Paper by Vincent Conitzer, et al: “Foundation models such as GPT-4 are fine-tuned to avoid unsafe or otherwise problematic behavior, so that, for example, they refuse to comply with requests for help with committing crimes or with producing racist text. One approach to fine-tuning, called reinforcement learning from human feedback, learns from humans’ expressed preferences over multiple outputs. Another approach is constitutional AI, in which the input from humans is a list of high-level principles. But how do we deal with potentially diverging input from humans? How can we aggregate the input into consistent data about ”collective” preferences or otherwise use it to make collective choices about model behavior? In this paper, we argue that the field of social choice is well positioned to address these questions…(More)”.
We Need To Rewild The Internet
Article by Maria Farrell and Robin Berjon: “In the late 18th century, officials in Prussia and Saxony began to rearrange their complex, diverse forests into straight rows of single-species trees. Forests had been sources of food, grazing, shelter, medicine, bedding and more for the people who lived in and around them, but to the early modern state, they were simply a source of timber.
So-called “scientific forestry” was that century’s growth hacking. It made timber yields easier to count, predict and harvest, and meant owners no longer relied on skilled local foresters to manage forests. They were replaced with lower-skilled laborers following basic algorithmic instructions to keep the monocrop tidy, the understory bare.
Information and decision-making power now flowed straight to the top. Decades later when the first crop was felled, vast fortunes were made, tree by standardized tree. The clear-felled forests were replanted, with hopes of extending the boom. Readers of the American political anthropologist of anarchy and order, James C. Scott, know what happened next.
It was a disaster so bad that a new word, Waldsterben, or “forest death,” was minted to describe the result. All the same species and age, the trees were flattened in storms, ravaged by insects and disease — even the survivors were spindly and weak. Forests were now so tidy and bare, they were all but dead. The first magnificent bounty had not been the beginning of endless riches, but a one-off harvesting of millennia of soil wealth built up by biodiversity and symbiosis. Complexity was the goose that laid golden eggs, and she had been slaughtered…(More)”.
On the Manipulation of Information by Governments
Paper by Ariel Karlinsky and Moses Shayo: “Governmental information manipulation has been hard to measure and study systematically. We hand-collect data from official and unofficial sources in 134 countries to estimate misreporting of Covid mortality during 2020-21. We find that between 45%–55% of governments misreported the number of deaths. The lion’s share of misreporting cannot be attributed to a country’s capacity to accurately diagnose and report deaths. Contrary to some theoretical expectations, there is little evidence of governments exaggerating the severity of the pandemic. Misreporting is higher where governments face few social and institutional constraints, in countries holding elections, and in countries with a communist legacy…(More)”
Democracy and Artificial Intelligence: old problems, new solutions?
Discussion between Nardine Alnemr and Rob Weymouth: “…I see three big perspectives relevant to AI and democracy. You have the most conservative, mirroring the 80s and the 90s, still talking about the digital public sphere as if it’s distant from our lives. As if it’s something novel and inaccessible, which is not quite accurate anymore.
Then there’s the more optimistic and cautionary side of the spectrum. People who are excited about the technologies, but they’re not quite sure. They’re intrigued to see the potential and I think they’re optimistic because they overlook how these technologies connect to a broader context. How a lot of these technologies are driven by surveying and surveillance of the data and the communication that we produce. Exploitation of workers who do the filtering and cleaning work. The companies that profit out of this and make engineered election campaigns. So they’re cautious because of that, but still optimistic, because at the same time, they try to isolate it from that bigger context.
And finally, the most radical is something like Cesar Hidalgo’s proposal of augmented democracy…(More)”.
The Formalization of Social Precarities
Anthology edited by Murali Shanmugavelan and Aiha Nguyen: “…explores platformization from the point of view of precarious gig workers in the Majority World. In countries like Bangladesh, Brazil, and India — which reinforce social hierarchies via gender, race, and caste — precarious workers are often the most marginalized members of society. Labor platforms made familiar promises to workers in these countries: work would be democratized, and people would have the opportunity to be their own boss. Yet even as platforms have upended the legal relationship between worker and employer, they have leaned into social structures to keep workers precarious — and in fact formalized those social precarities through surveillance and data collection…(More)”.
Global Contract-level Public Procurement Dataset
Paper by Mihály Fazekas et al: “One-third of total government spending across the globe goes to public procurement, amounting to about 10 trillion dollars a year. Despite its vast size and crucial importance for economic and political developments, there is a lack of globally comparable data on contract awards and tenders run. To fill this gap, this article introduces the Global Public Procurement Dataset (GPPD). Using web scraping methods, we collected official public procurement data on over 72 million contracts from 42 countries between 2006 and 2021 (time period covered varies by country due to data availability constraints). To overcome the inconsistency of data publishing formats in each country, we standardized the published information to fit a common data standard. For each country, key information is collected on the buyer(s) and supplier(s), geolocation information, product classification, price information, and details of the contracting process such as contract award date or the procedure type followed. GPPD is a contract-level dataset where specific filters are calculated allowing to reduce the dataset to the successfully awarded contracts if needed. We also add several corruption risk indicators and a composite corruption risk index for each contract which allows for an objective assessment of risks and comparison across time, organizations, or countries. The data can be reused to answer research questions dealing with public procurement spending efficiency among others. Using unique organizational identification numbers or organization names allows connecting the data to company registries to study broader topics such as ownership networks…(More)”.
