What if knowledge exists but nobody knows? [Forum special series, Artificial intelligence]

This article is part of a special series of summaries of keynotes and presentations from the KM Triversary Forum 2025, and also part of the artificial intelligence series.

AI as a catalyst for knowledge transfer in sustainability networks: reflections from the Green Growth Knowledge Partnership experience. By Gianguglielmo Calvi, Senior Knowledge Management Systems Expert, Green Growth Knowledge Partnership.

When it was announced that the Library contained all books, the first reaction was unbounded joy … That unbridled hopefulness was succeeded, naturally enough, by a similarly disproportionate depression. The certainty that some bookshelf in some hexagon contained precious books, yet that those precious books were forever out of reach, was almost unbearable.

Jorge Luis Borges, The Library of Babel¹

Since the Paris Agreement on climate change in 2015, global interest in green growth, green economy, and circular economy has increased dramatically, as has the volume of research, case studies, and technical knowledge products in these domains. Zorpas and colleagues² documented that the journal Sustainability alone published nearly 17,000 papers in just 2022, a 21% increase over the previous year.

Unfortunately, the rapid growth and decentralised development of the body of knowledge around sustainable economies is leading practitioners in these domains straight into Borges’ library.

In this article I am thus sharing how we have been trying to address such challenge at the Green Growth Knowledge Partnership (GGKP)³ and specifically how we ended up developing an AI powered Knowledge Brokerage API able to respond to one of our primary driving questions: can AI accelerate the transfer of knowledge across multiple knowledge hubs and the understanding of complex matters among different target audiences?

What we mean by knowledge here

If there is something we all know as knowledge management (KM) practitioners is that the word “knowledge” carries so many different meanings and it is open to so many interpretations that its distinction shapes everything that follows.

The KM literature, from Polanyi⁴ through Nonaka and Takeuchi⁵ to Alavi and Leidner⁶, draws a foundational distinction between tacit knowledge, the personal, experience based understanding that individuals carry without being able to fully articulate it, and explicit knowledge, which can be codified and transmitted in structured form. Nonaka and Takeuchi described organisational knowledge creation as a continuous spiral of conversions between these forms, from socialisation and externalisation through combination and internalisation. Alavi and Leidner developed this point further: knowledge is not information stored in a repository but information combined with experience, context, and reflection, applied within a specific domain of action.

In this article, partially inspired by the people-centric definition in the D-A-CH Wissensmanagement Glossar 2020^7,8, I refer to knowledge as what is conveyed to and retained by a target audience when that audience engages with a knowledge product in response to a question or while exploring a domain of interest. A knowledge product is a digital artefact designed to make expertise available in transmissible form. It can be a research paper, a case study, a policy brief, a webinar recording, a set of guidelines, a best practice document, a data visualisation, or a conversation with a human expert in an online community of practice. Each format represents a different act of externalisation: a research paper codifies empirical findings; a policy brief translates those findings into the language of decision making; a webinar adds elements of dialogue and shared context; a thread in a community of practice holds insights on how to operationalise a new technology.

While collecting, producing and distributing knowledge products is necessary, it is certainly not sufficient if we expect knowledge to trigger actions. Knowledge becomes actionable when a target audience understands how the information conveyed leads to desired changes, and what concrete actions those changes require. A policymaker who reads a circular economy report possesses information. That information becomes actionable knowledge when the policymaker grasps how its findings apply to their specific legislative, institutional, and economic context, and can identify the steps needed to translate that understanding into decisions.

In this regard, allow me to bring into the picture the 3 Sphere Model⁹ developed by Pavel Kraus and Gil Regev within the Swiss Knowledge Management Forum, as a tool offering a useful lens for visualising this progression. The model identifies three zones of KM activity: the Information Sphere (repositories, taxonomies, search engines), the Action Sphere (where information begins to be contextualised and connected to decisions), and the Knowledge Sphere (the domain of conversation, storytelling, and internalised understanding). Most KM investment concentrates in the Information Sphere. The highest contribution to organisational success, however, comes from the Knowledge Sphere, where knowledge manifests in the interaction between people. I believe that the passage between these zones is where the most consequential and possibly most difficult work of knowledge management takes place.

This is precisely the passage that GGKP is trying to accelerate with AI.

GGKP: a knowledge broker for the green economy

GGKP is a global partnership of organisations and experts, co-founded by the Global Green Growth Institute (GGGI), the Organisation for Economic Co-operation and Development (OECD)¹⁰, the United Nations Environment Programme (UNEP), the United Nations Industrial Development Organization (UNIDO), and the World Bank Group and counting more than 90 international Knowledge Partners. Its mandate is to provide actionable knowledge for the transition to an inclusive green and circular economy, acting de-facto as a knowledge broker in these domains. GGKP collects, produces, curates, and disseminates thousands of knowledge products and connects thousands of subject matter experts across many thematic areas.

Historically, a big part of its operations concentrated in the Information Sphere: building repositories, developing taxonomies, classifying resources, maintaining search interfaces. This infrastructure is essential, but by itself it produces primarily a well ordered library, not actionable knowledge. The challenge here is not about a lack of externalised material, it is about how to move that material toward the audiences and contexts where it can be internalised and acted upon.

The GGKP Knowledge Brokerage API: architecture for the passage from content to action

At the centre of our AI integration effort sits the GGKP Knowledge Brokerage API¹¹, a system designed to bridge the gap between vast repositories and expert communities of sustainability knowledge and the diverse audiences who need to access, understand, and act upon that knowledge. The system is built around six interconnected components, each addressing a distinct stage of the knowledge journey.

Taxonomy Management. The API provides full programmatic access to the GGKP taxonomy: over 2,800 published terms across 9 domains, organised in hierarchical relationships extending from broad categories (value chains, waste, chemicals) down to highly specific sub terms such as individual persistent organic pollutants. The taxonomy is not a static classification scheme. It is a living knowledge structure, collaboratively maintained via a Taxonomy Management System¹² by domain experts across the partnership, and it encodes the interpretive architecture that makes every subsequent stage of the pipeline meaningful. A robust taxonomy is necessary, fundamental, to ensure that the AI understanding of a domain is grounded on recognized, stable and yet evolving concepts.

Automatic Classifier. The classifier processes incoming documents and assigns candidate taxonomy terms automatically. Taherdoost and Madanchian¹³ identify automated classification and content tagging as among the most mature AI applications available to knowledge management today. Research on deep learning for document classification has demonstrated that multimodal neural network architectures can achieve accuracy exceeding 96% on hierarchical, domain specific taxonomies. In practice, our classifier transforms a time consuming manual bottleneck into a rapid, semi automated process: the AI generates candidates classifications that are scored against the taxonomy, pre-selected and eventually validated or amended by a domain expert.

Search Federator. The Federated Search¹⁴ component allows content connected to the GGKP network to surface within partner organisations’ own platforms through federated search. A researcher querying a partner portal can access relevant GGKP resources without knowing that GGKP exists as a separate entity. The knowledge finds the user. This is where the API begins to move beyond the Information Sphere: by embedding knowledge discoverability within the systems people already inhabit, it reduces the friction between content and the contexts where that content can become actionable.

Expert’s Bots. Domain specific question and answer agents that help curators and end users navigate the knowledge base conversationally. When a curator is working on a new policy brief about waste management, the bot can surface related case studies, complementary research, and contextual references that the curator might otherwise miss. For end users, the bots represent a step toward the Knowledge Sphere: they transform the interaction from a static document retrieval into something closer to a conversation with a knowledgeable colleague. The ambition is that engaging with a GGKP knowledge product, enriched by AI powered context and dialogue, begins to approach the impact of a conversation with an expert or a story told by a storyteller.

