I recently presented on positive digital development (PDD) to a group of ICT4D doctoral and more senior researchers.  I came away with a clear sense that the idea was seen as having value but that researchers wanted more concrete ideas through which it could be understood and put into practice.

A term that comes up repeatedly in relation to PDD is “hope”; something which reflects an aspiration for a better future and for better uses of digital in order to achieve that future.  And – through a paper I wrote with Shyam Krishna, “ICTs and Hope for Development” – this offers a concrete idea that could be one route to conceptual operationalisation of PDD.

Drawing on Snyder’s hope theory, we define hope in the paper as “the goal-directed capacity to both find routes to these goals (pathway) and the motivation to pursue goals (agency)” though it can be rather more-broadly understood from Bloch as the anticipation of a not-yet-realised future, oriented towards transformative change.  The paper develops a set of dimensions of hope based around “the Subject of hope (who hopes); the Object of hope (what they hope about); and the Enaction of hope (what is seen to be required to put hope into practice)”.

But the core model of relevance to PDD links ICTs and hope as shown below.

This sees hope as both an input to but also an output from ICT4D interventions:

• “Hope for development is a psycho-social input factor influencing adoption and uptake of ICTs by impacting expectations and motivations.
• Hope results as an output factor of ICT adoption and usage: ICT users may become more hopeful about the future as a result of their usage.
• That created hope then acts via a feedback mechanism to influence further ICT adoption and use (or non-use) of technology within any ICT4D initiative.”

We then test the model, using it as an analytical frame through which to evaluate evidence of a well-known ICT4D initiative: One Laptop Per Child.  This provides greater detail for each aspect of the model (see below) and supports the value and validity of the model for understanding hope in ICT4D, though also showing the need for primary and ideally longitudinal research into how hope changes before, during and after ICT use.

For PDD, the value of the model is its analytical power.  It doesn’t try to make a simple association between ICTs and positive impacts but – per the definition of hope – asks whether digital helps create credible goals, workable pathways and strengthened agency.  It therefore offers a way to study the positive potential of digital development without collapsing into naïve optimism.

Looking forwards, the model can be used to enact a hope-related thread in future positive digital development research.  Potential research questions it can be used to answer include:

• How do different stakeholders imagine hopeful digital futures, and what forms of agency and pathway do they see as necessary to realise those futures?
• When and how does engagement with digital technology increase people’s sense of agency, capability and future possibility?
• How are hopes for digital development distributed across the wider ecosystem of users, intermediaries, designers, funders and policymakers, and what happens when those hopes align or diverge?

These and others can be part of PDD research investigating the conditions under which digital development creates grounded hope; analysing how collective and actionable possibilities for better futures are formed in relation to digital and development.

These questions sit at the intersection of two well-established bodies of knowledge.

The first is the broader literature on spatial data infrastructure (SDI) and governance.  This has documented, across many development contexts, how the availability and quality of geospatial data shapes the capacity of institutions to plan effectively including responses to environmental risks.[i]  This literature also shows how the ability to make stakeholder claims spatially explicit is increasingly central to inclusive land management.

The second is a body of work specific to mountains, which has consistently found that these regions are data-scarce relative to their governance and environmental significance: monitoring is not only limited but unevenly distributed, datasets are fragmented and often inaccessible, long-term monitoring is rare and subnational spatial data layers frequently exist at insufficient resolution for the planning decisions that actually need to be made.[ii]

Against this backdrop, the 2025 MRI Mountain Governance Working Group Global Survey provides a rare empirical window onto how geo/GIS data infrastructure is working (or not) in specific mountain governance contexts across the globe.  Twenty sites participated, spanning ranges from the Karakoram to the Cantabrians, and the Andes to the Eastern Arc of Tanzania.  The survey asked respondents to rate the effectiveness of 26 categories of GIS dataset in supporting decision-making, alongside a prior question on whether an accessible national or regional GIS database exists at all.

The response pattern to that prior question is worth noting.  Of the twenty sites, only nine completed the GIS section.  Two indicated that a national/regional database was not available and a further nine skipped the section altogether.  This non-completion by almost half the sample cannot simply be attributed to survey fatigue as other sections of the survey were completed by all respondents.  It more plausibly reflects a meaningful absence of the geodata infrastructure the questions were about.

Among the nine sites that did engage with further detail, the ratings of individual data layers reveal a pattern (see figure below).  Layers grounded in standardised, internationally- or nationally-produced data – administrative boundaries, nature and protected area inventories, terrain morphology, cadastral records – score consistently highest, all rated at or above 3.0 on a four-point scale of effectiveness in supporting mountain region decision-making.  Those dependent on specialist in-situ monitoring or operational management systems – environment and energy, geology, natural hazards, hunting and fishing – cluster at the bottom, several rated at or below 2.33.

Figure: Mean GIS feature decision-support effectiveness rating, descending order

That weakness in natural hazards GIS takes on particular significance when read alongside responses from a separate section of the survey on natural hazard risk.  Wildfires, torrential rains and flooding, and landslides were rated as presenting somewhat high to high risk across the sample.  The Sendai Framework for Disaster Risk Reduction explicitly calls on countries to disseminate risk information using geospatial technology and to make hazard-exposure data freely available.[iii]  Yet the spatial data layer most directly relevant to meeting these obligations, which was natural hazard mapping, is among the weakest in the survey.  The mismatch between the level of hazard exposure and the quality of the GIS infrastructure meant to support hazard governance in mountain areas is arguably the single most policy-relevant finding in the data.

Two further results from adjacent sections of the survey reinforce the picture.  When asked whether soil functions are adequately mapped at their research site, ten of seventeen responding sites said no; only two said yes.  When asked whether spatial planning instruments for groundwater management are effectively implemented, the ratio was identical.  These are concrete deficits because poor soil and catchment-area mapping directly constrains the land management decisions that are a key part of mountain sustainability planning.

More broadly, four governance characteristics heavily dependent on good spatial information  – climate change adaptation, disaster risk reduction, cross-level governance coordination and transparency in government decisions – were all rated below the midpoint of the scale across the full sample of twenty sites.  This is a correlation, not a demonstration of causation, and there are plainly many drivers of governance weakness in mountain areas beyond data infrastructure.  But the consistency of the pattern is at least suggestive of a relationship between geodata gaps and planning shortcomings that warrants further investigation; in particular the first two which are so central to the relation between mountain regions and environment.

Twenty sites is a small and self-selected sample, and the nine who completed the GIS section represent a subset with a probable bias towards places where GIS infrastructure is sufficiently established to prompt engagement.  The findings therefore almost certainly understate the scale of spatial data gaps globally.

