Charting the AI for Good Landscape – A New Look

Curated on Posted on by Stefaan Verhulst

Article by Perry Hewitt and Jake Porway: “More than 50% of nonprofits report that their organization uses generative AI in day-to-day operations. We’ve also seen an explosion of AI tools and investments. 10% of all the AI companies that exist in the US were founded in 2022, and that number has likely grown in subsequent years.  With investors funneling over $300B into AI and machine learning startups, it’s unlikely this trend will reverse any time soon.

Not surprisingly, the conversation about Artificial Intelligence (AI) is now everywhere, spanning from commercial uses such as virtual assistants and consumer AI to public goods, like AI-driven drug discovery and chatbots for education. The dizzying amount of new AI programs and initiatives – over 5000 new tools listed in 2023 on AI directories like TheresAnAI alone – can make the AI landscape challenging to navigate in general, much less for social impact. Luckily, four years ago, we surveyed the Data and AI for Good landscape and mapped out distinct families of initiatives based on their core goals. Today, we are revisiting that landscape to help folks get a handle on the AI for Good landscape today and to reflect on how the field has expanded, diversified, and matured…(More)”.

Smart Cities:Technologies and Policy Options to Enhance Services and Transparency

Curated on Posted on by Stefaan Verhulst

GAO Report: “Cities across the nation are using “smart city” technologies like traffic cameras and gunshot detectors to improve public services. In this technology assessment, we looked at their use in transportation and law enforcement.

Experts and city officials reported multiple benefits. For example, Houston uses cameras and Bluetooth sensors to measure traffic flow and adjust signal timing. Other cities use license plate readers to find stolen vehicles.

But the technologies can be costly and the benefits unclear. The data they collect may be sold, raising privacy and civil liberties concerns. We offer three policy options to address such challenges…(More)”.

Data Commons: The Missing Infrastructure for Public Interest Artificial Intelligence

Curated on Posted on by Stefaan Verhulst

Article by Stefaan Verhulst, Burton Davis and Andrew Schroeder: “Artificial intelligence is celebrated as the defining technology of our time. From ChatGPT to Copilot and beyond, generative AI systems are reshaping how we work, learn, and govern. But behind the headline-grabbing breakthroughs lies a fundamental problem: The data these systems depend on to produce useful results that serve the public interest is increasingly out of reach.

Without access to diverse, high-quality datasets, AI models risk reinforcing bias, deepening inequality, and returning less accurate, more imprecise results. Yet, access to data remains fragmented, siloed, and increasingly enclosed. What was once open—government records, scientific research, public media—is now locked away by proprietary terms, outdated policies, or simple neglect. We are entering a data winter just as AI’s influence over public life is heating up.

This isn’t just a technical glitch. It’s a structural failure. What we urgently need is new infrastructure: data commons.

A data commons is a shared pool of data resources—responsibly governed, managed using participatory approaches, and made available for reuse in the public interest. Done correctly, commons can ensure that communities and other networks have a say in how their data is used, that public interest organizations can access the data they need, and that the benefits of AI can be applied to meet societal challenges.

Commons offer a practical response to the paradox of data scarcity amid abundance. By pooling datasets across organizations—governments, universities, libraries, and more—they match data supply with real-world demand, making it easier to build AI that responds to public needs.

We’re already seeing early signs of what this future might look like. Projects like Common Corpus, MLCommons, and Harvard’s Institutional Data Initiative show how diverse institutions can collaborate to make data both accessible and accountable. These initiatives emphasize open standards, participatory governance, and responsible reuse. They challenge the idea that data must be either locked up or left unprotected, offering a third way rooted in shared value and public purpose.

But the pace of progress isn’t matching the urgency of the moment. While policymakers debate AI regulation, they often ignore the infrastructure that makes public interest applications possible in the first place. Without better access to high-quality, responsibly governed data, AI for the common good will remain more aspiration than reality.

That’s why we’re launching The New Commons Challenge—a call to action for universities, libraries, civil society, and technologists to build data ecosystems that fuel public-interest AI…(More)”.

These Startups Are Building Advanced AI Models Without Data Centers

Curated on Posted on by Stefaan Verhulst

Article by Will Knight: “Researchers have trained a new kind of large language model (LLM) using GPUs dotted across the world and fed private as well as public data—a move that suggests that the dominant way of building artificial intelligence could be disrupted.

Article by Will Knight: “Flower AI and Vana, two startups pursuing unconventional approaches to building AI, worked together to create the new model, called Collective-1.

Flower created techniques that allow training to be spread across hundreds of computers connected over the internet. The company’s technology is already used by some firms to train AI models without needing to pool compute resources or data. Vana provided sources of data including private messages from X, Reddit, and Telegram.

Collective-1 is small by modern standards, with 7 billion parameters—values that combine to give the model its abilities—compared to hundreds of billions for today’s most advanced models, such as those that power programs like ChatGPT, Claude, and Gemini.

Nic Lane, a computer scientist at the University of Cambridge and cofounder of Flower AI, says that the distributed approach promises to scale far beyond the size of Collective-1. Lane adds that Flower AI is partway through training a model with 30 billion parameters using conventional data, and plans to train another model with 100 billion parameters—close to the size offered by industry leaders—later this year. “It could really change the way everyone thinks about AI, so we’re chasing this pretty hard,” Lane says. He says the startup is also incorporating images and audio into training to create multimodal models.

Distributed model-building could also unsettle the power dynamics that have shaped the AI industry…(More)”

AI action plan database

Curated on Posted on by Stefaan Verhulst

A project by the Institute for Progress: “In January 2025, President Trump tasked the Office of Science and Technology Policy with creating an AI Action Plan to promote American AI Leadership. The government requested input from the public, and received 10,068 submissions. The database below summarizes specific recommendations from these submissions. … We used AI to extract recommendations from each submission, and to tag them with relevant information. Click on a recommendation to learn more about it. See our analysis of common themes and ideas across these recommendations…(More)”.

Updating purpose limitation for AI: a normative approach from law and philosophy 

Curated on Posted on by Stefaan Verhulst

Paper by Rainer Mühlhoff and Hannah Ruschemeier: “The purpose limitation principle goes beyond the protection of the individual data subjects: it aims to ensure transparency, fairness and its exception for privileged purposes. However, in the current reality of powerful AI models, purpose limitation is often impossible to enforce and is thus structurally undermined. This paper addresses a critical regulatory gap in EU digital legislation: the risk of secondary use of trained models and anonymised training datasets. Anonymised training data, as well as AI models trained from this data, pose the threat of being freely reused in potentially harmful contexts such as insurance risk scoring and automated job applicant screening. We propose shifting the focus of purpose limitation from data processing to AI model regulation. This approach mandates that those training AI models define the intended purpose and restrict the use of the model solely to this stated purpose…(More)”.

Technical Tiers: A New Classification Framework for Global AI Workforce Analysis

Curated on Posted on by Stefaan Verhulst

Report by Siddhi Pal, Catherine Schneider and Ruggero Marino Lazzaroni: “… introduces a novel three-tiered classification system for global AI talent that addresses significant methodological limitations in existing workforce analyses, by distinguishing between different skill categories within the existing AI talent pool. By distinguishing between non-technical roles (Category 0), technical software development (Category 1), and advanced deep learning specialization (Category 2), our framework enables precise examination of AI workforce dynamics at a pivotal moment in global AI policy.

Through our analysis of a sample of 1.6 million individuals in the AI talent pool across 31 countries, we’ve uncovered clear patterns in technical talent distribution that significantly impact Europe’s AI ambitions. Asian nations hold an advantage in specialized AI expertise, with South Korea (27%), Israel (23%), and Japan (20%) maintaining the highest proportions of Category 2 talent. Within Europe, Poland and Germany stand out as leaders in specialized AI talent. This may be connected to their initiatives to attract tech companies and investments in elite research institutions, though further research is needed to confirm these relationships.

Our data also reveals a shifting landscape of global talent flows. Research shows that countries employing points-based immigration systems attract 1.5 times more high-skilled migrants than those using demand-led approaches. This finding takes on new significance in light of recent geopolitical developments affecting scientific research globally. As restrictive policies and funding cuts create uncertainty for researchers in the United States, one of the big destinations for European AI talent, the way nations position their regulatory environments, scientific freedoms, and research infrastructure will increasingly determine their ability to attract and retain specialized AI talent.

