AI-Assisted Knowledge Translation in One Health: Ethical Challenges for Participatory Governance

AI-assisted knowledge translation streamlines research communications for One Health initiatives addressing antimicrobial resistance and zoonotic diseases. What is gained in operational efficiency may come at the expense of authentic participation. By determining which evidence becomes visible and actionable, AI systems restrict epistemic pluralism, sideline essential stakeholder knowledge, and entrench existing disparities in global health.

Artificial Intelligence (AI) as Knowledge Translation (KT) Infrastructure in One Health. Image generated with ChatGPT by OpenAI.

Knowledge translation (KT) is crucial in collaborative and multi-sectoral initiatives aimed at addressing challenges related to antimicrobial resistance (AMR) and zoonotic disease outbreaks. The One Health approach, which informs research collaborations and public health interventions targeting significant global health threats, requires effective integration and strategic coordination across human health, animal health, environmental health, and health policy systems. This interdisciplinary domain is complicated by the coexistence of diverse institutional structures and heterogeneous knowledge traditions, where participatory research governance has increasingly been recognized as an essential operational principle. Additionally, high-level frameworks and international bodies have embraced inclusivity, stakeholder engagement, and equity as core normative dimensions of ideal governance for addressing antimicrobial resistance.

AI-assisted knowledge translation often contradicts foundational values of openness, transparency, and pluralism in the pursuit of efficiency. AI-powered research tools are fundamentally transforming how knowledge is generated across disparate systems, often making autonomous decisions that can impact which scientific evidence shapes global health policy and practice. This growing role of AI as a non-human actor in knowledge generation and policy-making has raised concerns, particularly within the One Health domain, where tensions between participatory mechanisms and automated systems are especially pronounced.

AI as Knowledge Translation Infrastructure in One Health

A quick disclaimer before I delve into the analysis of some emerging ethical and epistemic concerns about the quick AI integration in One Health research. This article builds on an invited presentation I gave during Canada's 2026 One Health Conference in Calgary, Alberta earlier this year. My talk was included in a session on the ethical use of AI in knowledge translation and research communications in a One Health context, which was part of a training initiative targeting trainees (graduate students, post-docs, and HQPs) from Canadian Antimicrobial Resistance Network (CAN-AMR-Net) and Canopy, the Canadian One Health Training Program on Emerging Zoonoses. The panel brought together multidisciplinary and complementary expertise on AI ethics, with topics ranging from broader ethical refection on opportunities and risks associated with using AI tools in knowledge transfer and communication to practical concerns surrounding algorithmic bias, reliability, transparency, and authorship. I focused largely on ethical considerations regarding the challenges automated research tools pose to community-based approaches to knowledge creation. My major concern was that the positioning of AI tools as KT infrastructure within the One Health framework has the potential to undermine normative standards for participatory governance. Given the constraints of a short conference presentation, here I expand on the ethical analysis and case studies of AI-powered surveillance dashboards, automated evidence synthesis, and predictive risk models.

It is hardly an overstatement to assert that AI systems do not process data in an unbiased manner and should therefore be subject to rigorous scrutiny by researchers and practitioners before their full integration into a One Health framework. AI tools for health research are currently mediating the synthesis of knowledge and influencing which key issues are prioritized and communicated across health systems. In this context, algorithmic capabilities have naturally become central to knowledge translation and research communication. They also support the development of practical solutions for antimicrobial resistance within the One Health framework.

A recent scoping review of 343 articles has highlighted that AI is increasingly leveraged to tackle AMR and zoonotic diseases. AI systems can support rapid identification of resistant pathogens, improve surveillance and early warning systems, integrate diverse datasets across human, animal, and environmental health, and provide support for drug discovery and antibiotic stewardship. The study, published in One Health Outlook in 2025, has also established major challenges associated with AI models, including data standardisation issues, limited algorithmic transparency, infrastructure and resource gaps, ethical and privacy concerns, and difficulties in real-world implementation and validation. It specifically highlights that opaque "black box" models, inequitable data representation from high-income countries, and unclear legal liability for AI-driven errors present major ethical hurdles, prompting calls for ethical, human-centric governance. These developments illustrate the need to balance accelerated AI adoption in KT with complex ethical challenges of integrating responsible AI into global health governance.

