Towards a public infrastructure for Artificial Intelligence

Before discussing models, algorithms, data centres, cloud or computing capacity, we must address the deepest level of Artificial Intelligence: the construction of meaning. An AI system does not interpret reality directly, but the representations built of it. These representations consist of data, attributes, categories, relationships, events, classifications and interpretative rules. When these elements are not defined according to shared criteria, the machine can process enormous quantities of information without truly understanding the meaning they take on in the administrative, healthcare, industrial, economic or legal context in which they are used.

Semantics is the level that gives meaning to data; ontology, by contrast, is the formal structure that organises that meaning in a computable way. In scientific terms, an ontology describes a knowledge domain by identifying its fundamental concepts, the relationships that connect them, the attributes that characterise them, the rules through which new information can be derived and the logical constraints that prevent inconsistent or incompatible interpretations. An ontology is therefore not simply an advanced vocabulary, but a true architecture of knowledge that allows people, organisations and intelligent systems to share not only the same language, but also the same interpretation of reality.

The quality of Artificial Intelligence depends directly on the quality of this representation. Two administrations may use the same term with different meanings; two hospitals may classify the same condition according to different criteria; a Region, a Ministry and a Municipality may describe the same administrative procedure with incompatible information models. In all these cases the problem is not the quality of the data, but the quality of the meaning associated with the data. A database can be complete, up to date and formally correct, yet remain semantically fragile. The algorithm will then produce apparently consistent results, built on unstable conceptual categories. A significant part of algorithmic risk arises precisely from this incomplete or ambiguous representation of the knowledge domain, and precedes the functioning of the model itself.

Building a public infrastructure of meaning requires a semantic architecture structured on five progressive levels. The first is the definition of a controlled vocabulary, through which the fundamental terms of each domain are standardised. The second concerns the construction of taxonomies that organise concepts into coherent hierarchies. The third level is the ontology proper, in which relationships, properties and dependencies between concepts are formalised. The fourth involves transforming the ontology into a knowledge graph, making these relationships queryable and usable by computer systems. The fifth introduces the inferential level, enabling AI systems to detect anomalies, check consistency and produce new knowledge through automated inference. This path marks the transition from simple document digitisation to the construction of a true knowledge infrastructure. It is the precondition for Public Administration to evolve towards a model of computable institutional intelligence.

Building public ontologies inevitably requires multidisciplinary governance. Defining meaning cannot be left to computer scientists alone, as it involves organisational, legal, economic, social and linguistic expertise. The Regions can play a territorial coordination role by identifying priority domains; Public Administrations contribute their knowledge of administrative procedures and data; universities and research centres bring expertise in knowledge engineering, artificial intelligence, data science, law, economics and computational linguistics; technology companies develop semantic representation tools, while businesses in the various economic sectors transfer the operational language of their domains. Professional orders help formalise specialist practices, and the Third Sector, together with the Community Juries for AI, checks that the categories adopted correctly represent the needs of people and local areas.

In this context, Fondazione ENIA could act as a methodological link between all the actors involved, supporting the construction, validation and evolution of the public semantics of Artificial Intelligence. The Congress therefore proposes launching a Public Semantic Atlas for Artificial Intelligence, conceived as a permanent programme of research and collaboration. The project could be developed by mapping priority domains, collecting existing vocabularies from administrations, businesses and universities, creating pilot ontologies in the main areas of public interest, turning them into interoperable knowledge graphs and then testing them with predictive and generative AI systems to measure their effects in terms of interoperability, accuracy, traceability and reliability. The ultimate goal is to build a public infrastructure of meaning, a shared asset that enables institutions, businesses and citizens to take part in building a trustworthy, interoperable and transparent Artificial Intelligence ecosystem. Because, in the end, algorithms can process data; only shared semantics enables them to process knowledge.

If ontologies are the language through which Artificial Intelligence interprets the world, Edge Computing is the physical infrastructure that allows it to operate in a distributed, resilient way, close to where data is produced. The underlying idea is simple: instead of continually transferring huge amounts of information to large centralised cloud infrastructures, a significant share of processing is carried out directly at the data source. Hospitals, schools, factories, energy networks, airports, ports, transport systems, town halls, environmental sensors and critical infrastructure can thus become intelligent nodes of a distributed computing network. In this model, data is not simply collected and transmitted, but can be validated, filtered, anonymised, semantically enriched and used to produce local inferences before being shared, if necessary, with other infrastructures. This approach reduces latency, limits network traffic, improves operational continuity and strengthens control over strategic data.