Taken together, however, the survey already provides evidence grounded in specific mountain sites from four continents that the SDI divide familiar from the wider development literature is reproduced, and perhaps amplified, in mountain governance contexts.  The operational data layers most needed for adaptive governance are precisely those that are weakest.  Addressing this will require more than technical investment: it will require understanding why these gaps persist across such varied institutional and geographic settings, and what governance arrangements are most effective at building the in-situ capacity that standardised global datasets cannot substitute for.[iv]

—

[i]  McCall, M.K. (2003) Seeking good governance in participatory-GIS: a review of processes and governance dimensions in applying GIS to participatory spatial planning, Habitat International, 27(4), 549–573 https://doi.org/10.1016/S0197-3975(03)00005-5

Ros-Tonen, M.A.F., Willemen, L. & McCall, M.K. (2021) Spatial tools for integrated and inclusive landscape governance: toward a new research agenda, Environmental Management, 68, 611–618 https://doi.org/10.1007/s00267-021-01547-x

[ii]  Shahgedanova, M., Adler, C., Gebrekirstos, A., Grau, H.R., Huggel, C., Marchant, R., Pepin, N., Vanacker, V., Viviroli, D. & Vuille, M. (2021) Mountain observatories: status and prospects for enhancing and connecting a global community, Mountain Research and Development, 41(2) https://doi.org/10.1659/MRD-JOURNAL-D-20-00054.1

Ly, A., Geschke, J., Snethlage, M. A., Stauffer, K. L., Nussbaumer, J., Schweizer, D., … & Urbach, D. (2023). Subnational biodiversity reporting metrics for mountain ecosystems. Nature Sustainability, 6(12), 1547-1551 https://doi.org/10.1038/s41893-023-01232-3

[iii]  UNDRR (2015) Sendai Framework for Disaster Risk Reduction 2015–2030, United Nations Office for Disaster Risk Reduction https://www.undrr.org/publication/sendai-framework-disaster-risk-reduction-2015-2030

[iv]  This post develops from an overview of findings from the whole survey available at: https://mountainresearchinitiative.org/news/what-matters-in-global-mountain-governance-in-connection-to-spatial-planning/

Image: NASA/JPL/NIMA https://science.nasa.gov/photojournal/srtm-colored-height-and-shaded-relief-laguna-mellquina-andes-mountains-argentina/

Originally published in the Mountain Digital Futures newsletter: https://www.linkedin.com/newsletters/7400288732999733248/

As an ICT4D researcher, I have been interested in how AI is shifting from a mere industrial tool into a core factor that reshapes governance structures. To understand how political power actively embeds emerging tech into the core of the state’s machinery, I believe China’s upcoming “15th Five-Year Plan” (2026-2030) offers a window.

Within China’s highly institutionalised governance system, a Five-Year Plan is far more than a simple vision statement. As the backbone of China’s state-led economic planning system, it dictates large-scale resource allocation and exerts a decisive coordinating role over macroeconomic governance and all levels of government through the establishment of binding targets. Reading through this newly launched strategic document, for me, what stands out most is the government’s explicit push to build a “new form of intelligent economy”. Here, I see the state intentionally positioning AI as a foundational infrastructure to achieve a coordinated layout: “accelerating high-level technological self-reliance” and “building a modern industrial system”.

By unpacking the power flows and governance networks embedded within this strategic document, I hope to share my insights into how state capacity is being systematically planned and rewired in the era of the intelligent economy.

The Bureaucratic Shift

To make sense of China’s AI governance trajectory, it helps to look at where administrative power is shifting. Back in 2017, the State Council’s “New Generation AI Development Plan” first articulated the vision to “promote the deep integration of artificial intelligence with various industrial sectors to form…… intelligent economy”. For years thereafter, AI was primarily viewed through the lens of research and innovation, falling under the jurisdiction of the Ministry of Science and Technology. Its integration with economic development was usually just a sub-chapter in broader digital economy plans.

However, observing the run-up to the “15th Five-Year Plan”, we can identify a pivotal bureaucratic realignment: the primary jurisdiction over AI has shifted from the Ministry of Science and Technology to the National Development and Reform Commission (NDRC). And this doesn’t happen in a vacuum. It is accompanied by the establishment of the National Data Administration, as well as the nationwide rollout of local government data administration, which were fully established by the first half of 2024 to lead digital planning at the local level.

To my mind, in the context of China’s national governance, this is not just an administrative reshuffling, but broadcasts a clear political signal. The NDRC is not an institution focused on technology; rather, it is an institution focused on the national macroeconomy, responsible for dictating national economic planning and steering massive resource allocation. So this shift signifies that the state now positions AI no longer merely as a key technology or business project, but as a foundational infrastructure, serving as a critical channel for macroeconomic governance and mobilisation of national resources.

The Blueprints

At the core of the “15th Five-Year Plan” lies a dual imperative: “accelerating high-level technological self-reliance” while simultaneously “building a modern industrial system”. And it has also for the first time proposed its development direction: intelligent, green and integrated. To achieve this, the official strategy outlines a massive renovation of physical and digital infrastructure. Policymakers plan to integrate traditional transport and power grids with the national computing network, establish trustworthy data spaces, and even reconfigure the higher education system to secure a pipeline of talent. These foundational layers are designed to pave the way for upgrading strategic emerging industries (e.g. intelligent connected vehicles) and future industries (e.g. embodied AI and quantum technology).

Within this expansive architecture, the policy framework proposes deploying the “AI+” initiative systematically across six key domains, embedding intelligent systems into every facet of society. In Scientific Technology, the goal is to forge AI-driven research paradigms and scientific large models. In Industrial Development, the plan pushes AI deep into the operational loops, from dispatching power systems and accelerating agricultural digitalisation to deploying intelligent terminals across services. In Consumption and Public Welfare, the blueprint envisages the popularisation of AI-native hardware (such as AI phones and computers), the application of intelligent supplementary tutors and diagnosis, and the deployment of embodied AI in hazardous labour environments. In Governance, the strategy outlines a multi-agent public security system collaboratively managed by “natural persons, digital persons, and intelligent robots” (a profound shift in statecraft that warrants deep scrutiny). Lastly, in Global Cooperation, the document extends outward, proposing international AI cooperation organisations and a globally open open-source technological ecosystem.

Table 1: Four-stage evolutionary logic of “AI+”

(Source: Liu, Z. et al. (2025). ‘AI+’ Initiative: Concept Evolution, Mechanism of Action, Governance Logic and Development Path [in Chinese]. E-Government, (11), pp.2–28.) (Translated by the author.)

What conceptually elevates this blueprint from a standard digitalisation roadmap is a distinct “four-stage evolutionary logic” constructed by Chinese policy researchers (Table 1) (Liu et al. 2025). This framework clarified how policymakers conceptually map the trajectory of technological integration. While the initial stages focus on unidirectional empowerment, the national strategy explicitly claims to transcend these phases. It targets what the researchers term the High Stage (AI+), which purportedly demands “full-domain reshaping”. Ultimately, the blueprint points toward an idealised Final Stage (AI⁺), where AI-native enterprises drive an exponential increase in total factor productivity, culminating in a profound “transformation of production relations”.