 

Ethical Tensions between Automation and Participation

AI tools offer several significant opportunities for knowledge synthesis and translation. They can accelerate evidence synthesis, improve access to complex knowledge systems, support cross-sector communication, and enable scalable training and decision-support systems. However, there are well-recognized ethical concerns about AI algorithms across various academic disciplines, with most critiques emphasizing their inability to ensure reliability, transparency, bias, and accountability. My contention is that these common ethical failures of AI systems do not fully capture the emerging ethical and epistemic tensions of automating KT processes within a One Health framework.

The deeper issue involves a growing tension between two distinct models of knowledge generation. On one side, participatory governance frameworks emphasize public deliberation, meaningful engagement of stakeholders and local communities as decision-makers, and equitable inclusion of multiple forms of expertise. This plurality of expertise encompasses scientific knowledge, Indigenous and local knowledge, community experiences, and applied policy solutions. Integrating diverse perspectives is crucial. It enhances contextual accuracy and builds public trust, both of which are essential for effective One Health governance. Consequently, participatory knowledge systems play a vital role in ensuring that research contributions are translated into socially legitimate policy decisions.

AI-enabled knowledge translation often prioritizes efficiency and standardization. While AI systems can distill complex evidence into simplified signals, they frequently standardize research outputs to facilitate interpretation and scalability. They also heavily rely on datasets created in the Global North, reproducing structural and power imbalances. Additional challenges arise from the "black box" problem associated with algorithmic decision-making. AI tools tend to obscure uncertainty and diminish epistemic diversity, restricting opportunities for genuine deliberative processes in search of consensus. In certain instances, automated outputs may create an illusion of consensus, even when the underlying knowledge remains profoundly contested.

AI-Assisted Knowledge Translation Tools

These dynamics are already visible in AI-enabled knowledge translation systems deployed across One Health research domains pertaining to antimicrobial resistance (AMR) and zoonotic diseases. Examples include AI-generated evidence summaries for policy, automated outbreak or AMR risk dashboards, generative AI communication tools, and AI-powered training systems. While these tools play an important role in translating complex knowledge, they also raise an important question: "Are these systems supporting participatory knowledge processes, or gradually replacing them with automated authority signals?"

To illustrate this tension more concretely, I briefly highlight three types of KT tools that have gained prominence in global health initiatives for disease monitoring and AMR prevention:

• HealthMap: A free automated, online system that utilizes machine learning to monitor global outbreak signals and generate real-time surveillance maps.

• EPIWATCH: An epidemic intelligence platform that employs natural language processing to detect early signals of disease outbreaks.

• WHO Global Antimicrobial Resistance Surveillance System (GLASS): A global collaborative framework that aims to standardize data collection and analysis of antimicrobial resistance and antimicrobial consumption to tackle drug-resistant infections worldwide.

These examples encompass the spectrum of data processing: from unstructured crowd-sourcing (HealthMap) and computational linguistics (EPIWATCH) to institutional, top-down data harmonization from global ministries (GLASS). AI systems thus significantly enhance capabilities for detecting and monitoring global health threats and can help governments and health agencies build better prevention policies. However, they not only translate complex data into interpretable risk signals but also demonstrate how algorithmic systems can shape what knowledge becomes visible and actionable.

HealthMap

HealthMap, launched by the Computational Epidemiology Lab at Boston Children's Hospital, is a global disease surveillance platform that employs automated data mining and machine learning techniques to monitor emerging disease outbreaks. It aggregates data from a diverse array of sources, including news reports, official public health alerts, and online media. The system effectively translates these heterogeneous signals into real-time outbreak maps and alerts, making it a valuable tool for researchers, public health agencies, and international organizations in tracking emerging infectious diseases, particularly zoonotic outbreaks. From a knowledge translation (KT) perspective, tools like HealthMap have transformative potential for public health decision-making as they can quickly process and synthetize vast volumes of dispersed information into interpretable risk models.

However, this process highlights the participation-automation tension I discussed earlier. AI systems' reliance on algorithmic filtering tends to prioritize signals that can potentially exclude local contextual knowledge. It can also obscure uncertainties inherent in the underlying data sources. Surveillance dashboards like HealthMap automate the identification and prioritization of outbreak data, raising critical questions about how algorithmic filtering influences perceptions of what constitutes a credible risk.

