Artificial intelligence and science communication: a new era unfolds

Current uses of AI in science communication

Although many experts remain cautious about the use of AI in science communication, several successful examples already show its potential. Take Duncan Yellowlees, for example — a science communicator with dyslexia. He relies on Otter.ai to convert his speech into text before using ChatGPT to structure his ideas into blog posts. As he puts it: “I find writing long text really annoying, so I speak it into OtterAI, which converts the speech into text. It’s my thoughts, but I haven’t had to write them down.”

Another example comes from Professor David Markowitz at Michigan State University. His study found that AI-generated summaries of scientific articles were actually easier to understand than those written by the researchers themselves. The participants found the simpler texts easier to understand, which in turn boosted their trust in scientists.

Today, researchers who embrace AI are using it to perform all kinds of tasks: drafting articles and abstracts (ChatGPT), summarizing scientific papers (Perplexity, Semantic Scholar), creating images (Midjourney, ChatGPT image generator), and translating articles (ChatGPT). Some are even taking it a step further by translating video conferences with synchronized dubbing (HeyGen). A powerful tool which strengthens international collaboration among researchers.

Limits and risks of using AI in Science

Despite its potential, AI is not without flaws. AI tools do not possess a human understanding of the context. Instead, it predicts the next most likely word. This method can lead to mistakes, known as “hallucinations” — seemingly convincing but incorrect responses which are not grounded in real data. According to studies conducted in 2024, the hallucination rate in abstracts produced by ChatGPT-4o is around 40%, though it is still better than competing AI models which can reach a staggering 70% error rate.

AI-generated images, created with tools like Midjourney, are increasingly used in science communication. However, their lack of reliability remains a major concern. These illustrations can contain physically impossible elements and be misleading, even while appearing credible. For example, a DNA structure generated by an AI might look visually impressive but include serious structural errors, or a prokaryotic cell might be mistakenly depicted with a nucleus even though they have none in reality. Even animal depictions can go wrong: in an image generated by Midjourney, a hammerhead shark’s eyes are misplaced. They are in fact located at the far ends of its cephalofoil — the hammer-shaped extensions of its head — not above its mouth.

Another important issue concerns copyright. In most countries — including France and the United-States, at least for now — works created by generative AI without meaningful human creative input are not protected by copyright and fall within the public domain. While this has little impact on materials such as slides and posters, the lack of copyright protection can become a major problem for scientific journals and books, where legal protection is crucial.

Can AI really improve science communication?

AI is making the editing of scientific papers and popular science writing faster than ever. Models such as ChatGPT or Claude 3.7 Sonnet improve grammar and style, generate summaries, and adapt the content for different audiences. They help to spot inconsistencies and to write annotations. However, as AI does not always grasp the context, it can lead to oversimplified content and a loss of depth in scientific reasoning. Excessive automation could also impoverish the originality of texts and decrease critical thinking skills.

Tools like Perplexity and Semantic Scholar speed up information retrieval by analysing large datasets. They identify relevant articles and produce concise summaries. However, their algorithms can overlook important nuances or provide outdated or inaccurate data. A study published in March 2025 reports that 37% of Perplexity’s citations are erroneous, despite it being one of the best tools for these tasks. Therefore, thorough double-checking generated data is still 100% necessary.

The increasing use of AI in the field of science is expected to increase the amount of publications and research. However, quantity and quality don’t always align. The surge of articles could spread unreliable information, make it harder to detect biased data, and require stricter oversight of AI-generated content. Currently, regulations require researchers to disclose the use of AI in research, but compliance largely depends on the authors’ integrity. To address these challenges, new AI tools are already in the works to detect erroneous data and analyse auto-generated content.

Discover our Generative AI and Communication training program

AI and science communication in the near future

As we have seen, AI is transforming the way scientific knowledge is shared. In the coming years, it is expected to be increasingly used to develop interactive platforms that tailor scientific content for different audiences.

But AI isn’t just transforming communication — it’s reshaping science itself. Experts predict that by 2030, AI will become an essential component of both scientific research and dissemination. McKinsey forecasts report that across all sectors, employees believe AI could take over up to 30% of their tasks within a year — and scientific professions are no exception.

Dario Amodei, founder of Anthropic, views science, particularly biology, as a key area for AI application. He foresees that AI will eventually not only analyse data but also act as a “virtual biologist” capable of performing the repetitive tasks of researchers.

More than 40 researchers, including scientists from EPFL in Switzerland, are already hard at work building an AI-driven virtual cell — an advanced neural network designed to model the behavior of molecules, cells, and tissues in various states. This system could simulate biological functions, predict how the body reacts, and even help to design new treatments.

Another factor to consider is the rapid pace of progress in generative AI tools. A major evolution is already underway! At the end of 2024, Meta unveiled LCMs (Language-Concept Models), a new generation of language models that operate at the level of concepts rather than individual words. Unlike LLMs, such as ChatGPT, which generate text based on the probability of the next word, LCMs manipulate more complex units, juggling whole sentences and abstract ideas. They are better at capturing the meaning of information and at processing longer texts more effectively.
For now, these models are not publicly available, but it is only a matter of time. These technologies are expected to make scientific communication more accessible, personalized, and effective than ever before.

AI: a new ally for scientific and technical communicators

AI is already transforming scientific communication. It automates routine tasks, accelerates information retrieval, and facilitates writing. However, despite its advances, it must be thoughtfully used to ensure that the content produced remains reliable and relevant.

Experts believe that AI’s role in science communication will only grow stronger, but that critical reviews will remain essential. Striking the right balance between automation and human input will be key, as AI cannot replace intuition, creativity, or deep contextual understanding. With rigorous oversight and an awareness of its limitations, AI can become an invaluable ally, helping scientific communicators to save precious time while delivering high-quality content.