For this reason, Edge Computing is a fundamental component of infrastructure sovereignty. The issue goes well beyond technology. Whoever controls processing capacity also controls response times, security levels, service continuity, technological dependencies and the resilience of the whole system. Infrastructure sovereignty does not mean isolation from the major global operators, but the ability to choose deliberately which processes to keep under public or industrial control, which data to process locally and which services to entrust to external infrastructures. For a region with such intense productive activity as Lombardy, this choice directly concerns healthcare, advanced manufacturing, mobility, energy, finance, research and public administration. A health system that uses algorithms for diagnostics, territorial monitoring or triage cannot rely solely on remote infrastructure. Similarly, a smart factory must be able to process data from robots, machines and sensors in real time, while a civil protection network must keep working even with degraded connectivity or events that compromise central infrastructure.

Technically, a modern Edge architecture develops as an integrated chain. At its base is the sensor layer, made up of IoT devices, industrial systems, medical equipment, cameras and monitoring networks. Next comes a pre-processing layer, in which data is normalised, checked, anonymised and prepared for use. Local inference then allows Artificial Intelligence models to operate directly on the ground, producing classifications, forecasts, alerts and decision support without necessarily relying on central infrastructure. Edge-cloud orchestration then coordinates the distribution of data and models, determining which information stays local and which must be shared. Alongside these are layers dedicated to security, governance and traceability of operations, up to the semantic layer, which links this pillar directly to the previous one. Indeed, a distributed network of Edge nodes creates real value only if all of them interpret data according to shared ontologies. Without common semantics, each node generates local knowledge that is hard to integrate with that produced by the others.

For a public decision-maker, designing an Edge infrastructure means building a network of territorial nodes distributed across hospitals, universities, industrial hubs, regional data centres, operations centres, energy networks and strategic infrastructure. Each node must be able to acquire data, verify its provenance and quality, apply security criteria, run AI models close to the information source, keep evidence of the operations performed and communicate with public, private or sovereign cloud infrastructures. The result is a hybrid architecture in which computing capacity is distributed according to operational needs: local when timeliness and control are paramount, central when more processing power is needed, federated when knowledge must be shared between territories and organisations. Building such an infrastructure requires stable collaboration between regional institutions, local administrations, universities, polytechnics, technology companies, critical infrastructure operators, telecommunications operators, manufacturing industry, standardisation bodies and the authorities responsible for cybersecurity and compliance. Each contributes complementary expertise: from architecture design to defining standards, from network management to model development, through to assessing security and organisational impact.

In this framework, Fondazione ENIA could act as a methodological link between the various actors, fostering dialogue between technology, governance, semantics, AI Literacy and innovation policy. The Congress therefore proposes launching the Italian Network of Edge Nodes for Public AI, an experimental programme to identify priority domains, design interoperable territorial nodes, define common technical requirements, integrate the architectures with public ontologies, test high-impact use cases and progressively assess results, security and models for regional and national scalability. The ultimate goal is to equip the country with distributed computing capacity that enables a safer, more resilient use of Artificial Intelligence, consistent with the needs of local areas. Digital sovereignty is not built through rules alone, but through infrastructure capable of keeping strategic processes under control while ensuring interoperability, security, operational continuity and decision-making autonomy. In this perspective, Edge Computing is the natural complement to the semantic infrastructure, turning shared meaning into distributed operational capacity.

Having defined a common language through ontologies and built computing capacity through distributed infrastructure, the third pillar concerns people’s ability to understand, govern and use these tools knowingly. Artificial Intelligence is not just a technology, but a cognitive transformation that changes the way individuals, organisations and institutions produce knowledge, make decisions and interpret reality. AI Literacy is therefore the cognitive infrastructure of the AI ecosystem. It does not coincide with traditional digital literacy, nor with the simple ability to use Artificial Intelligence applications or platforms. Rather, it concerns understanding the principles that govern how intelligent systems work: how models are built, what data they use, how they learn, what their limits are, what errors they can produce and what conditions determine their reliability. This awareness is now a strategic factor for economic competitiveness, institutional resilience and the quality of democracy.

Every technological revolution has been accompanied by a corresponding cultural evolution: literacy supported industrial society, computing accompanied the digital transformation, and AI Literacy is the precondition for the new knowledge economy. What sets Artificial Intelligence apart is its ability to produce content, formulate hypotheses, synthesise information, assist decision-making and generate new knowledge. This creates a new form of inequality, less and less linked to access to technology and more and more to the ability to understand how it works. The cognitive infrastructure must therefore develop along six complementary directions. The first concerns understanding the computational foundations of AI, learning models, data and inferential processes. The second concerns the quality of information: the provenance of data, its reliability, traceability and contextualisation. The third concerns the semantic dimension, i.e. the ability to understand how ontologies, taxonomies and knowledge representations are built, which links this pillar directly to the previous one on public semantics. The fourth concerns ethical and legal aspects, including responsibility, transparency, accountability, protection of rights and the management of algorithmic risks. The fifth concerns the organisational dimension, needed to integrate Artificial Intelligence into decision-making without creating new vulnerabilities. Finally, the sixth concerns the strategic dimension, essential for understanding the impact of AI on competitiveness, innovation, work, economic security and technological sovereignty.

