Public Interest AI Projects Dataset
Authors: Theresa Züger, Hadi Asghari, Birte Lübbert
Date: 30.01.2025
CC-BY-4.0

Motivation

The dataset was created to collect information about projects in the field of public interest AI. There is an existing research gap, since focus in the discourse is mainly on for-profit companies that build AI-systems. Also, the discourse on AI for good often does not provide detailed information about the projects it refers to and their set up regarding funding or organizational connection. This data aims at contributing to fill this research gap. We were especially interested in what fields and towards which goals public interest oriented AI is being developed. Therefore, the data we collected is categorized with the SDGs. This does not mean that the public interest and the SDGs are the same, but we believe they can serve as categories to help understand the landscape of projects.

The named authors collected, validated and categorized this dataset. They were part of the Federal Ministry for Education and Research (BMBF) funded research project “Public Interest AI” which was a project between 2020-2024 at the “AI & Society Lab” at the Berlin based research institute Alexander von Humboldt Institute for Internet and Society.

Composition and Collection Process

The instances in this dataset represent a project that develops or uses AI in systems for the public interest. It does not contain information about individual people. For us this means that it serves “the long term survival and well-being of a collective, construed as a public”, which is a definition by Barry Bozeman (2007) and is not profit driven at its core. A more detailed description of the understanding about what is “public interest” you can find in this paper, that reflects on public interest theory in its relation to AI, or on the website of the project. The dataset is a not representative sample of projects. However, it is the biggest such collection of public interest AI projects that we know of, since it requires a lot of detailed research and work. The projects were collected in two manners. The first methodological approach was a network-based search method (called “Network Search” in the Dataset) to identify relevant projects. Leveraging the professional networks and research of the research group members, information about projects was collected through research, direct communication and referrals. Any project identified through these interactions was documented and vetted against the minimal definition in the dataset. The categorization was based on the online documentation of the project. This approach was considered to be the most effective and reliable means of capturing relevant information, given the lack of pre-existing datasets or comprehensive repositories specifically focused on public interest AI projects. The second approach employed a structured web-based search methodology to identify relevant projects (called “Google Search” in the Dataset). Initially, a predefined set of search terms was used to query an online search engine. The first 20 search results returned were systematically reviewed, and for each result, hyperlinks within the identified pages were followed up to a depth of two levels. Each page at this depth was examined for the presence of relevant projects, which, if found, were documented in this dataset. This approach ensured a consistent and replicable process for capturing a broad yet focused scope of information.

Each project consists of characteristics assigned by the authors. The only unprocessed data is the link to the project website. Partially there is information missing within some projects and categories, because this information was unavailable/ there was not enough information about to clearly say e.g. what kind of AI technology was being used. The dataset is mostly self-contained, with the exception of the link to the project website/website with more information on the project. Even though we can not guarantee that the access to this website will always be available, we assume that research with this dataset can also be done without a working link. Furthermore it could also be an interesting finding about lacking sustainability of projects if the links from 2024 do not exist in the future.

The second data set “Public Interest AI Projects - connected to Democracy” is a subset of the larger data set “Public Interest AI Projects” and was assembled by reducing the categories to the SDGs that we see in connection with democracy: Peace, Justice and Strong Institutions; Reduced Inequalities; Quality Education; Partnership for the goals and Gender Equality. We then manually went through the projects with a four eye principle to reaffirm that we see a connection of each individual project to democracy. However, this connection is our interpretation.

Column description

Name: The first column depicts the name of the project.

SDG category: The second and third column contain the assigned SDG category/categories. The sustainable development goals are 17 global objectives that were adapted by the United Nations (UN) in 2015 as a “[...] universal call to action to end poverty, protect the planet, and ensure that by 2030 all people enjoy peace and prosperity.” (UNDP, n.d.). Projects can fit more than one SDG category. In our dataset, we used the four-eyes principle to decide which one or two categories the project objective fell into. We had an 18th category here called “not SDG” for projects working in the public interest sector, but not matching any of the SDG’s. While we are aware that the SDG’s are not the same as the public interest it is an attempt to create an overview with known categories.

For or non-profit: The next column, ‘for or non profit’, was used to categorise whether the developed application is for or non-profit at core (to the best of our understanding form its documentation online) or non-profit or and whether the project is part of a research effort (which is very often the case in the projects we found).

