AI-Powered Virtual Labs: New Framework Aims to Transform Science Museums

Science museums have long served as gateways to scientific curiosity, offering visitors hands-on encounters with phenomena that would otherwise remain confined to textbooks or research facilities. Now, a team of researchers in Mexico has proposed a structured framework for bringing artificial intelligence into these informal learning spaces through virtual laboratories, an effort that could reshape how millions of visitors engage with sustainability science. The study, published as an open-access review in Discover Sustainability, emerges from the early conceptual design of LúmiaLab, an interdisciplinary virtual laboratory being developed in connection with the Museum of Environmental Sciences in Guadalajara, Mexico. Rather than presenting a finished product or empirical evaluation, the work delivers something arguably more foundational: a preliminary conceptual map of what AI-supported virtual laboratories in museums should consider before a single line of code is written.

The research team, led by Jorge Sanabria-Z of the School of Architecture, Art and Design at Tecnológico de Monterrey, together with Pamela Olivo, Amadeo Arguelles of the Instituto Politécnico Nacional, and Inna Artemova of the Universidad de Guadalajara, set out to address a conspicuous gap. Science museums are increasingly positioned as venues for non-formal sustainability education, yet systematic frameworks for integrating AI-supported virtual laboratories into these environments remain scarce. Virtual laboratories themselves are not new; simulations and remote-access experiments have been used in formal education for years. What distinguishes the new framework is its focus on the museum context, where audiences are heterogeneous, visits are voluntary and often brief, and the goals blend entertainment, education, and civic engagement around socioecological challenges such as climate change, biodiversity loss, and resource management.

Methodologically, the study took a Delphi-oriented approach, a technique traditionally used to build consensus among experts through iterative rounds of reflection and feedback. In this case, the process was internal and designer-led, combining two rounds of multidisciplinary reflection with synthesis assisted by a large language model. The authors used GPT-4, developed by OpenAI, as a methodological instrument to synthesize expert inputs into preliminary dimensional frameworks and to propose supporting AI tools and references across the two iterative rounds. Crucially, the team documented that all LLM-generated outputs were critically reviewed, discussed, and refined by the expert panel, with references independently verified against academic databases. This transparent handling of AI assistance reflects a growing awareness in the research community that generative models can accelerate synthesis but must remain under rigorous human oversight, particularly when the output is meant to guide institutional design decisions.

From this consensus process, the researchers identified thirteen dimensions that collectively characterize AI-integrated virtual laboratories in museum settings. These are contextual understanding, operational sustainability, research support, educational design, audience personalization, technological infrastructure, physical-digital interaction, operability, ethics, participation, accessibility, interculturality, and international collaboration. Each dimension is paired with possible AI capabilities and intended outcomes, creating a matrix that institutions can use as a checklist during planning. The breadth of the list is telling: it signals that building a virtual laboratory is not merely a software engineering problem but a socio-technical challenge that touches pedagogy, ethics, cultural inclusion, and long-term institutional viability all at once.

Several of the dimensions deserve closer attention because they highlight tensions unique to the museum environment. Contextual understanding refers to the system’s capacity to ground scientific content in the local ecological and social realities of visitors, something particularly relevant for a museum in Guadalajara, a metropolitan region facing pressing water and environmental pressures. Audience personalization leverages AI’s ability to adapt explanations, difficulty levels, and interactive scenarios to individual visitors, from schoolchildren encountering a concept for the first time to adults with prior scientific training. Physical-digital interaction addresses the hybrid nature of the museum experience, where virtual experiments must complement rather than replace tangible exhibits, and where the flow between gallery floors and digital interfaces determines whether the technology feels integrated or intrusive.

Operability and technological infrastructure form the practical backbone of the framework. Virtual laboratories powered by generative AI demand reliable connectivity, computational resources, and maintenance plans that many cultural institutions, especially in low- and middle-income countries, must budget for carefully. Operational sustainability extends this concern into the future, asking how a laboratory will be updated, staffed, and financed beyond its launch. The inclusion of these dimensions in a conceptual framework is a quiet but important corrective to technology-driven enthusiasm, reminding designers that a dazzling AI demonstration that fails after six months serves neither the museum nor its audience. The framework’s emphasis on operability suggests that ease of use for museum educators, not just visitors, was treated as a design priority.

Ethics, accessibility, and interculturality round out the human-centered dimensions, and their presence reflects contemporary debates about the deployment of AI in public-facing institutions. Ethical considerations in this context include data privacy when systems personalize content, the transparency of AI-generated explanations, and the risk of presenting algorithmic outputs as authoritative scientific truth. Accessibility demands that virtual laboratories serve visitors with disabilities, a legal and moral requirement that is often easier to state than to implement in immersive digital environments. Interculturality pushes further, asking how a laboratory can respect and incorporate Indigenous and local knowledge systems alongside Western scientific frameworks, a question of particular resonance in Mexico, where the intersection of diverse knowledge traditions and formal science education is an active area of scholarship and policy.

Participation and international collaboration connect the framework to the broader movement of citizen science, in which members of the public contribute observations, data, and even research questions to genuine scientific projects. The study explicitly frames AI-supported virtual laboratories as platforms for citizen science participation oriented toward socioecological challenges, meaning that visitors would not merely consume simulations but potentially engage with real environmental data and contribute to ongoing research efforts. International collaboration, meanwhile, acknowledges that socioecological phenomena such as migration, drought, and pollution cross borders, and that museums can act as nodes in transnational networks of shared virtual infrastructure and open educational resources. The framework was supported by the Secretaría de Ciencia, Humanidades, Tecnología e Innovación (SECIHTI), Mexico, under its 2025 Call for Humanistic Research, grant reference IH-2025-G-125, awarded to the LúmiaLab project.

The authors are notably candid about the limits of their contribution. The framework is contextually grounded in the LúmiaLab project and the Museum of Environmental Sciences in Guadalajara, but it has not yet been independently or empirically validated. It is offered as a structured theoretical-practical reference for subsequent co-design, piloting, and external validation, rather than as evidence of educational effectiveness or generalizability. This honesty matters in a research landscape where AI-in-education claims often outrun the data. The team also disclosed a second form of AI assistance: a large language model, Claude, developed by Anthropic, was used during manuscript preparation to assist with structural organization, drafting of specific sections, and language editing, based on the authors’ conceptual directions and prior drafts. The authors reviewed, verified, and revised all such content, and no AI tool is listed as an author, with the researchers taking full responsibility for the scientific content and the accuracy of references.

For museum professionals, educators, and technologists watching the rapid evolution of generative AI, the study offers a sober starting point rather than a spectacle. Its thirteen dimensions function less as a blueprint than as a set of questions that any institution hoping to build an AI-supported virtual laboratory should be able to answer: Who is the audience and what do they need? How will the system respect ethical boundaries and cultural diversity? Can the institution sustain the technology over time? And how will anyone know whether the laboratory actually improves learning or engagement? By framing these questions before implementation, the Mexican research team has contributed a reference point for a field that is being built in real time, one in which the museum, that most public of scientific spaces, may become a proving ground for how artificial intelligence and human curiosity can productively meet.

Subject of Research: Conceptual framework for AI-supported virtual laboratories in science museums for sustainability and citizen science education

Article Title: Guidelines for AI-supported virtual laboratories in science museums

Article References: Guidelines for AI-supported virtual laboratories in science museums. (n.d.). https://doi.org/10.1007/s43621-026-04907-x

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04907-x

Keywords: virtual laboratories, science museums, artificial intelligence, generative AI, citizen science, sustainability education, non-formal education, socioecological challenges, LúmiaLab, Guadalajara, Delphi method, educational design