Reading Between the Lines

Of course, as with any large-scale institutional transformation, moving from a conceptual blueprint to grounded implementation inevitably generates administrative friction. Currently, the governance system is navigating a complex transitional overlap: legacy “digital” policies have not yet exited the stage, while new “intelligent” directives are concurrently being rolled out. Plus, even though the top-level design calls for the “efficient supply of computing power, algorithms, and data”, the administrative mechanisms required to orchestrate these resources remain fragmented and have yet to be fully streamlined. As a result, local implementers are navigating massive new infrastructure projects by relying on a familiar, pragmatic tradition: “crossing the river by feeling the stones”. To some extent, the top policymakers are also waiting for potential practical cases to emerge from the grassroots. The top level is responsible for setting the direction; as for how to implement it, it relies on local authorities to innovate based on local conditions. Successful local cases are taken as pilot projects and then promoted nationwide. This is the consistent approach of China’s governance.

From the perspective of ICT4D, I think there are some critical structural questions about this strategy we can pay attention to. The first one is about spatial inequality. If we examine this blueprint through the lens of the five-layer AI architecture recently conceptualised by Nvidia CEO Jensen Huang: As the top-down pushing for “intelligent infrastructure”, resource-rich western regions are poised to become the physical bedrock, which provides energy, land, and data centres required for the upper compute and model layers. The eastern coastal regions are positioned to monopolise the higher, value-added layers of intelligent applications and algorithmic services. This potential spatial division of labour suggests that the future “digital divide” might be about structural dependency. The West serves as the energy extraction zone for the intelligent economy, while the East reaps the immense cognitive and economic dividends. So, is the “intelligent economy” forging a new developmental paradigm, or just projecting the old models of the industrial and internet eras onto a new technological driver? Despite the rhetoric of an unprecedented transformation, we may be witnessing a cyclical repetition of spatial inequality, simply upgraded for the AI era.

The second question comes from the workforce. While deploying embodied AI in hazardous labour environments offers a positive social safeguard, the AI revolution inevitably accelerates labour displacement. The plan does blueprint the attractive technological empowerment and industrial upgrading, yet we still expect more discussions on labour rights protection and safety nets to come to the stage (although it is fair to acknowledge that detailed social safeguard measures are typically addressed in parallel, domain-specific policies rather than a Five-Year Plan). Still, this structural labour displacement is not new; we have witnessed similar pain during the mechanical automation of the industrial era and the digital optimisation of the internet era. So we have to ask how this new intelligent economy will absorb the traditional workforce squeezed out by technological upgrading. Have we learned lessons from the past to proactively design a pathway for a just transition? Or are we once again leaving the most vulnerable workers to bear the costs?

The third question relates to data rights. The blueprint’s vision of a multi-agent governance system, which will highly possibly enhance state capacity and predictive management. But again, we ask the old-school but important concern: how to find a delicate balance between the systematic data collection and process and individual privacy? Furthermore, we care about the global spillover. This infrastructure-first developmental model contrasts with the risk-based regulatory approaches commonly observed in the West. As China looks outward, it might present an alternative pathway for the Global South. Then, will the export of this governance model and underlying digital infrastructure foster empowerment, or create new forms of technological dependency?

Close

China’s AI strategy under the “15th Five-Year Plan” is not only a technological iteration of productive forces, but a systemic endeavour aimed at rewiring the underlying logic of national governance. As intelligent agents are set to be increasingly embedded within governance systems, the state is likely to restructure its patterns of resource extraction and predictive management. Over the next five years, I look forward to seeing how this ambitious blueprint translates into grounded practice, particularly how the state navigates the above structural frictions.

The deep integration of AI into the state machinery will unfold into a multitude of possible trajectories and complex social interactions. But how this profound transformation will ultimately reshape national governance remains a vast and open question. This ongoing evolution is not just a reality unfolding in China. It provides a critical site through which the future trajectories of global digital governance may be observed and contested.

Further Information

15th Five-Year Plan for National Economic and Social Development [in Chinese]

‘Largest Infrastructure Buildout in Human History’: Jensen Huang on AI’s ‘Five-Layer Cake’ at Davos

AI Is a 5-Layer Cake

Forum: Technology in China’s 15th Five-Year Plan

Note: The policy content cited in the blog was translated by the author. This blog gives the views of the author featured and does not necessarily represent the views of the Centre for Digital Development as a whole.

Across hill and mountain uplands, digital tools are increasingly central to how ecosystems are monitored and managed: from drones mapping upland restoration, to sensors tracking glacial melt and biodiversity, to citizens submitting wildlife sightings via mobile apps.  However, the promise of this digital turn will only be realised if conservation actors pay close attention to three interrelated issues: data quality, data praxis and data justice.

Data Quality: Getting the Foundations Right

It’s an obvious point that high-quality data is the foundation of any effective digital conservation system.  At the University of Manchester, we use CARTA to define data quality: completeness, accuracy, relevance, timeliness and appropriateness of presentation.[1]  Shortfalls from CARTA are a generic problem for conservation data with, for example, “gaps and biases in biodiversity data” potentially leading to “misleading results” and “misplaced conservation action”.[2]

In mountain regions these problems can be exacerbated by various factors.  In many regions, inaccessibility is a challenge to data gathering; for example, steep terrain, remoteness and conflict all make this difficult.[3]   Continuity is also challenging.  Many efforts generate data in the short term but then die when funding ends.[4]  Community and other citizen science datasets, likewise, can surge but then fall rapidly away.[5]

Even if these barriers can be overcome, without better attention to spatial coverage, temporal depth and contextual interpretation, the risk is precise, credible-looking numbers which are, in practice, misleading.  Investment in technology therefore needs to be matched by investment in data stewardship that ensures metadata standards and quality assurance processes.

Data Praxis: From Data to Decision to Results

Good data alone does not guarantee good outcomes.  What matters is the praxis: the process of turning raw data into useful information, informed decisions, effective actions, and measurable results, as per the information value chain model.[6]

For sure, there are mountain-related examples that run the full chain, from resolving legal conflicts to creation of protected areas for endangered species.[7]

Effective praxis such as this depends on institutional capability: analysts able to interpret data, managers willing and able to act, and monitoring systems that feed outcomes back into decision-making.  In reality, though, there may be clear data, and sometimes even clear information but the chain can quite often stall when decision-makers do not want the new data or when action requires cross-organisational coordination, negotiation with local communities, higher budgets, etc.

What is needed, then, right from the start of conservation initiatives, is a data praxis analysis: analysing every step in the information value chain; for example, to identify resource requirements, barriers and enablers to try to ensure that investments pay off.  One basis for this analysis can be the expanded information value chain model.[8]

Data Justice: Whose Data, Whose Benefit?

Finally, conservation’s digital future in mountain regions must be a just one.  Research on conservation data warns that “Ignoring justice concerns in the generation, analysis and use of … data risks perpetuating inequities in whose knowledge counts, whose questions are asked and whose interventions are supported”.[9]  Mountain conservation data landscapes are populated by national agencies, NGOs, researchers and local communities but data power is unevenly distributed.

We can understand this using the CRAB model: analysing the control, representation, access and benefits of specific conservation datasets.[10]  Armed with this understanding, one can look to develop innovative approaches such as “data gardening” that help to fill data gaps and improve conservation decision making.[11]

Conclusion Digital will be central to the future of conservation in mountain regions.  But the critical work is not the hardware.  It is improving data quality, building effective praxis linking data to decisions and results, and embedding data justice to ensure that benefits are more equitably shared.  The approaches and models identified above can provide some foundation for these necessary changes: do get in touch if these connect to work you are doing with mountain communities.

[1] Heeks, R. (2006) Implementing and managing e-government: an international text, Sage Publications

[2] Bowler, D. E., Boyd, R. J., Callaghan, C. T., Robinson, R. A., Isaac, N. J., & Pocock, M. J. (2025) Treating gaps and biases in biodiversity data as a missing data problem. Biological Reviews, 100(1), 50-67; Bowler, D. & Boyd, R. (2024) Dealing with gaps in biodiversity monitoring data, UKCEH Blogs, 18 Sep

[3] Han, X., Guo, Y., Mi, C., Huettmann, F., & Wen, L. (2017). Machine learning model analysis of breeding habitats for the black-necked crane in Central Asian Uplands under anthropogenic pressures. Scientific Reports, 7(1), 6114; Bettineschi, C., Magnini, L., Giovanni, A., & De Guio, A. (2022). Clearence cairnfields forever: combining AI and LiDAR data in the Marcesina upland (northern Italy). Post-Classical Archaeologies, 12, 49-68

[4] Shahgedanova, M., Adler, C., Gebrekirstos, A., Grau, H. R., Huggel, C., Marchant, R., … & Vuille, M. (2021). Mountain observatories: Status and prospects for enhancing and connecting a global community. Mountain Research and Development, 41(2), A1

[5] Biber, E. (2011) The problem of environmental monitoring. University of Colorado Law Review 83, 1

[6] Heeks, R. (2018) Information and Communication Technologies for Development, Routledge

[7] Bratu, I. A., & Dincă, L. (2021). Using GIS for management of conflicts in natural protected area in Cindrel mountains. In MATEC Web of Conferences (Vol. 343, p. 09011). EDP Sciences; Ma, C., et al. (2020). Transboundary conservation of the last remaining population of the cao vit gibbon Nomascus nasutus. Oryx, 54, 776–783.

[8] Heeks, R. (2014) The Data Revolution Will Fail Without A Praxis Revolution, ICT4DBlog, 14 Aug

[9] Pritchard, R., Sauls, L. A., Oldekop, J. A., Kiwango, W. A., & Brockington, D. (2022) Data justice and biodiversity conservation. Conservation Biology, 36(5), e13919

[10] Heeks, R. (2025) Assessing Data Inequality: The CRAB Approach, ICT4DBlog, 4 Mar

[11] de Albuquerque, J. P., Anderson, L., Calvillo, N., Cattino, M., Clarke, A., Cunha, M. A., … & Trajber, R. (2023) Dialogic data innovations for sustainability transformations and flood resilience: The case for waterproofing data. Global Environmental Change, 82, 102730

Image: Prasanjeet Yadav; https://www.ncf-india.org/high-altitudes

To answer that, I’m summarising here the paper – “Digital Economy Policy in Developing Countries” – written by myself and Rumana Bukht as part of the DIODE (Development Implications of Digital Economies) research network, funded by the UK ESRC.

I define the digital economy as “that part of economic output derived solely or primarily from digital technologies with a business model based on digital goods or services”[1].  On this definition, the digital economy makes up around 5% of global GDP and 3% of global employment.  It promises new forms of growth, new connections to global markets and new opportunities for firms and workers in the global South.  However, that potential remains only partly realised.

Three digital economy challenges

To explain why digital economy development is undershooting its potential, the paper identifies a set of structural constraints that policy has yet to adequately address.

These firstly arise from digital infrastructure.  While connectivity has improved, the foundations of digital economies remain fragile in many developing countries.  Electricity is still unreliable, broadband access is limited and costs are high relative to incomes.  Even where networks exist, device affordability and compatibility remain barriers.

The second issues arise with the digital ecosystem: the human and institutional infrastructure surrounding the technical.  Skills gaps are serious and exist across a range: not just basic digital literacy but the specialist capabilities in data analytics, cybersecurity and AI that the digital economy increasingly demands.  Financial systems are underdeveloped, leaving digital start-ups and SMEs without access to seed or growth capital.  And governance frameworks are often either outdated or absent: for example we cite lack of consumer protection legislation for online transactions, and gaps in cybercrime legislation.

The final and perhaps most important category is digital economy disbenefits.  By this, we mean the emerging downsides we see with digital economy growth.  Digital exclusion is widespread and becomes more significant as digitalisation spreads: women, rural communities and lower-income groups are among those most affected.  But “adverse incorporation” means that many of those who are included participate on highly unequal terms.  I’ve been particularly involved with platform workers in the global South who face under-payment and chronic precarity while platforms capture the majority of value.  Other threats include cybersecurity; a specific problem for developing countries which are seen as an ideal testing ground by hackers.

A framework for policy response

The paper maps policy responses across all three challenge domains, organising these in terms of policy issues, desired outcomes and specific instruments (see example below).

For instance:

On infrastructure: beyond direct public investment in power and telecoms, governments need independent regulation, transparent spectrum licensing and specific plans for mobile broadband.

On ecosystems: ICT curricula need embedding from primary through tertiary education; venture capital frameworks need creating; and legal frameworks for digital labour, e-transactions and consumer protection need updating or establishing from scratch.

On disbenefits: fair work standards for digital labour platforms, anti-monopoly regulation extended online, action to improve Universal Service Funds, and data privacy legislation are all essential but often lagging in developing country policy environments.

Beyond content to governance of digital economy policy

More than new policy instruments, though, what developing countries need is better governance of digital economy policy.  The paper calls for the establishment of “Digital Economy Policy Collaboratories”.

These are cross-cutting structures that bridge ICT ministries with finance, education, enterprise and sectoral ministries, and that incorporate private sector, civil society and community actors.  Such structures must combine clear leadership with genuine multi-stakeholder participation and experimental, iterative approaches.

The digital economy offers real opportunities for development.  But those opportunities will not be realised without policy action that is concerted, evidence-based, inclusive and adaptive.   Without that, the digital economy risks reinforcing existing inequalities rather than transforming them.

For more details, read the full paper here: https://diode.network/publications/

[1] https://research.manchester.ac.uk/en/publications/defining-conceptualising-and-measuring-the-digital-economy/

Recently, Artificial Intelligence (AI) has increasingly entered the core domains of digital urbanisation and governance in multiple areas. Rather than functioning solely as a supporting tool for data analysis, AI is emerging as a deeper technology of governance^[i]. Urban AI can be understood as an extension of smart city technologies and as a new phase that reshapes urban knowledge and the management of urban space^[ii]. In this sense, AI urbanism represents a key trend in the post-smart city era, in which algorithms no longer merely assist governance but begin to shape its very structure^[iii].

Meanwhile, this development raises major concerns regarding fairness, transparency, accountability and inclusivity^[iv][v]. The discussion of urban AI should therefore address questions such as whether citizens’ needs are being met, whether these technologies intensify existing inequalities, and whether city authorities have sufficient institutional capacity to regulate them effectively. This blog approaches AI as a form of dual expansion. It is rapidly spreading into areas such as transportation, public services, tourism fields^[vi], while also shifting the debate from technological efficiency to broader questions of urban governance and public responsibility. Therefore, the core is to understand how AI reshapes the logic and boundaries of urban governance.

The impact of AI in different sectors

The urban transport and mobility have consistently been the main area for AI adoption, where the AI is widely used for traffic forecasting, signal control, road management and the broader optimisation of urban operations ^[vii]. The AI is pushing city management from automation toward the autonomy, meaning that the decision-making is partially shifting from human manager to algorithmic systems^[viii]. As a result, the key issues are no longer simply whether AI is more efficient. More important questions are who sets the optimisation targets, who is accountable for system failures, and how algorithms affect different groups’ access to urban space and transport resources.

In public services and local government administration, the expansion of AI has also brought significant governance implications, including decision-making support, service responsiveness and resource management. The main challenge lies in readiness at the governance level, including organisational capacity, transparency mechanisms, accountability arrangements and public trust ^[ix]. Once AI is integrated into the public service system, government must explain the rationale for its usage, avoid causing systemic exclusion of specific groups, and uphold the procedural justice expected of the public sector. This is also where governance issues become acute.

In risk monitoring, public safety, AI is mainly used for anomaly detection, real-time alerts, incident response and support for law enforcement to enhance the efficiency. However, this area involves intensive surveillance, the processing of sensitive data and the high-risk exercise of public authority. Without a robust governance framework, it is difficult for urban AI to achieve the goals of inclusivity and sustainability ^[x]. The centre issue is whether is operation is transparent and subject to oversight. Urban management is shifting towards deeper logics of classification, ranking, prediction and behavioural discipline.

Conclusion and future work

The significance of urban AI lies not only in the expansion of its application areas, but also in reshaping the fundamental logic of urban governance in two levels. First, the logic of governance is shifting from procedural management towards data-driven prediction and real-time intervention. Second, AI is expanding the boundaries of governance through continuous data collection and risk forecasting, enabling more effective intervention. As a result, the boundaries between transport, service provision, management and monitoring are becoming increasingly blurred. The application of AI in cities represents a fundamental shift in governance models rather than a mere technological upgrade.

AI in urban settings is a subject that deserves further in-depth study, particularly in developing regions. Many cities face a combination of challenges, including uneven infrastructure, limited institutional capacity and significant social disparities. Future research could explore several directions, such as how power relations between technology providers, platforms and the public sector evolve when local government capacity is weak. Further research could also explore how transparency, accountability and redress mechanisms should be established as algorithms increasingly participate in public decision-making. 

^[i] OECD (2025) Artificial Intelligence for Advancing Smart Cities. Paris: OECD. 

^[ii] Caprotti, F. (2024) ‘Why does urban Artificial Intelligence (AI) matter for urban studies? Developing research directions in urban AI research’, Journal of Urban Affairs.

^[iii] Cugurullo, F., Caprotti, F. and Cook, M. (2024) ‘The rise of AI urbanism in post-smart cities: A critical commentary on urban artificial intelligence’, Urban Studies, 61(6), pp. 1168–1182.

^[iv] UN-Habitat (2025) Global Assessment of Responsible AI in Cities. Nairobi: United Nations Human Settlements Programme.

^[v] Wolniak, R., & Stecuła, K. (2024). Artificial intelligence in smart cities—applications, barriers, and future directions: a review. Smart cities, 7(3), 1346-1389.

^[vi] Yigitcanlar, T. et al. (2024a) ‘Artificial intelligence and the local government: A five-decade systematic literature review’, Cities, 150, 105151.

^[vii] Herath, H. and Mittal, M. (2022) ‘Adoption of artificial intelligence in smart cities: A comprehensive review’, Journal of Innovation & Knowledge, 7(4), 100076. 

^[viii] Cugurullo, F. (2020) ‘Urban Artificial Intelligence: From Automation to Autonomy in the Smart City’, Frontiers in Sustainable Cities, 2:38.

^[ix] Yigitcanlar, T. et al. (2024b) ‘Unlocking Artificial Intelligence Adoption in Local Governments: Best Practice Lessons from Real-World Use Cases’, Smart Cities, 7(4).

^[x] Ben Dhaou, S., Isagah, T., Distor, C., Ruas, I. C., & United Nations Human Settlements Programme (UN-Habitat). (2024). Global Assessment of Responsible Artificial Intelligence in Cities: Research and recommendations to leverage AI for people-centred smart cities. In Nairobi, Kenya. United Nations Human Settlements Programme (UN Habitat)

There’s a general assumption that it does, and that it’s problematic[1].  But evidence of this seems to be quite thin on the ground.  This post reviews a rather rough-and-ready but data-based approach to measuring the mountain digital divide at the level of individual countries.

Classifying Countries

One could seek to classify countries in terms of their mountainousness (is that a word?) using simple datasets like average elevation[2] or more complex datasets such as their ‘ruggedness’.  But that doesn’t tell you how many people are living in the mountain-y bits – maybe the upland areas are dramatic and plentiful, but no-one lives there.

Luckily, we have this wonderful article and associated dataset: “Global and country-level estimates of human population at high altitude” which breaks down the proportion of population in each country living within 500 metre bands, and provides a CSV dataset – thank you authors!  Using that, and admittedly rather arbitrarily, I divided countries into four categories:

– Mountain: 50%+ living ≥1,500m

– Hill: 50%+ living ≥500m (but not Mountain)

– Lowland: 75%+ living ≤500m (but not Flatland)

– Flatland: 95%+ living ≤500m[3]

There’s a table at the end that shows which countries go where.  When you read through the list a puzzled look may arise: Mexico as a ‘mountain’ country; Albania as a ‘lowland’ country?  Some of this is a corrective reminder not to confuse topography with population location, but other instances point to the limitations of this approach to categorisation.

Measuring the Mountain Digital Divide

Armed with this categorisation, it’s a fairly straightforward job to then compare that with digital indicators.  For this post, I chose the World Bank/ITU measure of individuals using the Internet (% of population).

Using this, I calculated the average percentage of individuals using the internet for each of the four categories; producing the results below (2022 figures):

There’s a clear gradient, and strong evidence for a country-level mountain digital divide with the higher-altitude-population countries being around about 20 percentage points behind the low-elevation-population countries in terms of internet use.

Controlling for Income

There are various factors that might explain these differences, and an obvious one is wealth.  To control for this, I used the World Bank’s dataset of GDP per capita, PPP (current international $) and ran a regression against ‘Mountain’ as the baseline for the other categories.

Doing this shows a very strong correlation between income and internet use (around US10,000 additional per capita GDP correlates to around 5 percentage points higher internet use).  Because there’s also a correlation between population altitude and national income, any internet use differences between the country categories largely disappear and certainly fall well below any test of statistical significance (p ranging from 0.73 to 0.96, with overall altitude f-test p=0.58).

There is no statistically significant independent altitude effect on internet use.  Put simply, mountain countries are poorer, and poorer countries have lower internet use.

Conclusions and Future Research

When we look globally, countries with significant populations at higher altitudes appear digitally disadvantaged: not just appear, they actually are disadvantaged.  But once income is controlled for, altitude largely disappears as an independent explanatory factor.  Mountain digital inequality, at least at national scale, seems primarily economic rather than geographic.

What next?

Two directions could be worth exploring further.  First, and easiest, one could use other digital indicators such as broadband, though my guess is that will likely show similar results.  Second, you could dig down to the sub-national level, if you can distinguish upland vs. lowland areas within a country and have ICT-related data at the same level.  That, though, would require much more complex data analytics.

Finally, if you spot a way to improve the approach taken in this post, do let me know.

Appendix: Country Classification by Population Altitude

Originally published in the Mountain Digital Futures newsletter: https://www.linkedin.com/newsletters/7400288732999733248/

Image Source: Nyani Quarmyne @ https://www.internetsociety.org/our-work/connectivity/community-centered-connectivity/

[1] https://link.springer.com/article/10.1007/s11629-009-1070-y; https://www.euromontana.org/wt-connectivity/

[2] Dataset download here: https://www.kaggle.com/datasets/mathurinache/list-of-countries-by-average-elevation

[3] Which therefore excludes countries with between 50% and 75% of their populations living below 500m

And why, despite its potential, has AI application not spread more widely?

According to the Food and Agriculture Organisation (FAO), the global population is expected to exceed 10 billion by 2060, significantly increasing food demand (FAO, 2017). At the same time, the conditions surrounding agriculture are not so optimistic. In this context, artificial intelligence (AI) has been attracting growing attention because of its potential to transform agriculture in many ways, such as crop monitoring and resource optimisation. However, in reality, many of these technologies remain at an experimental stage and are not yet widely adopted in practice in developing countries.

This article addresses this question by drawing on the findings of a systematic literature review, highlighting the real challenges of AI in agriculture.

Why is the application of AI to agriculture important?

First, I want to discuss why the AI application in agriculture is important. Agriculture is one of the most fundamental and essential industries to ensure a stable food supply.

Throughout history, we have continuously evolved agricultural techniques and production methods to meet the growing demand for food. By introducing new farming practices and production systems, we have managed to maintain food supply even as the global population has increased.

However, we are likely to face challenges in the future. The global population is rapidly increasing, and food demand is expected to rise. This challenge is especially serious in developing regions where population growth is expected to continue rapidly. For example, the population of Sub-Saharan Africa is projected to double by 2050 (Haub, 2013). As food demand grows alongside this population increase, ensuring stable food production will become an even more urgent issue.

In response to these challenges, many advanced technologies have been introduced in the agricultural sector in recent years. These include the use of genetically modified organisms, precision agriculture utilising satellite imagery and sensor data, and automation technologies. Among these technologies, AI stands out as having particularly significant potential. AI is already being applied across various areas of agriculture, including crop monitoring, irrigation management, disease diagnosis, image processing, and so on. In a field that has traditionally relied heavily on experience and intuition, AI enables data-driven decision-making and supports more efficient and precise agricultural production.

However, AI have so far been adopted mainly in developed countries. In developing countries, their use is often still limited to pilot initiatives rather than to large-scale implementation. Considering the rapid population growth expected in developing countries, using AI to improve agricultural productivity will be increasingly important. This means that AI has the potential to play a vital role in transforming agriculture and strengthening food security in the regions that need it most.

At present, however, its use remains limited. It is well known that factors such as lack of data and insufficient funding hinder the successful adoption of AI. But what is another factor that hinders the adoption of AI in agriculture besides the well-known challenges? To understand these, I am currently conducting a systematic literature review, and I would like to describe what I have found so far in the first half of the systematic literature review.

The following sections describe the research methodology and findings.

The research methodology

I use the PRISMA approach to identify and analyse what factors may also contribute to these challenges. I applied this approach to ensure the convergence of information across multiple sources, thereby strengthening the validity and reliability of the study’s findings (Okoli, 2015).  To understand factors,  I trialled several keyword search strings. Searches based on “artificial intelligence”, “AI”, “deep learning”, and “machine learning” in combination with agriculture-related terms such as “agriculture”, “farm”, “farming”, “food security”, and “agricultural” produced either too many broad results or results not specifically focused on developing regions. Therefore, geographical and development-related terms such as “developing countries,” “global south,” “Africa,” “Asia,” “Africans,” and “Asians” were added. As a result of the search, 65 articles were identified in Google Scholar, as shown below.

Based on the selected 65 articles, I have analysed 28 articles so far and found some challenges described below.

The Gap Between Technical Success and Real-World Impact

One of the biggest problems in current research is that technical success is often treated as if it were the same as real-world success.

Many studies show that AI models can perform very well in controlled settings. For instance, they may achieve high accuracy in detecting crop diseases, predicting yields, or recommending irrigation schedules. These results are important, but they only show that the technology works under specific conditions.

What they do not always show is whether farmers can actually use these tools in their daily work, whether the tools fit local farming conditions, or whether they lead to better outcomes such as higher income, lower costs or more stable production.

A disease detection model, for example, may be highly accurate in a lab environment, but farmers may not have smartphones, internet access, or the training needed to use it effectively. In the same way, an irrigation optimisation system may look promising in theory, but it may have little impact in places where irrigation infrastructure itself is weak or unreliable.

This means that good algorithm performance does not automatically translate into practical agricultural benefits. To understand the real value of AI, research needs to move beyond technical metrics and examine what happens after implementation: Do farmers continue using the technology? Does it improve productivity over time? Does it support better livelihoods? Without answering these questions, the promise of AI remains largely theoretical.

Lack of Socio-Economic and Political Perspectives and Absence of Theoretical Frameworks

One thing that is often missing from current research is a deeper discussion of the bigger picture. Many studies focus on whether AI tools are accurate or efficient, but pay less attention to how these tools fit into the real social and economic world of agriculture.

That is important because AI does not exist in a vacuum. When a new technology is introduced, its benefits are not shared equally by everyone. Farmers with more money, better access to infrastructure, or stronger connections to markets may benefit first, while smaller or poorer farmers may struggle to keep up.

There are also wider questions that go beyond productivity. For example, if AI systems collect large amounts of agricultural data, who owns that data? The farmers? The company providing the technology? Or the government? And if more farming decisions are influenced by digital tools, who ultimately has the power to shape those decisions?

AI can also reshape labour and rural life. It may reduce some burdens, but it may also change who gets work, who earns income, and who is excluded from new opportunities. Without looking at these broader issues, it is easy to assume that AI will automatically benefit everyone. But in reality, its impact depends heavily on local context, institutions, and existing inequalities.