Experts Locator. A service that connects users with relevant human expertise across the GGKP network, surfacing experts alongside contextually relevant knowledge products. Because the most effective knowledge transfer happens through human interaction, the API does not attempt to replace that interaction but to facilitate it. When the AI identifies that a user’s query requires depth, nuance, or contextual judgement beyond what automated systems can provide, the Experts Locator bridges the user to the appropriate human knowledge source. Similarly, when a set of knowledge products retrieved in response to a question points to relevant human experts in the network, those experts are surfaced together with those products.

Knowledge Synthesiser. A component that generates structured summaries and insights from large bodies of evidence. This is one of the most ambitious and complex services provided by the API, because synthesis is where the passage from information to actionable knowledge is most consequential. A policymaker does not need to read 200 papers on circular economy interventions. They need a synthesised understanding of what the evidence says about their specific policy context, what has been tried elsewhere, what worked and what did not, and what actions the evidence supports. The Knowledge Synthesiser is an early attempt to automate parts of this interpretive work, while preserving the human editorial oversight that ensures quality and contextual accuracy.

The five stage pipeline in practice

The six components of the GGKP API map onto an information-to-knowledge pipeline of five stages. Actionable knowledge does not appear at the end of such a pipeline as a finished product; it emerges progressively¹⁵ through how knowledge is handled at each stage. Each stage addresses a specific operational question, and each represents a point where AI either reduces friction or where human judgement remains irreplaceable.

Collation

Gathering potentially relevant publications from partner organisations, research institutions, and multilateral agencies; identifying new experts and knowledge communities around a subject matter; collecting the most reliable data and evidence on a development challenge. Given the almost exponential growth of sustainability knowledge, these tasks can overwhelm dedicated teams within weeks. The API supports automated crawling and metadata extraction, handling the repetitive, high volume work that frees knowledge workers to concentrate on quality assessment and contextualisation.

Classification

The Automatic Classifier and Taxonomy Management modules work together at this stage. The AI generates candidate taxonomy terms for each incoming resource. A domain expert validates. In practice, a UNEP document on chemicals, waste, and pollution touches multiple taxonomy branches simultaneously and requires judgement about primary versus secondary classification. The AI handles the volume here, while the human ensures the reliability and accountability.

Curation

A well classified document is findable, but that does not make it actionable. Curation enriches each resource with context: cross references to related work, summaries, editorial framing that makes the resource relevant for its intended audience. The Expert’s Bots and the Knowledge Synthesiser support curators by surfacing related resources and generating draft summaries. The deeper challenge, however, is interpretive: understanding what a specific audience needs to know, and why, within their particular decision making context. That interpretive work remains human.

Publication

A single finding on circular economy interventions may need to serve a policymaker, a project manager, an investor, and a student, each requiring different framings, levels of detail, and delivery formats. The GEF ISLANDS¹⁶ and GEF FARM¹⁷ knowledge portals and the Waste Management and Circular Economy Policy Support System¹⁸ platform represent early deployments of the API in service of audience specific delivery. GGKP’s country hubs extend this further: a hub for Kenya presents country specific data alongside thematic focus areas and an AI powered assistant that helps users navigate the available resources. The goal is to recombine and translate knowledge products so that they function less like static documents and more like a conversation with a knowledgeable colleague who understands the user’s context.

Dissemination

The Search Federator addresses the final stage: making GGKP knowledge products discoverable within partner organisations’ own systems. Interoperability demands shared standards, aligned taxonomies, and institutional willingness to open search interfaces. These are governance and relationship challenges as much as technical ones. But this is where AI powered KM holds some of its most important potential: not within any single organisation’s systems, but in the connective tissue between organisations, enabling knowledge to travel across institutional boundaries toward the contexts where it can become actionable.

What AI can do, and where the limits lie

The research literature on AI in knowledge management converges on a consistent set of findings that mirrors our operational experience at GGKP. Gelashvili-Luik, Vihma, and Pappel document precisely this pattern in a systematic review¹⁹ spanning forty peer reviewed studies.

AI is effective at high volume, consistency dependent tasks: crawling sources, extracting metadata, generating classification candidates against a defined taxonomy, identifying related resources across large collections, and producing draft syntheses. These are Information Sphere operations, and in these areas AI delivers speed and scalability that manual processes cannot match.

AI is not yet effective at the processes that create actionable knowledge: deep contextual understanding, interpretive nuance, audience specific framing, and the relational trust that enables knowledge to travel through conversation, mentoring, and community engagement. In Nonaka and Takeuchi’s terms, AI excels at combination (explicit to explicit) and can assist with externalisation, but socialisation and internalisation, the processes through which knowledge becomes truly personal and actionable, remain fundamentally dependent on human engagement. Jarrahi and colleagues²⁰ call this division of labour collaborative intelligence, a framing that I find productive because it shifts the question from what AI can replace to what it allows knowledge workers to concentrate on: the interpretive, relational, and strategic work that transforms information into actionable knowledge for specific audiences.

This constitutes a research agenda worth pursuing. Can AI accelerate not just the management of information but the passage of knowledge from its externalised form to its internalised, actionable state in the minds of specific audiences? Under what organisational conditions does this succeed? The GGKP API represents one operational attempt to answer these questions. The broader KM community needs many more.

Questions for the road ahead

What is your taxonomy, really? The quality of AI powered classification depends on the quality of the underlying knowledge structure, so maybe it is more important to structure knowledge first and introduce technology later.

Where does human judgement matter most? Identifying where automation genuinely frees knowledge workers versus where human interpretation creates actionable knowledge is a prerequisite for meaningful AI integration.

Can your knowledge products function like a conversation? The passage from static content to contextualised, audience aware interaction is the passage from information to actionable knowledge. AI powered bots, country hubs with interactive assistants, and federated search are early attempts to make this crossing. The question remains open: how close can a digital product come to the impact of a story told by a storyteller, or a conversation with an expert who understands your context?

What would federation look like in your ecosystem? If your users work in systems you do not control, how might you make knowledge findable within their existing workflows? The GGKP API’s Search Federator is one approach. The question is about architecture, but also about partnerships, standards, and trust.

How will you know if knowledge is becoming actionable? The meaningful measure is not processing speed or classification accuracy. It is whether the right knowledge reaches the right people in time to inform better decisions and inspire better solutions. How do we measure that? Can we?

An invitation

I am a firm believer in the transformative power of combining strong human relationships with intelligent technology. I am equally wary of the claim that AI will solve knowledge management’s hardest problems on its own. It will not. What it can do, when integrated thoughtfully into a well understood KM pipeline, is accelerate the journey of knowledge from its stored, classified form toward the people and contexts where it can generate understanding and action.

The GGKP Knowledge Brokerage API is one attempt to build the infrastructure for that journey. It is a work in progress. The larger trend, the emergence of AI as a potential catalyst for knowledge transfer across organisational and geographic boundaries, deserves sustained investigation by our community. I share this experience as an invitation: to map your own knowledge pipelines, to identify where the distance between content and action is greatest, and to approach AI integration with both ambition and intellectual discipline.

If we are serious about the transition to a sustainable, inclusive economy, then how we manage and share knowledge is not a supporting function. Working together to strengthen it, responsibly and with humanity, is one of the most meaningful contributions our community can make.

Biography:

Gianguglielmo Calvi is a computer scientist, knowledge manager, and professional based in Switzerland, founder of Heuristica and co-founder of EnQu Ideation. A Senior Knowledge Management Systems Expert at the Green Growth Knowledge Partnership (GGKP) and a board member of the Swiss Knowledge Management Forum, his former research in artificial intelligence and cognitive science has deeply influenced his career. Since 2008, he has dedicated himself to delivering transformative, efficient, and secure knowledge and information management solutions within complex organisational domains. In 2025 he published his first book How do we know how much we don’t know?

Presentation resources: PowerPoint slides.

Header image source: Created by Gianguglielmo Calvi using ChatGPT.

AI statement: AI was used to assist with editing this article for grammar and readability, and to help identify and verify open access research references. The substantive content, analysis, and conclusions are entirely the author’s own, based on the presentation delivered at the KM Triversary Forum 2025 (Session E2, 15 October 2025).

References and notes: