A Collaboration by Sandbox Labs, UPF-Barcelona School of Management, and Norrsken House Barcelona

Artificial intelligence is rapidly reshaping economies, public services, and everyday life, generating both societal benefits and risks. 

Yet control over how these systems are developed and deployed remains concentrated in a handful of private companies, creating a dependency on systems that are opaque, unequally accessible, and not designed to prioritize individual rights and societal well-being for governments and citizens. 

To counter this imbalance, we urgently need an alternative vision: one in which AI systems are controlled by citizens through public ownership, funding, operation, and regulation to serve the public interest. And where governments incorporate AI technologies into public administration, doing so in ways that ensure they also serve the public interest and protect people's rights.

However, this alternative vision of building AI in the public interest remains relatively unknown, and even fewer people possess the practical know-how to implement it.

Turning this vision into reality requires a new generation of leaders who are equipped to navigate the fast-moving field of AI. One where technology collides with politics, economics, law, and ethics all at once. 

A collaborative partnership between higher education and the private sector

While public AI and public-interest AI generate enthusiasm and support within specific communities, they remain largely unknown to the broader public. 

To help bridge this gap, a collaboration between Sandbox Labs, the UPF-Barcelona School of Management, and Norrsken House Barcelona launched the first edition of the Public-Interest AI Accelerator. 

Unlike traditional accelerators focused on scaling tech startup growth, this nine-week, in-person pilot program was designed to accelerate the knowledge, awareness, and confidence of current and future public leaders to build a world where AI is developed to serve the public interest. 

The program brought together a small cohort of diverse students from non-technical fields, including policy, law, education, and marketing, to engage directly with an international panel of technology thinkers, researchers, and practitioners spanning computer science, law and digital rights, health, policy, and neuroscience. 

Designed to provide students with technical, practical, and critical perspectives on AI, the accelerator program served as a vital testing ground for a pedagogical design that challenges learners to move beyond theoretical concepts into real-world, localized decision-making, learning from inspiring work around the world.

Throughout the program, three key insights emerged on how to move past abstract principles and toward concrete execution.

1. Visibility, Scale, and the Public AI Stack

Building an alternative to commercial AI is an uphill battle in a market overwhelmingly dominated by concentrated private capital, a reality highlighted by Joshua Tan, interim CTO of Current AI. 

To disrupt this momentum, we must first map what public AI infrastructure actually looks like.

Alek Tarkowski, Co-founder of the Open Future Foundation and Felix Sieker, Project Manager of reframe [Tech] – Algorithm for the Common Good at Bertelsmann Stiftung, introduced the public AI framework, which envisions AI built on transparent governance, democratic accountability, and open-access to key components of the AI stack. 

They also establish the idea of a “gradient of publicness”, which describes different components of AI systems along a continuum from fully private to fully public ownership, allowing policymakers to diagnose where current systems stand and where public control might be increased and public interest emphasized. 

Rather than remaining purely theoretical, this vision is already being localized. Albert Caňigueral, manager of the Barcelona Supercomputing Centre (BSC)’s AI factory, demonstrated how the BSC acts as an essential public infrastructure, connecting researchers, public sector actors and small and medium enterprises with critical compute power and technologies they would not otherwise be able to access. 

Public AI infrastructure serves as a channel for actors seeking to conceptualize, build, and deploy public-interest AI. At the same time, pursuing AI for public interest can also be done through leveraging private infrastructure, provided the right ideas, skills, and systems are assembled.

For example, Ana Freire, Vice-dean for Social Impact and Academic Innovation at UPF-BSM, showcased the STOP project, a public-interest AI initiative that leverages machine learning to detect early signs of mental distress on social media platforms and to connect vulnerable youth with life-saving services.

2. Shifting the Levers of Power: Democratic Oversight and Public Procurement

The direction of innovation isn’t inherently written into code; it is shaped by power. If technological progress is to serve the collective, inclusive institutions must intentionally distribute that power.

To build an equitable and desirable AI future, public leaders must break out of the reductive binary of being simply "pro-AI" or "anti-AI". As Javi Creus, Founder of Ideas for Change, noted, the direction of innovation isn’t inherently written into code; it is shaped by power. If technological progress is to serve the collective, inclusive institutions must intentionally distribute that power.

This institutional responsibility is historically urgent and a challenge Europe is uniquely positioned to address. Europe possesses deep experience in directing technological development and deployment to safeguard public interests, a legacy it can now apply to the current AI governance crisis. 

Simona Levi (Founder of Xnet: Institute for democratic digitalization) reminded the cohort that major breakthroughs in information technology historically trigger a push-and-pull dynamic between democratic openings and authoritarian control. 

The internet, for instance, initially democratized access to information and brought about new modes of social and political engagement, only to later experience recentralized control and increased censorship. 

Public AI should not be judged merely by its ability to limit AI’s risks through regulation, but by its capacity to help societies learn, coordinate collective action, and improve systemic economic and social outcomes. 

