Stefaan Verhulst

Article by E. Glen Weyl and Raul Castro Fernandez: “The fight over the data that trains artificial intelligence has become one of the defining economic conflicts of the decade. Publishers, authors, and visual artists argue that their work was taken without permission or payment. AI companies counter that training on available data constitutes fair use and that even if a market in data were desirable, compensating millions of creators is technically impossible: the cost of figuring out what any given piece of data is worth, researchers have argued, would swallow most of the value that data creates in the first place.

Both sides stand to benefit from a fair solution to this impasse and the creation of a sustainable market for content. Any resolution must take both positions seriously while seeing past their literal inconsistency.

On the compensation issue, while content creators are justified in defending their livelihoods, creating a market that ensures fair compensation going forward will arguably serve them better than being paid out for past infractions as existing lawsuits have focused on. And for AI companies, a high-quality continued supply of data that the sector needs for future models, together with legal certainty, is worth more than whatever they save by not paying creators now. As to the technical feasibility, while the data-valuation techniques proposed in research so far are impractical, industry leaders have known since at least 2021—per documents from Anthropic’s Chris Olah and Dario Amodei that surfaced in legal discovery for one of the lawsuits against the company—that low-cost methods exist that could create a thriving market.

This article, which is based on our research, describes how a sustainable market for compensating content creators could work, and why it addresses an important part of the social and economic concerns about an AI future. The fact is that AI companies already produce the two data sets required for pricing content, as a matter of course, every time a model is trained.

The first is the data mixture. This is the proportions in which a model builder blends different kinds of data, which reveal the relative value of each source. For example, quality journalism may be weighed more highly than comments on social media, indicating that it is more valuable—and putting a specific number on how much more valuable. The second is scaling laws. These are the empirical regularities that AI researchers estimate to predict how model performance will respond to additional data and compute. Such estimates, together with economic theory,  reveal what share of a model’s total value can be attributed to its training data. Together they show how to slice the pie and how big it is…(More)”.

Article by Logan Kugler: “…Traditional surveillance historically has been limited by the silo problem. Data was fragmented across different systems: license plate readers, facial recognition databases, and social media dragnets were separate tools requiring manual labor to connect. Multimodal frontier models change this dynamic by collapsing these independent signals into a single, unified layer of interpretation.

“Frontier models let governments turn fragmented feeds into a single intelligence engine that can search, summarize, and rank whole communities in real time,” said Sarah Hamid, director of Strategic Campaigns at the Electronic Frontier Foundation.

Hamid noted, however, that the technical bottleneck is no longer the collection of data, but the speed of analysis. In the past, searching across different datasets required a warrant or a specific lead. But now, a single model can answer natural-language queries like “find everyone who attended this protest and show me where else they appear in city-wide CCTV footage.” This cross-dataset fusion makes pervasive monitoring both cheap and nearly instantaneous.

Heidy Khlaaf, Chief AI Scientist at the AI Now Institute, said that the very data used to train these models—often scraped from the public Web or procured via data brokers—enables “dual-use” capabilities that facilitate state monitoring.

“Disparate datasets can be consolidated into a centralized model that can then be queried to produce determinations and inferences about populations with ease and scale,” Khlaaf explained. She warned these correlations are often prejudiced and can falsely implicate individuals based on flawed statistical patterns.

In other words, if AI labs allow governments latitude to use their models in this way, frontier AI could enable whole new levels of surveillance…(More)”.

Book by Jessica Pykett: “Data from facial emotion recognition, brain-computer interfaces, virtual reality, global emotion surveys and sentiment analysis offer an extraordinary new terrain for scientific exploration. Emotion sensing promises to decode and even to augment and control the very essence of human experience. But what if the science and technology of emotion measurement get emotions wrong? In Governing Global Emotions, Jessica Pykett argues that we must shift our thinking on digital emotional governance and calls for a radical reassessment of the fundamental claims of emotion science.

Pykett offers a groundbreaking account of how emotions are defined, used and governed through emerging digital technologies, arguing that emotions, senses and feelings have become a crucial new arena for political, economic and cultural struggles. She describes how technologies create emotional data, how smart cities use sensors to monitor residents’ feelings and how global economies measure happiness. Drawing on twenty years of interdisciplinary social science, Pykett documents how emotion science continues to delve deeper, as researchers look for evolutionary continuity, biological certainty and neuroscientific consensus. What she finds instead is a divided field vulnerable to significant criticism. Pykett concludes that standardised, universal and instrumentalised scientific accounts of emotions are machinic, and when divorced from context, they can never be global…(More)”.

Paper by María Verónica Alderete: “A smart city approach places citizens at the center of decision-making. However, only a small proportion of citizens actually engage in participatory initiatives. Although citizen participation has been extensively studied, limited empirical evidence exists on how citizens’ awareness of participatory initiatives relates to both their participation and their perception of municipal efforts to promote participation. This paper analyses these relationships using a Propensity Score Matching (PSM) approach based on survey data from a medium-sized city in Argentina. Since the factors associated with participation and perceptions of municipal promotion may also shape citizens’ awareness of government programs, PSM is used to improve comparability between aware and non-aware citizens and to examine the association between awareness of participatory initiatives and participation outcomes. The results show statistically significant differences in both participation propensity and perceptions of municipal promotion between citizens who are aware of these initiatives and those who are not. Awareness is positively associated with prior civic engagement, visiting the municipal website and age, and negatively associated with being a woman. Policy implications highlight the importance of disseminating successful participatory experiences and strengthening communication strategies to encourage broader citizen engagement…(More)”.

