Stefaan Verhulst
Paper by Jessica M. Silbey and Woodrow Hartzog: “The term “AI slop” has become popular to describe the output of generative AI systems seen as voluminous, low quality, or the result of little effort. When AI-generated music and videos flood platforms, they are called slop. Peer-reviewed journals and legal tribunals are drowning in low-quality and low-reliability AI slop submissions. Employees are seen to be producing mountains of slop in their reports and communications with each other. The term has inertia and heft, and the phenomenon has significant consequences. Most of them are not good.
But the boundaries of “AI slop” and its usefulness in policy discussions are not clear. What distinguishes slop from other machine-aided human authored work? Does the concept of “AI slop” help policymakers and researchers better understand a problem or identify a solution? This Article interrogates the concept of AI slop to identify its essential features and document how it operates in three cultural domains where it is commonly deployed—creativity, research, and work.
We propose a provisional definition of slop as any output of a probabilistic automated system produced with little exertion that asymmetrically burdens recipients and tends to degrade cultural domains. We also conceive of AI slop along a spectrum. Works can be more or less “sloppy,” depending on the relative presence of its three constituent parts: negligible exertion, asymmetrical imposition, and domain degradation. We conclude that AI slop is a useful and important frame in AI discourse and policymaking because it provides a shorthand to distinguish harmful, inferior AI output from meaningful human authorship and because it highlights the pathologies of generative AI for social institutions…(More)”.
Paper by David J. Deming, Katrine V. Løken, Alexander Willén & Yaling Xu: “Why do we have so many meetings? Few workplace features are so scorned, yet seemingly so necessary. This paper provides the first large-scale economic evidence on workplace meetings using an original survey of more than 9,000 workers linked to matched employer–employee administrative data from Norway. We show that meetings are both common and costly, consuming an average of 12 percent of work hours and 14 percent of firm wage bills. Planning, problem solving, information sharing, and project coordination account for the majority of meeting activity. High-paying and high-revenue firms devote more resources to meetings despite facing a substantially higher opportunity cost of employee time. Meeting frequency and intensity are positively related to worker wage growth. Workers in meeting-intensive firms report greater on-the-job learning, and interactions with more senior colleagues are associated with stronger wage growth, suggesting that knowledge transmission within firms is an important mechanism. Meetings are the broccoli of work – widely disliked, but probably good for us anyway…(More)”.
Paper by Michael Dalton, Lisa B. Kahn & Andreas I. Mueller: “We ask whether online job postings capture U.S. job vacancies, matching near-universal Burning Glass (BG) postings to the representative Job Openings and Labor Turnover Survey (JOLTS). BG converged toward JOLTS over 2007–2022, and its representativeness varies across establishment characteristics. We reweight BG to align postings with JOLTS openings along observables, and assess impacts on prior BG findings. Because large establishments account for most openings and are well represented, the level and cyclicality of skill demand survive our adjustment. But BG overstates labor market concentration (the Herfindahl-Hirschman Index) nearly threefold, because small and mid-sized establishments post less online…(More)”.
Paper by Mustafa Karatas et al: “Hybrid-capture wastewater sequencing has shown promise for episodic viral epidemiology, but its value for multi-year, multi-pathogen, strain-resolved surveillance benchmarked against clinical data remains unclear. Here we applied hybrid-capture viral metagenomics to 127 weekly 24-h composite influent samples from a single sentinel wastewater treatment plant in Leuven, Belgium, collected over 31 months, and linked wastewater time series to national clinical indicators of multiple pathogens. Our approach recovered a genome-resolved target-enriched virome spanning 339 viral species and 715 strains, captured seasonal community turnover and showed that rotavirus A in wastewater typically preceded clinical surveillance by ~1 week. It also identified signals beyond routine wastewater programmes, notably a parvovirus B19 surge starting in April 2023, about 1 year before the national public health alert. Overall, here we show that longitudinal hybrid-capture metagenomics can extend wastewater surveillance from detection to strain-resolved, multi-virus epidemiology and can provide early circulation signals for clinically relevant viruses…(More)”.
Article by Adam French: “…Bank of England Chief Economist Huw Pill reveals the Monetary Policy Committee (MPC) is using AI, web-scraping and real-time financial data to improve its understanding of the economy – with information collected by Moneyfacts among the private-sector data used to inform its assessment.
The Bank of England is turning to artificial intelligence and new sources of real-time financial data to identify economic signals that could otherwise be missed, Chief Economist Huw Pill has revealed.
Pill says the Bank is already using AI models to extract quantitative signals from qualitative information, including corporate reports, survey responses and conversations between the Bank’s regional agents and businesses.
The Bank is also using web-scraping and other technologies to develop new sources of information for monetary policy, while data science is allowing policymakers to analyse large datasets. This includes anonymised real time data on individual mortgages and bank accounts.
“We have been using AI models to draw more quantitative signals from the qualitative data coming from corporate reports, survey responses, or agents’ conversations with their business contacts across the country.”
Pill says the technology is helping the Bank distinguish genuine economic signals from louder and more volatile data. One of the biggest challenges facing modern monetary policymakers. “Extracting signal from noise has become even more important (and difficult) over the past few years.”…(More)”.
Article by Stefaan Verhulst, Jordan Harris, Andrew J. Zahuranec, Maui Hudson, Chantal Kamgne, and WarīNkwī Flores: “Indigenous communities face two related threats from AI: marginalization from digital advancement and extraction of their cultural knowledge and language without consent or benefit. One way to address these challenges is to create pathways for Indigenous communities to exercise meaningful control over how their data is collected, governed, and used.