The Ethics of Advanced AI Assistants
Paper by Iason Gabriel et al: “This paper focuses on the opportunities and the ethical and societal risks posed by advanced AI assistants. We define advanced AI assistants as artificial agents with natural language interfaces, whose function is to plan and execute sequences of actions on behalf of a user – across one or more domains – in line with the user’s expectations. The paper starts by considering the technology itself, providing an overview of AI assistants, their technical foundations and potential range of applications. It then explores questions around AI value alignment, well-being, safety and malicious uses. Extending the circle of inquiry further, we next consider the relationship between advanced AI assistants and individual users in more detail, exploring topics such as manipulation and persuasion, anthropomorphism, appropriate relationships, trust and privacy. With this analysis in place, we consider the deployment of advanced assistants at a societal scale, focusing on cooperation, equity and access, misinformation, economic impact, the environment and how best to evaluate advanced AI assistants. Finally, we conclude by providing a range of recommendations for researchers, developers, policymakers and public stakeholders…(More)”.
The End of the Policy Analyst? Testing the Capability of Artificial Intelligence to Generate Plausible, Persuasive, and Useful Policy Analysis
Article by Mehrdad Safaei and Justin Longo: “Policy advising in government centers on the analysis of public problems and the developing of recommendations for dealing with them. In carrying out this work, policy analysts consult a variety of sources and work to synthesize that body of evidence into useful decision support documents commonly called briefing notes. Advances in natural language processing (NLP) have led to the continuing development of tools that can undertake a similar task. Given a brief prompt, a large language model (LLM) can synthesize information in content databases. This article documents the findings from an experiment that tested whether contemporary NLP technology is capable of producing public policy relevant briefing notes that expert evaluators judge to be useful. The research involved two stages. First, briefing notes were created using three models: NLP generated; human generated; and NLP generated/human edited. Next, two panels of retired senior public servants (with only one panel informed of the use of NLP in the experiment) were asked to judge the briefing notes using a heuristic evaluation rubric. The findings indicate that contemporary NLP tools were not able to, on their own, generate useful policy briefings. However, the feedback from the expert evaluators indicates that automatically generated briefing notes might serve as a useful supplement to the work of human policy analysts. And the speed with which the capabilities of NLP tools are developing, supplemented with access to a larger corpus of previously prepared policy briefings and other policy-relevant material, suggests that the quality of automatically generated briefings may improve significantly in the coming years. The article concludes with reflections on what such improvements might mean for the future practice of policy analysis…(More)”.
Unleashing collective intelligence for public decision-making: the Data for Policy community
Paper by Zeynep Engin, Emily Gardner, Andrew Hyde, Stefaan Verhulst and Jon Crowcroft: “Since its establishment in 2014, Data for Policy (https://dataforpolicy.org) has emerged as a prominent global community promoting interdisciplinary research and cross-sector collaborations in the realm of data-driven innovation for governance and policymaking. This report presents an overview of the community’s evolution from 2014 to 2023 and introduces its six-area framework, which provides a comprehensive mapping of the data for policy research landscape. The framework is based on extensive consultations with key stakeholders involved in the international committees of the annual Data for Policy conference series and the open-access journal Data & Policy published by Cambridge University Press. By presenting this inclusive framework, along with the guiding principles and future outlook for the community, this report serves as a vital foundation for continued research and innovation in the field of data for policy...(More)”.
The AI That Could Heal a Divided Internet
Article by Billy Perrigo: “In the 1990s and early 2000s, technologists made the world a grand promise: new communications technologies would strengthen democracy, undermine authoritarianism, and lead to a new era of human flourishing. But today, few people would agree that the internet has lived up to that lofty goal.
Today, on social media platforms, content tends to be ranked by how much engagement it receives. Over the last two decades politics, the media, and culture have all been reshaped to meet a single, overriding incentive: posts that provoke an emotional response often rise to the top.
Efforts to improve the health of online spaces have long focused on content moderation, the practice of detecting and removing bad content. Tech companies hired workers and built AI to identify hate speech, incitement to violence, and harassment. That worked imperfectly, but it stopped the worst toxicity from flooding our feeds.
There was one problem: while these AIs helped remove the bad, they didn’t elevate the good. “Do you see an internet that is working, where we are having conversations that are healthy or productive?” asks Yasmin Green, the CEO of Google’s Jigsaw unit, which was founded in 2010 with a remit to address threats to open societies. “No. You see an internet that is driving us further and further apart.”
What if there were another way?
Jigsaw believes it has found one. On Monday, the Google subsidiary revealed a new set of AI tools, or classifiers, that can score posts based on the likelihood that they contain good content: Is a post nuanced? Does it contain evidence-based reasoning? Does it share a personal story, or foster human compassion? By returning a numerical score (from 0 to 1) representing the likelihood of a post containing each of those virtues and others, these new AI tools could allow the designers of online spaces to rank posts in a new way. Instead of posts that receive the most likes or comments rising to the top, platforms could—in an effort to foster a better community—choose to put the most nuanced comments, or the most compassionate ones, first…(More)”.