To ensure AI’s benefits are broadly shared, systems must be governed within democratic digital frameworks that safeguard human rights by design. Echoing this, co-founder of SharedAgendas, Tatiana Fernandez, emphasized that public AI should not be judged merely by its ability to limit AI’s risks through regulation, but by its capacity to help societies learn, coordinate collective action, and improve systemic economic and social outcomes. 

Advancing this alternative AI framework requires navigating the tension between the enormous demand for training data, which lies at the core of AI technology, and individual rights protections. 

Legal experts Migle Laukyte (Associate Professor at UFP) and Maja Bogataj Jančič (Founder of the Open Data and Intellectual Property Institute) demonstrated that, without safeguards, data collection and use underpinning AI systems can override fundamental rights like privacy, or interfere with intellectual property, and introduce bias into information and decision-making systems. 

Severe risks ensue, ranging from corporate manipulation and state surveillance to automated discrimination that denies marginalized communities essential opportunities.

Governments are not passive consumers, but rather powerful architects of the AI market. 

Crucially, governments already possess market levers to combat these threats, though they are underutilized. The Associate Dean of Education and Academic Affairs and Professor at UPF Barcelona School of Management Rodrigo Cetina-Presuel, highlighted that governments are not passive consumers, but rather powerful architects of the AI market. 

Because vendors will ultimately build what governments demand at scale, the state can effectively enforce the ethical rules of the game for private providers. By establishing strict public-interest criteria for purchasing tech, public entities can leverage massive public spending to directly shape market standards.

3. Anchoring AI in Collective Intelligence and Human Limits

Ultimately, public-interest AI must be fostered at all levels of decision-making, including the community and individual levels. On a community scale, evaluating an AI system's efficacy cannot rely solely on the technical performance metrics favored by Silicon Valley. 

Instead, B Cavello, Director of Emerging Technologies at Aspen Institute, introduced the concept of community-aligned benchmarks, which evaluate technology against the specific values, priorities, and self-defined needs of the communities for which they are designed. 

But scaling AI-powered collective intelligence requires a clear-eyed understanding of human vulnerability. Katherine d’Amico, Professor of the Practice in Applied Neuroscience at UPF-BSM, cautioned the cohort about the mismatch between the high-speed nature of AI environments and the cognitive constraints of human decision-making, which is bounded by emotion, resistance to change, and processing capacity. 

In high-velocity, data-driven landscapes, leaders must be careful not to default to intuitive and path-dependent “fast thinking". 

To protect the integrity of public-interest decision-making, institutional processes must intentionally build in friction, leaving dedicated space for the slow, deliberate, ethical reasoning required to solve complex problems.

Emerging Ideas: Translating Principles into Practice

The value of the accelerator’s pedagogical design was demonstrated in the final assignment, in which students moved beyond theory to propose concrete, real-world public-interest AI applications. 

The cohort successfully identified structural blind spots and socially relevant opportunities that tech-first learning approaches may dismiss. Students’ proposals tackled urgent and broadly shared societal challenges, including leveraging AI to improve public administration effectiveness, enhancing the relevance of university education and efficient labor market access, and scaling mental health services for youth and workers. 

Others framed their solutions in highly specific policy contexts, from addressing gaps in childcare access and quality in Mexico or the risks of AI misuse in Ecuador’s justice system, where mainstream AI market tools fail to comprehend Indigenous languages and cultural contexts, posing significant risks of bias and discrimination.

Where Next? 

Insights from global experts and our first student cohort tell us that robust policies are necessary to advance effective AI regulation and accelerate the design and implementation of public-interest AI use cases. 

Governments must pursue smart regulations that shape private AI, focusing on clear rules of the game to foster healthy competition without killing innovation. Policy frameworks must support the pooling of public and private investment and facilitate the testing and adoption of public-interest AI use cases, with a particular role for middle-power countries in Europe and beyond that share a vision for public-interest AI.

But none of these goals are achievable without ensuring that current and future public sector leaders have the right knowledge, skills, and confidence to engage with AI systems. Leaders need to understand the foundations of AI technologies, the structural realities of AI markets, where public AI and public-interest AI might fit, and the tools and competencies to foster the development, deployment, and adoption of AI technology in the public interest.

Reflecting on lessons from our first Public-Interest AI Accelerator, we believe effective AI education must pursue two directions. 

The hard work of translating principles into reality through best-practice analysis and hands-on experimentation is critically needed and should be baked into AI educational programs.

First, we need to push learners to think critically about the ethical use of AI and its impacts on communities and individuals, helping them identify both areas where regulation is needed and areas where new ideas and practices must be cultivated to promote and implement socially valuable uses of AI. 

Second, learners must understand the blockers and enablers that shape the concrete implementation of public-interest AI, whether technical, economic, political, organizational, or psychological. The hard work of translating principles into reality through best-practice analysis and hands-on experimentation is critically needed and should be baked into AI educational programs.

Those are ideas guiding our next steps, building on the inaugural cohort of our Public-Interest AI Accelerator and looking ahead to the program's next iteration. 

We look forward to feedback and collaboration across the public and private sectors to help create an alternative model of AI development that limits technological harm, fosters universal access to human-focused technological progress, and fairly distributes the benefits of AI-driven transformations.