Hashtags: The column ‘Hashtags’ is being used to describe the projects to help sort them better. The hashtags were given in an intuitive manner following the description of the projects and are supposed to give more fine grained categorization additional to the SDG category.

Index of Hashtags used: #AntiDiscrimination; #AccesstoKnowledge; #Accessibility; #AssuringProvision; #AggregatingKnowledge; #Biodiversity; #BuildingFootprints; #CircularEconomy; #CrisisResponse; #ClimateResearch; #CatastropheManagement; #ConnectingArt&Culture; #DataPrivacy; #Democracy; #DetectDisinformation; #DisplacementMonitoring; #DiscourseIntervention; #EnergyEfficiency; #Education; #EnsuringHarvest #EqualOpportunities; #EnvironmentalMonitoring; #GreenCityPlanning; #HumanRights; #HumanitarianResponse; #Inclusivity; #ImprovingTreatment; #Investigation; #Fairness; #KnowledgeTransfer; #KnowledgeInfastructure; #LegalSupport; #LaborMarketIntegration; #MentalSupport; #NatureConservation; #NaturePreservation; #OpenSourceInvestigation; #OceanCleaning; #OrganDonation; #PublicHealth; #PublicAdministration; #PublicDiscourse #PublicSafety; #ProtectingEcosystems; #Prevention; #ProtectionfromDomesticViolence; #ProtectCultualHeritage; #PolicyAdvice; #PreserveLanguage; #PreservingHistoricalMemories; #ReducingFoodwaste; #RememberingWitnesses; #SimpleLanguage; #SustainableMobility #SustainableIndustry; #sustainableAgriculture; #SustainableInfrastructure; #SustainableConsumption #SustainableEngergy; #SustainableHousing; #SocialJustice; #SupportingPhysicalWork; #Tourism #WaterQuality

AI/Technology: ‘AI/Technology’ is based on the description of the projects on their website, as far as we could tell. Therefore it’s our “best guess”, if you work in depth with that category, we recommend looking into it yourself deeper and maybe even contacting the projects individually.

Project Owner: ‘Project Owner’ displays the institution mainly responsible for the project, which we could find in the online documentation.

Activity Status: ‘Activity Status’ refers to the status of the project in 2024. Not in all cases a clear indication for the activity status could be found. The status might have changed. If you work in depth with this category we suggest confirming the status at the given time.

City, Country Legal: ‘City’ and ‘Country Legal’ are where the project is legally based.

Region Project: ‘Region Project’ refers to the scope the project has (e.g. if a tool only works in greek language, it has a scope of only Greek).

Funding: In ‘Funding’ we collected information that could be found on funding. Projects can have more than one funder, for such projects we noted the main funding source.

Link: In ‘Link’ the project website itself or a website with more information on the project are linked. Each link was accessed by us at the point of creation of the project in the dataset. However, it might not be active anymore, which could indicate that the project is not active, too.

Description: In ‘Description’ there is information on the aim/achievement of the project. Where possible, the self-description of the projects was used, if not we described it based on the information we had in our own words. If the description was in a language other than English, it has been automatically translated by Deepl.com.

In the spreadsheet for the Google Search, there is an additional column ‘Search Phrase’ in which you can see the term that was searched for, under what we found the project.

Uses
The dataset is used for the research of the Public Interest AI research group of the HIIG, and was partly presented in a poster presentation at Neurips 2024. We will continue our research and plan to publish results from an analysis of the dataset in 2025 in a conference paper. The data could be used by other researchers to better understand the landscape of public interest AI projects as well as NGOs active in this field.

Maintenance
The authors will maintain the dataset to the best of their ability. This might mean that updates are not regular and data might be outdated within the status quo of the dataset. Contact for questions: [email protected] If you want to contribute to this dataset by pointing to projects that are now yet included you can submit new projects to this Google Form and we will vet them and include them if they fit into our methodology.

Source as guideline for this datasheet: Gebru, T., Morgenstern, J., Vecchione, B., Wortman Vaughan, J., Wallach, H., Daumé III, H., & Crawford, K. (2021). Datasheets for datasets. arXiv. https://arxiv.org/abs/1803.09010