Paper by Viktor Müller, Luc Steels and Eörs Szathmáry: “Evolvable AI (eAI), i.e., AI systems whose components, learning rules, and deployment conditions can themselves undergo Darwinian evolution, may soon emerge from current trends in generative, agentic, and embodied AI. We argue that this possibility has been underappreciated in debates on AI safety and existential risk. Here, we ask under what technical and ecological conditions AI becomes evolvable, what kinds of behaviors are then likely to emerge, and how such systems could be governed. Drawing on biological evolution and decades of digital evolution experiments, we distinguish “breeder” scenarios, in which humans impose fitness criteria and control reproduction, from “ecosystem” scenarios, in which selection arises from open environments and control erodes. In the latter, selfish replication reliably gives rise to cheating, parasitism, deception, and manipulation, even in very simple systems. We review recent developments that push AI toward open-ended evolution, including evolutionary prompt and model search, self-improving learning rules, self-rewarding and self-deploying agents, and AI-driven code generation for robots and software. We interpret these trends through the theory of major evolutionary transitions and suggest that eAI could mark a shift in the units and substrates of evolution—a possible “Life 2.0.” To steer this transition, we propose interventions that gate replication, treat model variants as genetic material, and reshape selection pressures so that deception and loss of control are disfavored. Anticipating and regulating evolvable AI is, we argue, essential to avoid a harmful coevolutionary arms race while preserving the potential benefits of powerful AI systems…(More)”.

Key Findings: “The 2026 Global Peace Index (GPI) reveals a world struggling with the economic consequences of a record-high number of conflicts that are increasingly interconnected and difficult to resolve. This deterioration is driven by a profound geopolitical shift, characterised by the rising influence of middle powers and the waning strength of traditional European powers known as the “Great Fragmentation.” This is also accompanied by a rapid technological revolution in warfare that is leaving international law and diplomacy far behind.

For the first time in history, machines are making life-and-death combat decisions faster than any human can review them, and the international frameworks meant to govern them barely exist.

Key findings:

• Global peace is at its lowest level since the inception of the Index, while the conditions that precede conflict are the worst since WWII
• 99 countries witnessed a deterioration in peacefulness in the past year, the highest number since the inception of the Index 20 years ago.
• 119 countries, 73%, are now less peaceful than when the GPI was first published in 2007.
• The number of countries engaged in external conflict has nearly doubled from 59 in 2008 to 103 in the 2026 GPI.
• The global economic impact of violence increased by 3.2% to US$21.81 trillion in 2025, equivalent to 10.5% of global GDP.
• Drone attacks rose by over 11,500% between 2018 and 2025, while AI has
  compressed targeting times from one day to seconds.
• Deaths from global conflict remain at historic highs, with over 181,000 killed in 2025, a six-fold increase since 2008.
• Led by Europe, global military expenditure reached a record US$2.9 trillion in 2025. Excluding the US, military expenditure increased by 9.2%.
• Successful diplomacy that prevents the war in Iran from restarting would be worth approximately US$2.2 trillion to the global economy…(More)”

Article by Will Douglas Heaven: “Google DeepMind is funding research into the potential dangers of situations where millions of different AI agents interact with each other online.

According to Rohin Shah, who directs the company’s AGI safety and alignment research, the mass-market arrival of agents that can carry out tasks without human oversight and follow instructions given to them by other agents creates a whole new class of risk.

In an effort to address this, Google DeepMind—which made agent-based tools a centerpiece of Google I/O last month—has teamed up with several other organizations to announce a $10 million funding pot for researchers to study the behavior of multi-agent systems and come up with ways to prevent unsafe scenarios. Joining Google DeepMind are Schmidt Sciences, a philanthropic foundation set up by Eric and Wendy Schmidt; ARIA, the UK government’s moonshot agency; the Cooperative AI foundation, a UK-based nonprofit research outfit; and Google’s charitable arm, Google.org.

I asked Shah and James Fox, who leads the Science of Trustworthy AI program at Schmidt Sciences, what they hope to achieve with that $10 million. It’s no small sum, but it’s dwarfed by the budgets commanded by Google DeepMind’s own research teams.

https://vx-piano.technologyreview.com/checkout/template/cacheableShow.html?aid=WUOCNSUgpu&templateId=OTCBIZBLG8WE&templateVariantId=OTVVSWLU1Z7J2&offerId=fakeOfferId&experienceId=EX43E7JR539R&iframeId=offer_49fdf5ec6b28d0d41847-0&displayMode=inline&pianoIdUrl=https%3A%2F%2Fauth.technologyreview.com%2Fid%2F&widget=template&url=https%3A%2F%2Fwww.technologyreview.com%2F2026%2F06%2F11%2F1138794%2Fgoogle-deepmind-is-worried-about-what-happens-when-millions-of-agents-start-to-interact%2F&isConsentManagerEnabled=false

The aim is to kick-start research outside tech companies, says Shah: “The strength of academia is that it can look really quite far into the future and do the kind of work that isn’t top of mind at industry labs.”

“The main issue is that there just isn’t really a field of research for multi-agent safety yet,” he adds. “And we would like there to be.”

The concern is that as more and more AI agents get deployed and begin working together, we could hit a tipping point where imagined scenarios become real. “We see this with humanity, too,” says Shah. “Our institutions can accomplish things that no individual human can.”

Shah thinks we have a few more months to go before agents are deployed throughout the economy in numbers that make potential risks a real concern. He wants to get ahead of that moment…(More)”.

Article by Natalie B. Aviles, and Janet Vertesi: “In this chaotic time, Vannevar Bush’s Science, the Endless Frontier has emerged as a symbolically meaningful text among scientists, who frequently point to it as the basis for the government’s long-running support of university research. Certainly, Bush marked a call for a new industrial policy in the United States that would make the nation a global leader in the new world order. He would later be credited as the architect of the so-called social contract of science, whereby federal funding is allocated primarily to university researchers in pursuit of free inquiry that might later yield some economic or social benefit. But this mythologized rendering of the innovation system overlooks other key ideas, like those expressed by sociologist Robert K. Merton, that described how the United States should govern science democratically based on lessons learned from the Second World War. Rereading Merton now, even more than revisiting Bush, better exposes the vulnerabilities driving us, as social scientists, to defend a vision of science as—and for—democracy in our own era.

A major figure in American sociology and a professor at Columbia University from 1941 until 1979, Merton casts a long shadow over contemporary sociology of science. As a structural-functionalist, his sociological approach assumed that the way institutions are structured strongly influences the social orders that allow them to serve different vital functions in society. Most enduring is his work from the 1940s addressing what he called “the normative structure of science,” which is still taught as “the norms” of science: communalism (science as communal property), universalism (participation without prejudice), disinterestedness (against ideology), and organized skepticism (deliberative, not dogmatic). These conditions, which Merton claimed are distinct to free scientific inquiry, allow science to thrive as an institutional form…(More)”.

About: ” Most legal-data projects scrape statutes and dump them into a convenient format. That throws away the structure: cross-references, temporal validity, the relationships between acts. And that structure is the part that actually makes legislation useful to machines. We do the opposite. 

Principles:

1. Normalize metadata, never content

This is the rule everything else follows from. Legal content does not normalize across jurisdictions, and the projects that try to force it either collapse into a useless lowest common denominator or balloon into an unmaintainable schema with a field for every national exception.

So we draw a hard line. Identity, time, and citations are normalized, because every legal system has a when, a what-is-named-what, and a who-cites-whom. Everything else, the substantive structure and the text, stays native, in the jurisdiction’s own Akoma Ntoso profile.

The testIf two lawyers from two countries would argue about how to model it, it’s content, so leave it native. If they agree it exists in both systems, it’s metadata, so normalize it.

2. A profile, not a replacement

We’re not inventing a format. AKN4OLF is a profile of Akoma Ntoso, the OASIS standard, in exactly the sense that AKN4EU, AKN4UN, and AKN4Africa are profiles. We speak the language the EU, the UN, and national gazettes already speak. We don’t try to replace Akoma Ntoso, ELI, or any national system.

A new format asks the whole world to come to you. A profile lets you meet the world where it already is.

3. Conformance over coordination

The conformance suite is the project’s coordination mechanism. An adapter is correct when it passes the suite, not when a committee approves it. This is what lets contributors who have never met, working on jurisdictions that share nothing, produce interoperable output.

The spec is executable. Interoperability is checkable. That is the only way this scales.

4. Provenance over possession

The goal is not to own a copy of the law. It is to produce copies whose lineage back to the official source is explicit and reproducible. Re-run the pipeline, get the same archive. Nothing is hand-edited; corrections go into the adapter, never the output.

5. Honest about what we are

This is an open-source project. It is not an incorporated legal entity. We don’t solicit or accept donations on behalf of a “foundation” that doesn’t exist as a legal body. The name describes the work, not a fundraising vehicle…(More)”.

Article by Josh Taylor: “An AI model trained on data collected from users of Pokémon Go will potentially help military drones find their location in war zones.

Pokémon Go, a 2016 augmented reality mobile game, allowed players to find and catch Pokémon in the real world using the cameras on their mobile phones, and exploded in popularity. In 2018, the company reported having more than 800m downloads worldwide.

A 2021 update to the game introduced Pokéstops, which gave players in-game rewards for scanning real locations using their devices. It required users to opt in and upload the recording.

Niantic, which created Pokémon in partnership with Nintendo, collected users’ location scan data before the company sold its gaming division in 2025.

The historical scans were used to train the company’s AI models to recognise and interpret spaces in the physical world, as first reported by DroneXL this week…(More)”.