Co-Evolution of Scientific Text in the Age of Human-AI Collaboration

Pontus Wärnestål
6 min readJun 14

As human-AI collaborations continue to evolve, the way we create, consume, and interpret texts is poised for significant transformation. This shift has the potential to revolutionize the way we engage with content, particularly in the realm of scientific texts. In an era where traditional, linear and static “papers” are being questioned, it is crucial to explore how the format of texts will change due to the symbiotic relationship between humans and AI, such as large language models and GPTs.

Here are some possible ideas of how scientific texts could change, reaping the benefits of AI-powered possibilities.

I. Multi-Layered, Interactive Texts

There is really no reason to hold on to static, linear texts anymore. With digital tools, texts can be transformed into dynamic, multi-layered, interactive documents. Scientific texts are no longer a one-size-fits-all, but adapt to the expertise, preferences, and context of the reader. AI’s role in generating context-aware content allows for enriched visualizations and engaging interactivity. For instance, a scientific article on genome sequencing could adapt its presentation layer based on the reader’s knowledge level — a student might receive more explanations and glossary links, while a geneticist views the raw data and complex analyses.

Similarly, hypertextual connections made possible by AI can link ideas, sources, and supplementary information in a seamless way, encouraging holistic understanding of complex concepts. Imagine reading a climate change report where each referenced study is just a click away, and interlinked concepts such as greenhouse gases, carbon footprints, and global warming are visually mapped out. This would help in the peer review process as well.

Lastly, real-time updates and versioning powered by AI ensure that scientific texts are always up-to-date. A paper on the latest COVID-19 variant, for example, could automatically update with new research findings, ensuring the reader always has the most current information.

So, in summary, multi-layered interactive texts could have these characteristics:

1. Context-aware content: Scientific texts will evolve from linear, static formats into dynamic, multi-layered documents. These texts will adapt their presentation based on the reader’s expertise, preferences, and context, offering tailored explanations, visualizations, and interactivity.
2. Hypertextual connections: The integration of AI-generated hypertext will enable seamless navigation between related ideas, sources, and supplemental information. This will facilitate a deeper understanding of complex concepts by providing immediate access to contextual information.
3. Real-time updates and versioning: AI-powered systems will facilitate the continuous updating and versioning of scientific texts, ensuring that readers always have access to the most up-to-date information and insights.

A photographic rendering of a young houseplant against a neutral background, seen through a refractive glass grid and overlaid with an single neuron from a neural network diagram.
Image by Alan Warburton / © BBC / Better Images of AI / Plant / CC-BY 4.0

II. Enhanced Comprehension and Accessibility

AI collaboration plays a pivotal role in breaking down the barriers of complexity and technicality in scientific content. AI-assisted writing tools can generate simplified language and summaries for intricate ideas, bringing scientific discourse closer to the public. For example, an AI tool could generate a layman’s summary of a technical neuroscience text, making it comprehensible for non-scientists. “Papers” that adapt to the reader could be viewed as a democratization of knowledge.

Furthermore, the shift towards multimodal presentation leverages visuals, audio, and interactive elements to cater to diverse learning styles and preferences, enhancing comprehension making the texts more accessible. A physics article on quantum mechanics, for example, could provide interactive 3D models or simulations to visualize abstract concepts.

Lastly, machine-readable metadata integrated into texts can expedite AI processing and interpretation of information. A research paper on AI, when embedded with such metadata, could be rapidly scanned, summarized, and cross-referenced by other AI systems, enhancing knowledge dissemination and interconnectivity.

Enhanced comprehension includes:

1. Simplified language and summaries: AI-assisted writing tools will help authors generate simplified language and summaries for complex ideas, making scientific content more accessible to a broader audience.
2. Multimodal presentation: The collaboration between humans and AI will give rise to more engaging, multimodal content that incorporates visuals, audio, and interactive elements. This will enhance comprehension and cater to diverse learning preferences.
3. Machine-readable metadata: Scientific texts will be designed with both human and AI readers in mind, incorporating machine-readable metadata that facilitates the rapid processing and interpretation of information by AI systems.