The gender analysis in our study illuminates another dimension of competitive advantage. Contrary to the overall AI talent pool, EU countries lead in female representation in highly technical roles (Category 2), occupying seven of the top ten global rankings. Finland, Czechia, and Italy have the highest proportion of female representation in Category 2 roles globally (39%, 31%, and 28%, respectively). This gender diversity represents not merely a social achievement but a potential strategic asset in AI innovation, particularly as global coalitions increasingly emphasize the importance of diverse perspectives in AI development…(More)”

Make privacy policies longer and appoint LLM readers

Curated on Posted on by Stefaan Verhulst

Paper by Przemysław Pałka et al: “In a world of human-only readers, a trade-off persists between comprehensiveness and comprehensibility: only privacy policies too long to be humanly readable can precisely describe the intended data processing. We argue that this trade-off no longer exists where LLMs are able to extract tailored information from clearly-drafted fully-comprehensive privacy policies. To substantiate this claim, we provide a methodology for drafting comprehensive non-ambiguous privacy policies and for querying them using LLMs prompts. Our methodology is tested with an experiment aimed at determining to what extent GPT-4 and Llama2 are able to answer questions regarding the content of privacy policies designed in the format we propose. We further support this claim by analyzing real privacy policies in the chosen market sectors through two experiments (one with legal experts, and another by using LLMs). Based on the success of our experiments, we submit that data protection law should change: it must require controllers to provide clearly drafted, fully comprehensive privacy policies from which data subjects and other actors can extract the needed information, with the help of LLMs…(More)”.

Artificial Intelligence: Generative AI’s Environmental and Human Effects

Curated on Posted on by Stefaan Verhulst

GAO Report: “Generative artificial intelligence (AI) could revolutionize entire industries. In the nearer term, it may dramatically increase productivity and transform daily tasks in many sectors. However, both its benefits and risks, including its environmental and human effects, are unknown or unclear.

Generative AI uses significant energy and water resources, but companies are generally not reporting details of these uses. Most estimates of environmental effects of generative AI technologies have focused on quantifying the energy consumed, and carbon emissions associated with generating that energy, required to train the generative AI model. Estimates of water consumption by generative AI are limited. Generative AI is expected to be a driving force for data center demand, but what portion of data center electricity consumption is related to generative AI is unclear. According to the International Energy Agency, U.S. data center electricity consumption was approximately 4 percent of U.S. electricity demand in 2022 and could be 6 percent of demand in 2026.

While generative AI may bring beneficial effects for people, GAO highlights five risks and challenges that could result in negative human effects on society, culture, and people from generative AI (see figure). For example, unsafe systems may produce outputs that compromise safety, such as inaccurate information, undesirable content, or the enabling of malicious behavior. However, definitive statements about these risks and challenges are difficult to make because generative AI is rapidly evolving, and private developers do not disclose some key technical information.

Selected generative artificial antelligence risks and challenges that could result in human effects

GAO identified policy options to consider that could enhance the benefits or address the challenges of environmental and human effects of generative AI. These policy options identify possible actions by policymakers, which include Congress, federal agencies, state and local governments, academic and research institutions, and industry. In addition, policymakers could choose to maintain the status quo, whereby they would not take additional action beyond current efforts. See below for details on the policy options…(More)”.

Brazil’s AI-powered social security app is wrongly rejecting claims

Curated on Posted on by Stefaan Verhulst

Article by Gabriel Daros: “Brazil’s social security institute, known as INSS, added AI to its app in 2018 in an effort to cut red tape and speed up claims. The office, known for its long lines and wait times, had around 2 million pending requests for everything from doctor’s appointments to sick pay to pensions to retirement benefits at the time. While the AI-powered tool has since helped process thousands of basic claims, it has also rejected requests from hundreds of people like de Brito — who live in remote areas and have little digital literacy — for minor errors.

The government is right to digitize its systems to improve efficiency, but that has come at a cost, Edjane Rodrigues, secretary for social policies at the National Confederation of Workers in Agriculture, told Rest of World.

“If the government adopts this kind of service to speed up benefits for the people, this is good. We are not against it,” she said. But, particularly among farm workers, claims can be complex because of the nature of their work, she said, referring to cases that require additional paperwork, such as when a piece of land is owned by one individual but worked by a group of families. “There are many peculiarities in agriculture, and rural workers are being especially harmed” by the app, according to Rodrigues.

“Each automated decision is based on specified legal criteria, ensuring that the standards set by the social security legislation are respected,” a spokesperson for INSS told Rest of World. “Automation does not work in an arbitrary manner. Instead, it follows clear rules and regulations, mirroring the expected standards applied in conventional analysis.”

Governments across Latin America have been introducing AI to improve their processes. Last year, Argentina began using ChatGPT to draft court rulings, a move that officials said helped cut legal costs and reduce processing times. Costa Rica has partnered with Microsoft to launch an AI tool to optimize tax data collection and check for fraud in digital tax receipts. El Salvador recently set up an AI lab to develop tools for government services.

But while some of these efforts have delivered promising results, experts have raised concerns about the risk of officials with little tech know-how applying these tools with no transparency or workarounds…(More)”.