Integrating LLMs with local knowledge bases assists sustainability, but neglects tacit knowledge and complexity

This article is part of three series: artificial intelligence, KM & SDGs initiative, and KM is sustainable development.

In September 2024, world governments endorsed the Pact for the Future¹ at the Summit of the Future – the most important global event for sustainable development this decade. The Pact for the Future is the blueprint for the reinvigorated United Nations (UN) Sustainable Development Goals (SDGs) for the next five years, and was prepared after the 2023 SDG Progress Report: Special Edition² highlighted lack of progress towards the SDGs.

The final Pact for the Future includes the following action, reflecting the key calls of a year-long collaborative campaign by RealKM Cooperative Limited (RealKM), the Knowledge Management for Development Community (KM4Dev), and Knowledge for Development Partnership (K4DP), as expressed in our open letter and submission and journal editorial³:

Integrating LLMs with local knowledge for sustainable development

A newly published paper⁴ proposes a role for ChatGPT-like large language models (LLMs) in supporting “the need for science, technology and innovation to be adapted and made relevant to local needs and circumstances” part of Pact for the Future Action 32.

To explore this potential, the paper authors use the example sustainability issue of carbon neutrality, which is an aspect of climate action. They state that LLMs are transformative AI tools offering the potential to revolutionize how we approach the economic, technological, social, and environmental issues of achieving carbon neutrality. They alert, however, that the full potential of these models in such a task is yet to be realized, hindered by limitations in providing detailed, localized, and expert-level insights across an expansive spectrum of subjects. The authors’ observations in this regard align with the long-recognised need for “actionable climate knowledge,” as I raise in a 2024 article.

To bridge these gaps, the paper authors introduce a framework that integrates local knowledge with LLMs, aiming to markedly enhance the depth, accuracy, and regional relevance of the information provided, as shown in Figure 1. The “local knowledge” in the framework is in the form of local knowledge bases (LKBs).

The authors examine the effectiveness of the framework from the perspectives of government, corporations, and the community. They find that the integration of local knowledge with LLMs enriches the comprehension of local specificities, and also facilitates up-to-date information that is crucial for addressing the specific concerns and questions about carbon neutrality raised by a broad array of stakeholders.

They conclude that their proposed LLM-LKB framework showcases significant potential in enhancing societal comprehension and participation towards carbon neutrality, and based on this, list the potential applications shown in Figure 2.

Neglecting tacit knowledge and complexity

The LLM-LKB framework looks to be of assistance in helping to localize sustainable development issues such as carbon neutrality. However, the “local knowledge” in the local knowledge bases (LKBs) consists only of explicit knowledge, which the paper authors describe as “incorporating various formats like PDF, DOC, TXT and CSV to cover an extensive range of sources.”

Completely missing is tacit knowledge. Indeed, despite “local knowledge” being a primary focus of the paper, “knowledge” and “local knowledge” are not even defined and described, when they should have been, drawing on the extensive body of knowledge management (KM) research. This means that discussion of different types of knowledge as they relate to local knowledge is absent from the paper, including from the limitations section which makes no mention of tacit knowledge having not been considered in the paper.

Tacit knowledge is a vital aspect of the multiple knowledges approach that is essential to the success of the Pact for the Future, as discussed in a previous RealKM Magazine article:

Ultimately, the SDGs will not be achieved by the UN or its Member States. Rather, the SDGs will – or won’t – be achieved by the billions of individual people across the world who need to be able to integrate the SDGs into their daily lives. But every one of those billions of people has different realities: different needs, different values, different goals, different work circumstances, different family situations, different community and cultural contexts, and so on. This manifests as a diverse array of knowledge – “multiple knowledges” – that must be part of the decision-making if the SDGs are to be achieved. …

However, different realities such as these can bring divergent and often conflicting perspectives, which makes sustainability complex.

As references to that article clearly show, the only way that pathways forward can be found in the face of these complex realities is through effective stakeholder engagement.

Supporting this localization emphasis, the UN has also published⁵ the Inter-Agency Policy Brief: Accelerating SDG Localization to deliver on the promise of the 2030 Agenda for Sustainable Development, which includes a focus on stakeholder engagement and partnerships. It states that:

Comprehensive stakeholder engagement and participatory governance can help foster a better understanding of local needs and priorities and cultivate a sense of shared commitment towards achieving sustainable development for all.

Meaningful engagement of stakeholders in SDGs localization processes should consider:
1) establishing listening and participatory mechanisms at local/community level, creating an enabling environment for participation in problem identification, policy making, implementation, monitoring and evaluation;
2) mapping and mobilizing stakeholder representatives from different sectors, harnessing diverse perspectives and local expertise to identify priorities and craft effective sustainable solutions;
3) facilitating sharing of information and data;
4) raising awareness about the SDGs and their local implementation; and
5) setting up or strengthening established platforms for follow up and review, enabling local leaders and communities to take ownership of local development efforts and foster partnerships and structural dialogue that support SDGs implementation.

Header image: A local landholder shares his tacit knowledge during a dairy farm field day, as part of the Smart Farms projects of the Hawkesbury-Nepean River Recovery Program. Source: © Bruce Boyes, CC BY-NC-ND 4.0.

References: