Stefaan Verhulst
Blog by Milena Jael Silva: “Open data remains a democratic achievement, and a necessary one. It widened access to information once held inside ministries, laboratories, firms or platforms, and gave public actors, researchers and citizens stronger grounds for scrutiny and reuse. Yet access is no longer where the governance problem ends.
Data become powerful through what happens after publication. They are cleaned, linked, modelled, ranked, mapped and translated into decisions. A portal may release a signal, while the machinery that turns that signal into authority sits elsewhere. The sharper question is therefore not only whether data are open. It is whether public institutions and relevant communities retain the capacity to make, inspect and contest the claims made from them.
Take a coastal municipality. It publishes drainage maps, flood records, shoreline observations, land-use data, infrastructure files and social vulnerability indicators. The portal is functional, the licences permissive and the metadata adequate. By conventional open data standards, the municipality appears compliant. An external provider then combines those public signals with remote sensing, proprietary modelling and a hosted interface. It sells the municipality a climate-risk dashboard that ranks neighbourhoods, assigns exposure scores, proposes investment priorities and makes some zones appear less viable. Elected officials can see the colours, planners can export the maps and consultants can cite the ranking. Yet the municipality cannot reproduce the classifications, inspect all thresholds or fully argue with the uncertainty. Who, at that point, governs the coast?
This is closed inference: a post-publication asymmetry in which data may be open, shared or technically accessible, while the capacity to transform them into authoritative interpretation remains concentrated, closed or insufficiently accountable. It appears downstream, where accessible signals become classifications, forecasts, priorities and a working basis for public decisions.
At this point, the issue is not that information is hidden. The data may circulate, the dashboard may be visible, and the report may be public. Still, the authority to say what the data mean may sit inside an analytical infrastructure that public actors do not command. Transparency shows that information exists; it does not necessarily reveal how significance is assigned, how uncertainty is handled, or how an output becomes a reason to act…(More)”.
Report by New America: “Ballot initiatives are often criticized as too expensive and vulnerable to wealthy interests, but cost alone is a poor measure of democratic value. Drawing on campaign finance data, academic research, and interviews with practitioners, this report examines the costs, benefits, and relative value of statewide direct democracy in the United States. We find that initiative campaigns can be expensive, particularly in a handful of large states, but their costs are often comparable to lobbying and candidate campaigns. Money influences the initiative process, especially at the ballot-access stage, but does not reliably determine outcomes. Rather than restricting direct democracy, policymakers and organizers should pursue reforms that lower barriers to participation, strengthen voter information, and preserve initiatives as a viable pathway for citizen-led policymaking…(More)”.
Blog by Lea Gimpel: “The open web is undergoing a systemic structural transformation. Today, digital public goods (DPGs) and the broader knowledge commons face a critical challenge: large-scale, automated extraction by AI crawlers and scrapers operating without reciprocity. This leads to ballooning infrastructure costs for often volunteer-run, underresourced open projects, and the erosion of trust in open knowledge, among other challenges. DPG product owners have to make delicate decisions to protect their resources while keeping them as open as possible.
To help stewards of open resources navigate this complex landscape, the DPGA Secretariat is proud to share two complementary assets developed in consultation with our community:
What it is: A position paper and set of recommendations compiled directly by DPG product owners across major open initiatives like Wikipedia, Open Food Facts, Storyweaver, Govdirectory, The Turing Way, among others.
Core Focus: This note describes the problems faced by DPGs firsthand and outlines practical steps for engaging with commercial AI entities to ensure data users respect licenses, robot policies, and terms of service, and contribute back to the ecosystem, while also detailing technical mechanisms to protect against AI exploitation.
What it is: An overview of available mechanisms to protect against AI exploitation and how they map against the current open definition, and, by extension, the DPG Standard. The playbook’s aim is to deepen understanding and conceptual clarity around the tension between maintaining open access and ensuring the survival of an open resource.
Core Focus: The playbook helps understand the different layers of interventions available to DPG product owners, ultimately showing that code-level mechanisms are a first line of defence, but solving the challenge of AI exploitation requires ecosystem-level solutions at the policy, legal, and governance levels.
Layers of Defence: It introduces a 6-layer “defence and reciprocity pyramid” classifying technical, legal, regulatory and institutional measures..(More)”
Paper by Talia Caplan and Bilal A Mateen: “In 1600, Queen Elizabeth I granted the East India Company a royal charter and a monopoly over trade across much of the known world. For more than a century, the relationship was mutually advantageous: the Crown received revenue, access to rare commodities, and unparalleled geopolitical reach; the Company received protection, legitimacy, and the coercive backing of a state. By the mid-18th century, it became something very different. The Company maintained its own army, territories, and foreign policy, and increasingly operated in ways that were counter to the objectives of the Crown, resulting at times (due to its unconstrained, singular objective of maximising profits) in the deaths of millions.
Today, the world’s largest technology firms are at the precipice of a similar transition: towards becoming firms that function as corporate sovereigns whose power derives from the control of indispensable digital infrastructure and the algorithms that shape the attention economy, rather than from physical territory.2 The concentration of power is stark. A single firm (Nvidia) supplies between 80 and 90% of the chips on which advanced artificial intelligence (AI) models are trained; a handful of hyperscale providers (Microsoft Azure, Amazon Web Services, Google Cloud) operate the data centres in which those chips run; and a small number of laboratories (eg, OpenAI, Anthropic, Google DeepMind, DeepSeek), almost all American-owned, use those data centres to produce the frontier AI models on which others build.
These layers are less separate than they appear. A dense web of cross-investment now binds them, with chipmakers taking equity in the laboratories that buy their processors, and cloud providers and laboratories acquiring stakes in one another while committing to purchase each other’s services; analysts have estimated that interlocking arrangements of this type are worth more than US$800 billion. The effect is a tendency towards vertical integration achieved through finance rather than a formal merger, drawing nominally competing entities into a single, mutually dependent interest.
We expect that two structural features will characterise AI firms’ transition to pseudo-sovereign status. These firms will increasingly occupy every position in their own oversight, building the systems, funding the safety evidence (based on evidence from a preprint), staffing the advisory bodies, and drafting the standards meant to constrain them. They will also become jurisdictionally mobile in ways territorial regulators and sovereign governments are not, using digital infrastructure and complex corporate structures to obfuscate the residency of data and software in an effort to escape true oversight.
In that context, a health system that adopts a frontier AI solution for triage, imaging, or clinical decision support—such as the Horizon 1000 initiative seeks to roll out in Rwanda—does not necessarily acquire a substitutable tool but rather creates a dependency on a supply chain it neither controls nor can reproduce. The risks are especially acute in the global health context because upfront philanthropic support, especially when provided by the frontier AI companies that stand to benefit most, increases the likelihood of three specific risks…(More)”.
Paper by Polly Mackenzie et al: “There is now a broad consensus among practitioners, funders, and policymakers that Britain is paying the costs for a decades-long depletion of civic life. A great deal of excellent work has begun to reverse this decline, but a key question remains unanswered: how do we fund the work of civic renewal at the scale that is needed?
The Marshall Plan for Civic Life is a landmark programme of research and deliberation, convened by Kinship Works and Demos, to answer this question – identifying viable financing mechanisms, and developing an overall funding architecture. The initial work has been supported by This Day and the Joseph Rowntree Foundation but we are building a coalition – adding partners and modules to deepen our understanding of the problem and possible solutions.
In this opening Discussion Paper, we share initial reflections on a range of financial mechanisms that look promising. The paper is intended to stimulate discussion and we would welcome feedback. This will be followed in Autumn 2026 with a broader paper, describing a potential architecture and supporting philosophy for a Civic Marshall Plan, alongside an estimate of the scale of the funding gap that needs to be filled…(More)”.
Paper by Kevin A. Bryan & Joshua S. Gans: “AI predicts; humans use its predictions to make decisions. These predictions are combined with human verification and analysis, queries to other statistical models, and so on. The economic value of an AI, therefore, depends on how it interacts with the surrounding decision environment. We describe the value of AI as part of this “composite experiment” where AI makes a coarse prediction of the state of the world, show what this means for optimal model training via a geometric argument, explain why optimal training can be discontinuous in economic variables, and study how heterogeneous users or monopoly model trainers affect these results. In particular, maximizing the unconditional accuracy of AI predictions is generally suboptimal…(More)”.
Report by UNESCO: “Media freedom and journalism are under attack around the world. AI-fuelled disinformation is confusing and polarizing audiences, government censorship is sharply rising, physical and online violence towards journalists is increasing, and the journalism business model is failing. At the same time, funding for public media is declining, and drastic cuts to international aid and media development budgets have reduced the amount of reliable, independent journalism available to audiences globally.
This report explains why these cuts and attacks matter – and what is at stake when journalism declines and disappears. It synthesises the latest academic research on the value of journalism and its role in economics, national security and crises. The report is focused on public interest journalism, that is: reporting that is independent, accurate and ethical, and that seeks to inform the public about important issues affecting their lives, enable debate, and hold power to account.
The evidence collected here demonstrates that this journalism can have a profound and positive impact on societies and individuals globally. Its role supporting democracy is very well established: journalism shares the information citizens need to cast meaningful votes, it is a check and balance on power, and it acts as a conduit between citizens and elected officials.
But journalism’s impact goes far beyond that. As the evidence in the report indicates: Journalism is an economic enabler: it can reduce corruption, lower the cost of doing business, encourage the fair and transparent distribution of resources, and support economic growth and development. Journalism supports national security: it makes societies more resilient to disinformation and the interference of malicious actors, and reduces the risk of conflict and war.Journalism improves the response to crises: it provides information that saves lives, and improves preparation and response to disaster and crises.Journalism is also a surprisingly cost-effective way to achieve these positive outcomes, and it offers a high return on investment. As the evidence in this review shows, every $1 spent on journalism, can result in more than $100 in savings to the public through improved public services and reduced corruption. Journalism’s most significant impact, however, is that it plays a preventative and protective role – as part of a healthy information infrastructure…(More)”.
Article by Camille François et al: “…It has been challenging to define openness in the context of FMs, as definitions of “open source” software do not easily translate to AI systems. However, shared nomenclature is essential to developing shared understandings, norms, benchmarks, and best practices. By shared nomenclature, we mean establishing common terms across researchers, developers, policymakers, and civil society organizations working to unlock the benefits and mitigate the risks of FMs. We use AI systems here as shorthand for systems built with FMs, while recognizing the limitations and critiques of the term AI. This work is also necessary to illuminate the range of potential design choices around openness throughout the AI stack, and to ensure that this conversation moves beyond a narrow focus on model weights.
In this article, we survey existing approaches to defining openness in AI models and systems. We also propose a descriptive framework to evaluate how each component across the FM stack contributes to openness, enabling normative definitions of openness in AI. We intentionally do not present a definitive list of requirements for openness. This work builds on a February 2024 workshop convened by Mozilla and the Columbia Institute of Global Politics, which brought together more than 40 leading scholars and practitioners working on openness and AI. These individuals—spanning open source AI startups and companies, nonprofit AI labs, and civil society organizations—focused on exploring what open should mean in the current era of foundation models…(More)”.
Paper by Neil D Lawrence and Jessica K Montgomery: “Public dialogues have produced clear demand signals for AI. These dialogues imagine innovations that improve our shared wellbeing and prosperity while developing under democratic control. AI development has largely advanced along another trajectory. We argue that this gap is a structural feature of an innovation system whose incentives are captured by the attention economy. Closing it requires a different driver of the innovation cycle. We propose an attention reinvestment cycle, in which efficiency gains accrue as freed time that can be invested in community innovation, with frontline professionals adapting and sharing tools that meet priority needs. Operating the cycle at scale requires institutional infrastructure—dialogic, absorptive, and distributive capacities—that the existing science-policy system has yet to develop…(More)”.
Release by University College London: “A major new resource that provides one of the most comprehensive pictures yet of what people are eating around the world has been introduced in a new study by a UCL and University of Oxford researcher.
From obesity and heart disease to climate change and food affordability, many of today’s biggest challenges are shaped by what we eat. But there is a surprisingly basic problem facing researchers and policymakers: We often don’t know with enough accuracy what people are actually consuming.
The new study, published in Nature Food and authored by Professor Marco Springmann (UCL Institute for Global Health as well as the Environmental Change Institute at the University of Oxford), introduces the Global Dietary Database for Impact Assessments (GDD-IA), which is freely available through an interactive online explorer, allowing users to investigate dietary patterns across countries and over time.
The GDD-IA combines information on food production, food waste, dietary surveys and human energy requirements to estimate what people eat from 1990 to 2020. It includes detail by age, sex and whether people live in urban or rural areas.
The resource has been designed to support research into some of the world’s most pressing questions: How do diets affect human health? What impact do they have on climate change and the environment? How affordable are healthy and sustainable diets for different populations?..(More)”.