EPIWATCH 

Another tool utilizing AI-driven data collection and synthesis is EPIWATCH, an epidemic intelligence platform developed by the Kirby Institute at the University of New South Wales (UNSW). It employs natural language processing to analyze global media reports in 41 global languages, detecting early signals of emerging disease and outbreak events. The online system gained attention during the early stages of the COVID-19 pandemic, by identifying signals of unusual pneumonia cases before many official alerts were issued. EPIWATCH illustrates how AI can help transform unstructured information from global media sources into structured alerts and risk maps.

While such AI-powered, open-source systems significantly enhance the speed and scale of epidemic intelligence, they cannot circumvent problems associated with algorithmic decision-making. EPIWATCH's heavy reliance on media reporting can introduce geographical and linguistic biases, and algorithmic filtering may influence which outbreak narratives gain visibility within global surveillance systems.

WHO Global Antimicrobial Resistance Surveillance System

Launched in 2015, GLASS seeks to standardize global AMR surveillance by consolidating data from various countries and displaying it through interactive dashboards for researchers and policymakers. This initiative shifts from traditional surveillance methods that rely solely on laboratory data to a more sophisticated analysis of epidemiological, clinical, and population-level information. The objective is to gradually integrate data from AMR surveillance in humans, which includes tracking resistance and the use of antimicrobial medications, as well as addressing AMR in the food chain and the environment.

Country Participation in WHO Global Antimicrobial Resistance and Use Surveillance System. Source: WHO, https://www.who.int/initiatives/glass/country-participation

The system operates through a series of specialized technical modules, serving as a powerful KT tool that transforms complex microbiological data into policy-relevant indicators and visual summaries. These outputs can effectively guide global responses to antimicrobial resistance (AMR). It is important to acknowledge that GLASS is fundamentally an institutional, human-governed framework. As outlined in the WHO GLASS Country Participation guidelines, the surveillance platform relies on active country enrollment, national focal points, and explicit capacity building.

GLASS also carries epistemic risks, particularly from the manner in which data is aggregated, processed, and displayed on interactive dashboards. When automated synthesis modules operate atop deeply uneven global surveillance infrastructures, they inadvertently smooth over structural discrepancies. By converting highly heterogeneous datasets into uniform visual indicators, the platform creates an illusion of comprehensive global consensus. But in many instances it's a false consensus. The standardized dashboard ends up silencing the very local contexts, structural gaps in health systems, and qualitative realities that the human components of the network are striving to highlight.

Responsible AI Integration in Knowledge Translation

AI-assisted knowledge translation in global health research should not be regarded as a neutral infrastructure for knowledge transfer, as the systems themselves play a crucial role in determining which evidence is deemed visible, credible, and actionable. We must ensure that meaningful opportunities for participation are preserved for researchers, practitioners, local communities, Indigenous knowledge holders, and other stakeholders whose insights may not be adequately represented in existing datasets. Instead of replacing deliberative processes with automated authority signals, AI should be designed and governed as a tool that enhances these processes.

Responsible integration of AI research tools within the One Health framework for antimicrobial resistance (AMR) and zoonotic diseases requires more than improving the technical accuracy and transparency of AI systems. It also requires careful examination of the power structures embedded in the production and interpretation of knowledge. In this regard, responsible AI in One Health transcends the governance of algorithms; it encompasses the governance of knowledge itself: determining whose knowledge is valued, who holds the authority to interpret it, and who is included in decision-making processes that affect communities and ecosystems.

The next generation of One Health researchers—graduate students, post-docs, and highly qualified personnel (HQPs) are at a crucial junction to navigate radical transformations within the research ecosystem. To prevent automated authority from completely overtaking participatory governance, future health leaders must recognize emerging epistemic challenges. It is no longer enough to be technically proficient in running AI models or deploying data pipelines. Trainees must be equally skilled in community engagement and critical data literacy. The next generation must learn to interrogate the data architectures they inherit, asking not only questions about the accuracy of predictive models but also whose reality was excluded from it. Only by fostering this critical reflexivity can we ensure that AI remains a tool that amplifies diverse human expertise rather than a gatekeeper that suppresses it.

Dr. Kalina Kamenova is an engaged scholar whose work focuses on the ethical, legal, and social implications of emerging biomedical technologies, including AI in healthcare and participatory governance. She is the Founder and Research Director of the Canadian Institute for Genomics and Society (Genomics4S).