Scientific accuracy is essential. But accuracy alone does not guarantee understanding.
A technically correct graph can still obscure its message. A research poster can contain excellent results and remain difficult to navigate. An infographic can look polished while distorting the science behind it. And an AI-generated summary can save time—or introduce errors that no one notices.
This tension between scientific rigor and clear communication was at the heart of the H2START Masterclass “Science in Sight: Visual Communication and AI Tools for Education and Science,” held on 24 April 2026 at Trakia University in Stara Zagora.
Led by H2START Prof. Ani Zlateva and science communicator Petar Teodosiev, it examined how people perceive information, how visualisation shapes interpretation, and how researchers can use tools such as NotebookLM, Canva, ChatGPT and Codex without surrendering scientific judgement.
We do not see information neutrally
Prof. Zlateva introduced the cognitive principles behind visual perception and, drawing on principles from Gestalt psychology, showed how people organize information through proximity, similarity, continuity, closure and figure–ground relationships.
The session connected these principles with scientific practice: how audiences read diagrams, identify hierarchy, follow processes and distinguish the main message from background detail.
The conclusion was clear: effective visual communication begins not with decoration, but with an understanding of how attention and perception work.
A clearer image is not automatically a truer one
The Masterclass also addressed visual ethics and the balance between aesthetics and scientific accuracy.
Through examples of photographs, enhanced images, infographics, charts, tables and posters, participants examined how visual choices can clarify evidence—or unintentionally distort it. The discussion highlighted a recurring problem in science communication: adding more information does not necessarily improve understanding.
For hydrogen research, this balance is especially important. The same content may need to serve researchers, students, industry, public authorities and wider audiences without losing its scientific meaning.
From scientific sources to usable knowledge
The practical sessions demonstrated how AI-supported tools can help researchers organise source material, extract key ideas and develop summaries, educational resources and presentation structures.
NotebookLM was used as an example of working from defined scientific sources rather than relying on unverified generated content. The focus remained on traceability: the final output must stay connected to the evidence behind it.
The outcome was not a promise of automatic communication, but a practical workflow in which AI assists with structure and speed while the expert verifies accuracy, context and relevance.
One result, several professional formats
Petar Teodosiev demonstrated how scientific content can be adapted into presentations, posters, infographics, learning materials and outreach visuals.
Using Canva, he covered the practical reuse of one source across several formats, including templates, image adjustment, infographic development and consistent visual identity.
The objective was not to turn researchers into designers, but to help them create clearer and more usable materials without weakening the logic of the research. This capacity directly supports H2START’s mission to connect science, education, industry and society.
Building useful AI tools without programming
The final session showed how researchers and educators can define their own AI-assisted tools through ChatGPT and Codex.
Examples included a flashcard generator, a structured academic reviewer and a scientific literature search workflow using established publication databases. The emphasis was on precise requirements: what sources the tool should use, what it must not invent, and how results should be structured.
AI accelerates. The expert decides.
One principle connected every part of the Masterclass: AI must support expertise, not replace it.
Using AI to draft a review that a researcher verifies is fundamentally different from delegating the academic decision to the system. Using it to identify possible weaknesses can improve a paper. Using it to produce work that is presented as original undermines both learning and academic integrity.
AI can accelerate the process. Responsibility remains with the user.
Making science visible is part of making it useful
“Science in Sight” treated visual communication not as decoration added after the research is complete, but as part of the scientific process itself.
A strong figure can reveal a pattern. A clear diagram can expose a weak assumption. A well-structured poster can open a conversation with a future collaborator. A responsible AI workflow can free time for analysis rather than replace it.
For hydrogen research and innovation, this capability matters. The field connects complex technologies with industrial decisions, public investment, education, infrastructure and policy. Progress depends not only on what scientists discover, but also on whether others can understand the evidence well enough to act on it.
That is why making science visible is part of building the research culture, skills and partnerships that the H2START Centre of Excellence needs.