That is why theory matters. It helps us move beyond the question of whether a tool works and ask a more important one: who actually benefits from it, and who might be left behind?

Geographic Bias in Research

Although many studies claim to focus on developing countries, most of the research is actually concentrated in Africa. While Africa is clearly an important region, this pattern may not simply reflect where the challenges are greatest. It may also reflect where the funding is.

Many AI and agriculture projects are supported by international donors, development agencies, and NGOs, and these organisations have long focused on Africa as a priority region. This means that research projects, pilot programs, and data collection efforts are more likely to take place there. As a result, researchers often follow the funding. This can shape both the location of studies and the types of problems being explored.

While this has generated important knowledge, it also means that other regions such as Asia and Latin America, receive far less attention. This creates an incomplete picture of how AI works in different agricultural contexts and limits the broader relevance of current research.

What This Review Suggests

This review reveals several key challenges in agricultural AI research in developing regions. Firstly, while many studies demonstrate high accuracy and performance under controlled conditions, this doesn’t necessarily translate directly into practical use by farmers, increased productivity or improved livelihoods. To properly assess the value of AI, it’s necessary to examine not only technological success but also its impact on real-world agricultural settings.

Secondly, existing research does not adequately address socioeconomic and political perspectives. The effectiveness of AI implementation is heavily influenced by infrastructure, institutions, inequalities, data ownership, and decision-making power; simply determining whether the technology works is insufficient. A stronger theoretical framework is needed to understand who benefits and who is left behind.

Thirdly, research is geographically biased, with a large concentration in Africa. This may be due not only to the severity of the challenges there but also to the influence of funding allocation by international aid and development agencies. This bias means that the potential of AI in diverse agricultural environments, including Asia and Latin America, is not being adequately captured.

Therefore, future research should not be limited to evaluating technical performance, but should take a more comprehensive approach that includes the effects in real-world situations, socioeconomic and political conditions, and regional diversity.

References

FAO (2017). The future of food and agriculture. [Online]. Available at: https://www.fao.org/global-perspectives-studies/resources/detail/en/c/458158/

Haub, C. K., Toshiko; (2013). ‘2013 world population data sheet’.

Okoli, C. (2015). ‘A guide to conducting a standalone systematic literature review’ Communications of the Association for Information Systems,37  p. 43.  DOI: 10.17705/1CAIS.03743.

This question matters because mountain communities worldwide face accelerating out-migration, particularly of young people.  While this “can help to reduce poverty and diversify livelihoods in mountains and beyond” it “has reached such a scale that depopulation and the seasonal absence of people of working age are widespread” with potential negative impacts “for the lives of those who stay behind, for the social fabric of mountain communities, and for the management of mountain ecosystems”[1].

From this source and others, the dynamics of mountain out-migration are well-established.  Limited economic opportunities, inadequate services and infrastructure, and the physical challenges of mountain life – increasingly exacerbated by the impact of climate change – drive people to leave.  But digital tech adds some complexities to this established pattern, from which I’ll explore three potential roles it can play: accelerator, enabler, and counter-force.

Digital accelerating out-migration

That problem of “inadequate services and infrastructure” applies to digital, with digital divides of both connectivity and skills being particularly strong for remote mountain communities.  These divides limit access to education, enterprise and public services, and make urban migration more attractive.[2]  For mountain youth, exposure through social media to urban lifestyles and opportunities can amplify dissatisfaction with limited local prospects, as they get an immediate – if distorted – sense of what life elsewhere looks like.[3]

Digital supporting out-migration

Digital connectivity supports out-migration by making it more sustainable for individuals and communities.  Mobile money and digital remittance platforms reduce transaction costs for migrants supporting families back home[4].  Video and audio calling maintains social bonds across distance in ways that were impossible a generation ago[5].  This digital infrastructure of migration means that leaving no longer requires the same degree of social rupture.  For mountain communities, this might preserve some economic and social links, but it also removes friction that might otherwise encourage people to stay or return.

Digital reversing out-migration

There is evidence that improved digital connectivity such as broadband can increase the reach, growth and employment of enterprises in rural areas[6].  Whether this could be linked to a reduction in out-migration is, however, questionable because of the strength of the accelerator and enabler effects.  One study from Spain, for example, finds “no evidence of a causal effect” between broadband growth and rural population change[7], while a similar study in China found “broadband creates ‘digital routes’ facilitating outmigration rather than ‘digital roots’ anchoring residents to rural areas”[8].

There’s a similar picture with the growth of interest in remote working; the idea that local people could undertake digital work in mountain communities, or even attract in-migration of digital nomads[9].  While this is both feasible and happening in some global North mountain communities, and digital nomads are setting up in well-connected beach resorts in the global South, there remain serious barriers to this as a strategy for most global South mountain regions[10].

Policy Priorities

A first priority is to strategically position mountain connectivity as integrated within a broader development architecture.  This means making services affordable and resilient, e.g. during extreme weather conditions.  It also means developing the skills to make productive use of the technology.  And it means recognising that connectivity alone does not reverse migration without complementary economic development and service provision.

A second priority is to envisage migration patterns as circular.  This means designing digital tools to help maintain productive links between mountain communities and those who have migrated out, potentially facilitating return migration or continued economic contribution such investment in local productive infrastructure and enterprise.

Of course, all this assumes that reversing mountain depopulation is a worthwhile goal and that abandonment to nature of (more remote) mountain regions should not be the intent.

Research Priorities

For researchers, key gaps remain around the actual versus assumed impact of digital connectivity on migration decisions in mountain contexts.  Much current understanding is extrapolated from lowland rural areas.  We lack longitudinal, causal evidence on which forms of digital investment genuinely reduce out-migration, which ones mainly facilitate mobility outwards, and which do both.  Understanding these dynamics requires mixed methods approaches that can capture both structural factors and individual decision-making across different mountain regions and cultural contexts.

A second gap is more political: research has not kept pace with the governance questions surrounding digital roll-out in mountain regions, including who bears the costs and who captures the gains since will also shape patterns of migration.  We thus need to know – alongside economic, social, and environmental impacts – what are the political implications of digital’s increasing interaction with mountain land, environment and labour markets.