A laptopogram displaying a dataset as cloud-like clusters of black blobs on a neutral background. There are three larger collections, almost resembling a map, with some data points leaking out into the negative space.
Philipp Schmitt / Better Images of AI / Data flock (digits) / CC-BY 4.0

III. Collaborative Authorship and Feedback Loops

The solitary academic working in isolation is outdated. Authorship should not be a solitary endeavor. Furthermore, AI tools are becoming more like co-authors, providing real-time suggestions, insights, and revisions. Consider a team of researchers drafting a paper on an AI model’s performance; an AI co-author could provide instant statistics, graphs, and revise language based on best writing practices.

AI systems can also facilitate feedback loops by collecting and synthesizing reader feedback. A paper on mental health interventions, for example, could evolve over time, integrating reader comments, new research insights, and even feedback from clinicians implementing the interventions. Indeed, the dynamic evolution of scientific papers based on reader feedback and real-time insights poses unique challenges to maintaining academic rigor and quality. It also necessitates a shift in how researchers are evaluated and credited for their work. For example, here are a few points worth thinking about:

a. Establishing Protocols for Revision: As scientific papers become more fluid, protocols for incorporating feedback and updates need to be established. These might include a thorough review process for any suggested changes, maintaining a version history to keep track of alterations, and ensuring transparency about who made changes and when.

b. AI-Assisted Quality Assurance: AI tools could assist in maintaining quality by screening updates for accuracy and relevance, checking against existing databases of scientific knowledge, and even predicting the impact of updates based on patterns in past successful revisions. AI can also help detect and mitigate biases in the feedback, ensuring a balanced representation of views.

c. Re-imagining Metrics of Success: The conventional metrics of academic success, often focused on the quantity of new papers produced, will need to evolve. Researchers could also be credited for the ongoing development and refinement of existing work. The impact of a paper could be measured not just at the point of publication, but over its lifecycle. New metrics could include the number of updates, the extent of interdisciplinary collaboration, or the paper’s adaptability over time in response to feedback and new findings.

The peer-review process can also be significantly streamlined by AI, enabling more efficient, transparent, and unbiased evaluations. AI systems could pre-screen submissions, check for plagiarism, and ensure adherence to guidelines, making the peer-review process faster and more reliable. This would bring science closer to real-time discovery and discourse. To maintain academic rigor, significant updates or additions to a paper based on feedback could undergo a mini peer-review process. This would ensure that any new content is vetted by experts before it becomes an official part of the document.

Collaborative authorship and feedback loops are all about:

1. Dynamic collaboration: Human-AI collaboration will enable the co-creation of texts in real-time, with AI systems offering suggestions, insights, and revisions based on a deep understanding of the content, context, and goals.
2. Feedback loops: AI systems will help authors gather and incorporate feedback from readers, enabling continuous improvement and refinement of content. This feedback will also inform the AI’s understanding of the subject matter, creating a virtuous cycle of learning and adaptation.
3. Open peer review: The integration of AI into the peer-review process will streamline the evaluation of scientific texts, enabling more efficient, transparent, and unbiased assessments.


The future of text formats in the age of human-AI collaboration holds immense potential for transforming the way we create, share, and engage with scientific content. Through multi-layered, interactive texts, enhanced comprehension and accessibility, and collaborative authorship, we can expect a new era of scientific communication that transcends the limitations of linear, static documents. This co-evolution of text formats will foster a deeper understanding of complex ideas, facilitate interdisciplinary connections, and democratize access to scientific knowledge.

I am eager to hear your thoughts on how we can update and improve the academic writing process and system. Please let me know what you think.

Pontus Wärnestål

Head of AI Design and Innovation at eghed. Deputy Professor (PhD) at Halmstad University (Sweden). Father of two. I ride my bike to work.