Communicating science means making complexity understandable — but what happens when clarity comes at the expense of uncertainty?
Read about current trends in science communication and science communication-related activities in the SciComm Close-Up, a monthly deep dive into science communication from the University of Maryland School of Graduate Studies’ Science Communication (SciComm) certificate program. The SciComm Close-Up is part of the program’s monthly newsletter SciComm Spotlight. Sign up for monthly updates on SciComm.
Has artificial intelligence (AI) ever answered one of your prompts using the word “maybe”? I asked my AI of choice, Perplexity, to count how many times it had used the words “maybe” and “perhaps” in several chats relating to scientific research. The answer was zero.
So why does AI, especially when debating complex topics with unknown answers such as science, not use the word "maybe"? It's because these large language models (LLMs) prefer direct, authoritative answers and statements. This applies to both how they give answers and which sources they use and cite to answer your questions.
I was in academia for 10 years before becoming a scientific writer, communicator, and marketer. As I studied, it was clear to me that the jobs of scientists were to understand the world as it was, but to do that, you have to find questions that didn't have a concrete answer yet. However, as a science communicator, I'm expected to make things clearer for the public. This often means shorter, direct statements with little room for uncertainty and nuance.
If you communicate science by writing, you probably have come across the Flesch-Kincaid Grade Level score, which tells you what U.S. school grade level your readers need to understand your text. And, to write more understandably, you’ve probably been told to write for lower grades.
The Flesch-Kincaid scale was developed in 1975 by the U.S. Navy to ensure that draftees could understand technical information quickly. By 1978, it was the official U.S. Department of Defense standard for assessing readability and determining the comprehension that the public had of complex materials.
And for years now, science communicators and writers have been told to “write simply,” using this grade-level scale as a benchmark for complexity.
But after years of making materials simpler to understand, with the expectation that this would increase literacy and comprehension, we've seen that the results are the opposite. About 54 percent of 16- to 74-year-olds in the U.S. read below a sixth-grade level. And U.S. literacy scores declined between 2017 and 2023.
Yes, writing simply and clearly is something we should do. People want a clear picture of the latest scientific developments. But not challenging the public with new words and concepts leaves them unprepared to deal with the full complexity of science, with its untold mysteries.
It all gets more complicated. When LLMs were invented, they were trained with mostly public material hosted on the Internet: pieces and content written following the advice of simplifying language. As a result, AI both prefers and outputs extremely simple, authoritative language.
For scientists and communicators, this creates a problem. Science pursues the answers to questions that don’t yet have them. Uncertainty is the norm.
How does a protein fold the way it does? Why does a specific molecule trigger an immune response? How does a drug actually kill a pathogen? These are complex questions whose answers often boil down to "we are not sure."
The difficulty of navigating uncertainty and complexity through words appears in every facet of my work, regardless of whether I am communicating with the general public, machines, or scientists. Let me give you a few examples.
The General Public
“So is this molecule good or bad?” my stepmom asks as she finishes her dessert. We're having lunch at a restaurant, and she decided to ask me about a newspaper article that she read, which describes how the molecule glutathione plays a key role in cancer progression. But glutathione is an antioxidant lauded for its health benefits. Both these statements together prompted confusion in my stepmother, because, after all, if the molecule is good for your cells, how could it speed up cancer?
I tried my best to explain that every cell in our body struggles with oxidative stress, and this molecule helps them with that. At the same time, cancer cells, which divide rapidly and have massive energy consumption, are under heavy oxidative stress, and glutathione helps them survive. She looked at me with annoyance because what she really wanted to know was if she should look for this kind of molecule in supplements and foods or if she should try to avoid it.
I had no direct answer for her.
The Machines
I look to my left and read the client's AI-generated instructions for the product's web page. They ask me to write a direct, authoritative answer to a frequently asked question. Specifically: “start with a direct statement and then follow with context. Make sure the total answer is 50 to 70 words.” Clear instructions.
I had done just that, and the company had sent my draft for review to their internal team of scientists.
I look to my right, and read the expert’s comments. They didn't love what I wrote. They thought it was incomplete and not true enough. Their rewritten answer had 120 words, way over the instructions' limit.
Frequently asked questions, or FAQs, are something more and more pages include because AI heavily pulls from these. When someone asks them a query that matches a frequently asked question on a page, they often draw from the FAQ’s answer to reply to the prompt.
I knew this company was trying to get cited by AI, but how could I give a short authoritative reply to this question when even their internal team of scientists agreed that there was more nuance than my answer reflected? I sighed, stopped the time tracker, rubbed my eyes, and put the PC on sleep mode. I needed to think about this one.
The Scientists
“The founders of the company are amazing,” the woman on my screen said, “but they are all scientists and … they have this problem. When a new possible customer joins the sales meeting, they go straight to the pitch deck. They barely even ask, 'How are you?' and then it's all science jargon from there.”
The woman was one of the few employees of a new green manufacturing company. She told me the product was world-class, but they really struggled to land clients. Now it was clear why.
I smiled at her and replied: "OK. I can help with this."
Final Words
Science communicators have always had to work between the scientific jargon and the often lower scientific literacy of the general public/non-scientist audiences. With AI, we have introduced a new problem: overly authoritative statements that don't really show the nuance and uncertainty behind science and its progress.
I know some who have not yet realized the challenge that this type of language poses for science communication.
I myself find it challenging to navigate the treacherous waters of language complexity to preserve both the main takeaways of science but also its uncertainty and complexity. I don't yet have a solution for how to do this consistently, but I would argue for caution when trying to make your writing more authoritative so LLMs use it, and when using AI outputs for any kind of scientific communication.
Our job of making science understandable does not end at communicating something. Challenging the audience to expand their scientific literacy is just as important for the long-term understanding of science and the continuation of scientific progress. Because, if we cannot keep facing the uncertainty, we can never find the answers we are looking for.
PS1: Further reading:
- Gal Yona, 2024. Can Large Language Models Faithfully Express Their Intrinsic Uncertainty in Words? ArXive.
- Belém et al. 2024. Perceptions of Linguistic Uncertainty by Language Models and Humans. Association for Computational Linguistics.
PS2: I thought I would try to answer some questions you may have after reading all this.
Frequently Asked Questions (FAQs)
How should I communicate scientific uncertainty?
It depends. Do your best!
At what reading level should I write so my audience understands me?
Depends on your audience.
Should I try to use complex language to improve my audience’s literacy or use simpler language, so they understand what I am trying to say?
Yes.
Darío Sánchez Martín, PhD, is a scientist and science communicator with a background in biotechnology and cell and molecular biology. A former university lecturer, researcher, and R&D consultant, he is the co-founder of Helixa Communications, where he works with biotech, pharmaceutical, and other life science organizations to turn complex scientific ideas into clear, engaging content for different audiences.