The Future of Research is an SCDI research initiative that explores how generative and agentic AI are changing the way scientific knowledge is produced, validated and funded. It develops infrastructure through which scientific rigor can reach the places where data is actually generated.
2026
Two documents have organized the modern research enterprise: the scientific article and the grant application. Both have worked as boundary objects. Each is flexible enough to serve universities, funders, publishers and researchers in very different ways, yet stable enough to coordinate a global system. Embedded Science starts from the observation that AI is taking away the effort threshold these objects depended on. When a formally adequate manuscript or application can be produced at negligible cost, the documents stop being credible signals of quality, competence and original contribution. Rather than defending the old system or automating it uncritically, the initiative takes a third path. Embedded Science is a mode of knowledge production in which rigor travels through validated infrastructure, instead of depending on a credentialed researcher being present at every analytical step. The researcher’s role shifts from being the interpretive instrument to validating the instrument and being accountable for what it produces. The initiative combines conceptual work on the future of research publishing and funding with the design and validation of computational infrastructure for AI-augmented qualitative and quantitative analysis.
Embedded Science, generative AI, agentic AI, knowledge production, boundary objects, research funding, scholarly publishing, research infrastructure, provenance
The research system has long relied on proxies for quality rather than direct assessment. Our own analysis suggests that more than 70 percent of time allocated to research is spent on activities that are not research: writing applications, reviewing, packaging findings into article formats and navigating editorial processes. Generative AI has made this problem acute. Article volumes have more than doubled since 2010 while the share of truly disruptive contributions has fallen. Journals are struggling to find reviewers, and funders are seeing sharp increases in applications without matching gains in quality. Metrics such as the h-index lose what little discriminatory power they had once both publications and citations can be inflated cheaply. Meanwhile, most organizations remain structurally excluded from producing trustworthy knowledge about their own practice.
The initiative works along three connected lines. The first is infrastructure. We design computational infrastructure for AI-augmented analysis. It uses a graph database to represent codes, categories and memos, and a provenance architecture based on the W3C PROV standard, so that every prompt, model inference, coding decision and human intervention can be logged and audited. The second is validation. We test this infrastructure stage by stage against established trustworthiness criteria, systematically comparing automated analysis with human analysis. We report openly what provenance does and does not allow a reader to check. The third is institutional design. This covers merit systems that credit datasets, protocols and provenance architectures as scientific contributions, and an epistemic commons of open standards and validated methods. It also covers new models for research funding, in which funders move from passively running open calls to actively orchestrating research. This line of work is being put into practice together with Vinnova and the Swedish Research Council (VR), where we are helping to redesign the systems through which Swedish state research funding is applied for and administered.
The initiative has developed its core concepts, Embedded Science and the Sciencecene, in a reflection note for the Scandinavian Journal of Information Systems and in a debate article in Organisation & Samhälle. Together with Vinnova and VR, the initiative is redesigning the application portal and case management system for Swedish state research funding. The work uses a radically faster, AI-enabled development approach in which developers and domain experts build together in short, intensive sprints. The aim is a new platform that can meet the requirements of a funding system in which AI agents take part on both sides. Applicants increasingly use agents to write applications, and funders need new capabilities to assess, match and orchestrate research in response. The first sprint, which focuses on the application portal, is underway, and the case management system will follow with the same approach. The development process is also being studied as a research case in its own right.
The initiative contributes to national and Nordic debates on how research quality should be assessed, and how research funding should be allocated, once AI has eroded the signal value of articles and applications. Through the collaboration with Vinnova and VR, these ideas are moving from debate into the infrastructure that Swedish research funding actually runs on. The collaboration also shows how public agencies can develop critical digital systems far faster and at a fraction of the conventional cost. It offers researchers, departments and funders practices they can adopt now. The first is to make analytical provenance the default. The second is to validate AI-augmented pipelines before scaling them. The third is to credit research infrastructure as scientific contribution. The fourth is to open protocols and validation results to a shared commons. Its longer-term aim is to extend rigorous inquiry to organizations that have been structurally excluded from producing trustworthy knowledge about their own practice, without removing human judgment or accountability.
Vinnova, Swedish Research Council (VR)
Contact our team to learn more or explore collaboration opportunities.