Originally published in the Mountain Digital Futures newsletter: https://www.linkedin.com/newsletters/7400288732999733248/

Image source: https://migrationrightslab.org/migration-and-gender-in-west-africa-ghana-brief/

[1] https://adaptationataltitude.org/wp-content/uploads/2023/05/bachmann_et_al_migration_and_smd_2019_low_0.pdf

[2] https://www.defindia.org/wp-content/uploads/2023/07/Connecting-Himalayan-Communities-An-Issue-Brief_PRINT.pdf. For rural areas more generally, see: https://academic.oup.com/jcmc/article/21/3/247/4065369 and https://www.oecd.org/en/data/insights/statistical-releases/2025/07/digital-connectivity-expands-across-the-oecd-but-rural-areas-are-falling-further-behind.html 

[3] https://www.researchgate.net/publication/399601704_Rural-Urban_Migration_of_Young_People_in_High_Andean_Communities_in_Peru_Imaginaries_and_Practices_of_Vulnerability_and_Social_Advancement. Noting that social media is just continuing a longer historical trend of media-based exposure to urban lifestyles: https://onlinelibrary.wiley.com/doi/10.1111/j.1468-2435.2010.00627.x

[4] E.g. https://migrantmoney.uncdf.org/resources/insights/integrating-remittance-and-mobile-wallet-services-a-case-study-of-ime-pay-in-nepal

[5] https://ict4d.org.uk/wp-content/uploads/2023/09/ict4d-rwp-1-nepal-v5-1.pdf

[6] https://ruralinnovation.us/press-releases/new-research-proves-that-providing-fiber-broadband-experiences-to-rural-communities-boosts-income-entrepreneurship-and-business-investment/; https://www.sciencedirect.com/science/article/abs/pii/S0743016725003237; https://publications.iadb.org/en/access-credit-and-expansion-broadband-internet-peru

[7] https://onlinelibrary.wiley.com/doi/10.1111/tesg.12596

[8] https://www.iza.org/publications/dp/17752/digital-roots-or-digital-routes-broadband-expansion-and-the-rural-urban-migration-in-china

[9] https://pub.nordregio.org/r-2024-7-remote-work-in-rural-areas/r-2024-7-remote-work-in-rural-areas.pdf

[10] https://pmc.ncbi.nlm.nih.gov/articles/PMC11387329/; https://wol.iza.org/articles/does-working-from-home-work-in-developing-countries/long

Read the paper: https://doi.org/10.1007/978-3-031-56481-9_14

The most useful move is not simply re-listing “barriers and enablers” (which the literature already does) but classifying sharing behaviour into three empirically derived patterns: enthusiastic, circumstantial, and unenthusiastic CTI sharing.

The enthusiastic group is described through mechanisms that look institutional and procedural: government-led dissemination channels, defined processes (e.g., policing/national intelligence practices), and practical awareness of regulatory responsibilities such as the General Data Protection Regulation (GDPR). These are not “warm feelings” about collaboration; they are routinised structures that make sharing normal, legible, and defensible.

The headline theoretical contribution is circumstantial sharing. In this mode, practitioners may perform rigorous investigation and mitigation work and may even recommend external disclosure; however, disclosure becomes contingent on organisational dynamics, especially senior management decision-making. This reframes CTI sharing away from an individual-level “willingness” narrative and toward a governance narrative: “Who has the mandate to share, and under what conditions?”

By contrast, the unenthusiastic group is anchored in political economy. Participants link reluctance to the capitalist nature of the cybersecurity market, weak inter-organisational trust, and the perceived risk that shared intelligence will advantage competitors or even adversaries. The paper is strongest when it treats this as rational behaviour under market incentives, rather than a moral failure to collaborate.

Figure 1. Conceptual map of CTI sharing behaviours factors and key impediments

The “impediments” also land with practical clarity: fear of regulatory penalties, cost (both paying for CTI and operationalising sharing), and competitive/adversarial exposure. Notably, the language/standardisation issue (“a way of describing things that everyone agrees on”) adds an interoperability layer: even when organisations want to share, they may not share effectively.

Summary table: behaviours factors and drivers (from the paper)

A constructive critique is that the study’s empirical base (nine UK interviews) is well suited to theory-building, but the paper could more explicitly separate (a) inhibitors of decision-to-share, (b) inhibitors of ability-to-share, and (c) inhibitors of value-from-sharing. That would sharpen how frameworks like Structured Threat Information eXpression (STIX), Trusted Automated eXchange of Intelligence Information (TAXII), Malware Information Sharing Platform (MISP) and governance (policies, liability, reporting lines) interact, rather than sitting in parallel sections.

One way to operationalise this critique is to reorganise the reported impediments into three analytically distinct stages of inhibitors in the sharing pipeline: (a) what prevents authorisation to disclose (decision-to-share), (b) what prevents effective packaging and transmission (ability-to-share), and (c) what prevents recipients from realising actionable benefit (value-from-sharing), which also makes it easier to see where governance dominates versus where technical frameworks add leverage.

(a) Inhibitors of decision-to-share

These factors block the “authorisation” step: even when analysts want to share, the organisation chooses not to.

Liability and penalty risk: fear of regulatory consequences (and uncertainty about what is permissible) pushes leaders toward risk avoidance.

Reputational exposure: concern that disclosure signals weakness, triggers customer agitation, or invites scrutiny.

Competitive/adversary calculus: worry that sharing benefits competitors or provides adversaries with insight into detection/response posture.

Governance bottlenecks: unclear reporting lines, slow approvals, or management override (“management decides whether to disclose”).

Framework–governance interaction: standards like STIX/TAXII don’t resolve “who” can approve sharing or “what” legal basis is acceptable; policies, liability models, and decision rights dominate here.

(b) Inhibitors of ability-to-share

These factors block the “execution” step: the organisation may be willing but cannot share efficiently or safely.

Lack of shared language / weak standardisation in practice: inconsistent vocabularies, formats, and levels of abstraction make “good CTI” hard to package.

Process immaturity: absence of repeatable workflows for sanitisation, classification, Traffic Light Protocol (TLP), de-identification, and release.

Tooling and integration gaps: limited capability to structure, transport, and operationalise CTI (where STIX/TAXII/MISP would normally help).

Resource constraints: time, staffing, and financial costs of producing usable intelligence for others.

Framework–governance interaction: STIX/TAXII/MISP increase “technical interoperability”, but they depend on governance primitives (data classification rules, sanitisation policy, ownership, accountable roles) to function reliably.

(c) Inhibitors of value-from-sharing

These factors undermine the “payoff” step: even if CTI is shared, recipients may not get actionable benefit, reducing motivation to sustain sharing.

Low signal-to-noise / questionable relevance: shared CTI may be too generic, too late, or not aligned with the recipient’s context.

Poor actionability: indicators without context (TTPs, confidence, provenance, mitigation guidance) don’t translate into defensive action.

Asymmetric exchange: perceptions of “free riding” or unequal benefit erode trust and long-term participation.

Commercial data control: when CTI is treated as a product, access limitations and licensing constraints reduce collective value creation.

Framework–governance interaction: structured formats can improve enrichment (confidence, provenance, relationships), but “value” ultimately depends on community norms, reciprocity rules, and outcome feedback loops (what helped, what didn’t).

In short, governance determines “whether” sharing happens (decision-to-share), frameworks and operational controls determine “how” it happens (ability-to-share), and community norms plus CTI quality determine “why” it is sustained (value-from-sharing).

Call to action: If you work on digital development, policy, or organisational governance, consider auditing your CTI-sharing pipeline against the three stages (decision-to-share, ability-to-share, value-from-sharing) and identify where governance, process, or tooling is the binding inhibitor. You can read the paper via the DOI link above, and I’m happy to discuss practical implications or collaboration opportunities. Feel free to email me a.n.muhammad@edu.salford.ac.uk or message me on LinkedIn Abubakar Muhammad Nainna, PhD