural language processing impressive? YES. Are they flawless? NO. Are they useful? ABSOLUTELY!
p.s.
I did not cover any of the ethical/legal issues and security risks of large language models here but that does not mean that they should be ignored. Plagiarism is not ok. Made up citations as provided by ChatGPT are really really bad. So, we do need clear standards for ethical and sustainable AI and legal regulations for this technology and an open and informed societal discussion to shape them.
Disclaimer: Besides a spell checker no computational tool was used to compose this article. Sometimes I enjoy writing the old-fashioned way.