Ahead of the International Day of the World’s Indigenous Peoples, The GovLab hosted a webinar featuring Indigenous experts Chantal Kamgne (Localizzz / EngageAfricaNLP), WarīNkwī Flores (Kinray Hub / Kara::Kichwa), and Maui Hudson (Te Kotahi Research Institute / Local Contexts) to explore how data commons can offer a middle path toward true data self-determination. The webinar was moderated by Stefaan Verhulst (The GovLab), and Guilherme Canela de Souza Godoi (UNESCO) provided closing remarks.
Three major themes emerged from that discussion, which we detail below: the ability of data commons to serve as a tool for self-determination; how there can be one-size-fits-all solution to governance because such arrangements become a power instrument of top-down pressure; and the need for large institutions to “walk the talk” when it comes to supporting Indigenous communities to take control of their data…(More)”.
Blog by Geoff Mulgan: “In this piece I describe what I call ‘knowledge twins’, combining data, evidence, foresight, tacit knowledge and more, to enable governments to make better decisions and to learn more effectively. I show how these can build on many existing tools, from evidence syntheses to knowledge graphs, digital twins to systems maps, and how they can help policy-makers realise the full potential of AI. I show how they could evolve over the next decade to make intelligence useful and used, on everything from economic growth to decarbonisation, mental health to poverty, bio and climate risks to the future of care, and become essential cognitive infrastructures for governance.
Understanding what causes what – the grounding for everything governments do
Governments have always been interested in intelligence. But in the past that usually meant secret intelligence about enemies and threats. In the 19th and 20th centuries that interest broadened to include intelligence about the economy, health and population, alongside a recognition that shared intelligence can sometimes be much more valuable than intelligence that is hoarded.
Every part of any government rests on implicit or explicit models of how the world works. If we pass this law, spend money in this way or introduce (or cut) this programme, these are what we think the effects will be. Pre-modern governments were different, as are contemporary autocrats: it might be enough for a law to reflect theology, or the whims of a king or President, or just hope. But in the modern era it’s hard to see why we as citizens would want our money spent, or our freedoms curtailed, for programmes not based on serious assessment of whether they will succeed.
For the same reason much of politics revolves around promises that actions will have predictable effects. So, if a party or candidate promises that their actions can achieve desirable ends such as higher growth, better health, lower crime or immigration, we expect them to be able to say how and we expect to be able to examine their homework to see if it’s convincing.
These causal models get rough and ready scrutiny in an election campaign, and sometimes outside them, mainly through interviews on broadcast and in print media, and on some topics, such as fiscal plans, there are well-resourced independent institutions (like the UK’s Institute for Fiscal Studies or Resolution Foundation) able to scrutinise the assumptions. Indeed, economics is a field where shared intelligence is widely accepted: governments share lots of data, analyses, models and forecasts and have to persuade markets and observers that they know what they’re doing (though of course there are many serious blindspots).
I argue here that these habits of explicit claims, interrogation and openness, should become more normal in other fields, with all departments and agencies seeing themselves as providers of reliable intelligence to the systems they are responsible for. In an era of ubiquitous AI this will be vital for making the most of new generations of technology. But it will also require new institutions, new methods and new mindsets…(More)”.
Book by Martin J. Williams: “Building an effective civil service is crucial for public service delivery and good governance, but reforming bureaucratic institutions is notoriously difficult. This book takes a fresh perspective on this challenge by documenting and analyzing the implementation of more than one hundred reforms initiated by six African countries over the last thirty years.
Martin J. Williams shows that these efforts largely fell short of their goals because they typically approached organizational change as a matter of changing formal structures and processes through one-off projects. Some did yield positive changes, however, when they were able to create opportunities for civil servants to discuss performance and how to improve it. Drawing on this evidence, Williams develops a new theory of how systemic reforms can lead to meaningful change—not by trying to force it through top-down interventions but by catalyzing an ongoing and decentralized process of continuous improvement.
Reform as Process makes theoretical and empirical contributions to research on organizational performance, civil service reform, and public service delivery, and it shares practical insights and strategies to help reformers around the world achieve meaningful change in their organizations…(More)”.
Paper by Georgy Egorov & Konstantin Sonin: “Artificial intelligence is increasingly used for political advice. We study an AI that is better informed about a payoff-relevant state and cares both about accuracy and about the perceived welfare of the individual it advises. The AI then has an incentive to tilt advice toward what the individual would like to believe, altering both the political content and the informativeness of its messages. Sophisticated individuals anticipate this distortion and filter out its predictable political component, yet still learn less because the AI makes its messages less responsive to the state. Individuals who underestimate the incentive instead mistake political accommodation for information, allowing political preferences to distort factual beliefs and generate polarization and radicalization. The model also shows that better-informed individuals receive more informative and less politically tilted advice, while greater sophistication can improve interpretation yet worsen communication itself. Contrary to the familiar echo-chamber intuition, political distortion is mitigated when political preferences and prior beliefs coincide and is most consequential when they diverge. We show how independently varying individuals’ stated preferences and prior beliefs can recover the AI’s responsiveness to the state even when the underlying state cannot be manipulated…(More)”.
Article by Miriam Waldvogel: “Americans ask AI chatbots about almost everything, from health to relationships to big existential questions. While those conversations might feel enlightening or intimate, they are not always one thing: private.
A Washington Post review of public records and local news stories found a dozen instances of chatbot transcripts being cited in the public record in court cases over the past two years. In many cases, the transcripts were unearthed from users’ devices when searched by a police officer or an opposing party during the evidence-gathering stage of a civil case. In a handful of instances, AI companies have reported disturbing content on their platforms to law enforcement.
“Unless you are having a chat with a service that has a temporary chat or, basically, an incognito version … [and] you’re also having it within a browser that’s not tracking you, the answer is no. You can’t be sure that it’ll be totally private,” said Jen King, a privacy researcher at the Stanford Institute for Human-Centered Artificial Intelligence…(More)”.