Together, these skills make up what can be called the national cognitive capital for Artificial Intelligence. Building it is now a strategic priority at least as important as developing digital infrastructure. In the coming years, the main divide will not be determined only by access to technology, but by the ability to understand it. Alongside the traditional digital divide, an AI Divide is gradually emerging, set to separate citizens, businesses and institutions able to govern Artificial Intelligence from those who will only use its interfaces, remaining dependent on decisions and logics developed by others. For a small business, this means being able to critically assess the tools it intends to adopt, avoiding technological dependencies and unsustainable investments. For a Public Administration, it means introducing intelligent systems while retaining control over decision-making and administrative responsibility. For citizens, it means developing the cultural tools needed to recognise information manipulation, synthetic content, bias, algorithmic hallucinations and new forms of disinformation. AI Literacy thus becomes a preventive measure of collective resilience and cognitive security.

Building this infrastructure requires coordinated action. The Regions can integrate AI Literacy into innovation, economic development and training policies; universities and research centres can develop interdisciplinary programmes combining computing, law, economics, organisation and social sciences; the school system is the first line of critical education on Artificial Intelligence; businesses and public administrations must promote permanent professional development programmes; professional orders can help spread specialist skills in the various fields of application, while the Third Sector plays a fundamental role in reaching the sections of the population most exposed to the risk of cognitive exclusion.

In this context, Fondazione ENIA could take on a methodological coordination role, supporting the definition of national standards, maturity metrics, certification pathways and shared AI Literacy models. The Congress therefore proposes launching a National Plan for AI Literacy and Cognitive Resilience, aimed at building a unified skills framework, promoting local literacy programmes, developing systems to assess cognitive maturity and supporting a permanent network of collaboration between schools, universities, businesses, public administrations and the Third Sector. If the first pillar builds meaning and the second creates the computing infrastructure, the third builds the ability to understand both. The societies that will lead the evolution of Artificial Intelligence will not necessarily be those with the most powerful models, but those that have invested in spreading knowledge, critical thinking and the capacity to govern technology. In this perspective, AI Literacy takes on the value of a true national strategic infrastructure, set to shape the quality of public decisions, the competitiveness of businesses and the country’s ability to face the transformation of the economy and society autonomously.

If the first pillar builds meaning, the second creates the computing infrastructure and the third develops cognitive capital, the fourth defines how these elements are coordinated within organisations and institutions. Governance is the organisational infrastructure of the AI ecosystem: the set of rules, responsibilities, processes and coordination mechanisms that turn a technology into a stable capability of the economic and institutional system. Every complex infrastructure requires a governance model. Power grids have dispatching systems, airports operate through control towers and railway networks are coordinated by operations centres that ensure safety, continuity and synchronisation. Artificial Intelligence requires a similar organisational capacity. The availability of data, computing capacity and widespread skills is not enough without a structure capable of directing, coordinating and monitoring the functioning of the whole ecosystem.

AI governance therefore cannot be reduced to compliance with current legislation. The AI Act is a fundamental reference for regulating the obligations and responsibilities of providers and deployers of Artificial Intelligence systems, but governing AI requires a broader perspective. It means designing decision-making models, assigning responsibilities, defining organisational roles, building reliable information flows, setting up monitoring systems and promoting continuous improvement throughout the life cycle of intelligent solutions. In this sense, governance is the nervous system of the entire national AI infrastructure. Artificial Intelligence profoundly changes how organisations work because it intervenes directly in decision-making. Whether it is granting credit, making a clinical diagnosis, managing traffic, predictive maintenance of infrastructure or allocating public resources, every algorithm introduces new decision criteria and redistributes responsibilities that previously belonged exclusively to people. It therefore becomes necessary to establish who guarantees data quality, who authorises the introduction of models, who assesses risk, who monitors their performance over time, who intervenes in the event of anomalies, who keeps documentary evidence and who ensures dialogue with the competent authorities. When these responsibilities are not defined in advance, technology evolves faster than the organisational capacity to govern it. This is one of the main critical issues of the current phase of AI diffusion. Many organisations are investing in buying intelligent tools, while only a limited number are building a real capacity to govern innovation.

To meet this need, governance must be conceived as a multi-level system. At the strategic level, governing bodies and senior management define vision, objectives, investments and risk appetite. The organisational level identifies roles, responsibilities and skills, involving figures such as the Chief AI Officer, Risk Manager, Data Protection Officer, information security managers and process owners. The operational level governs the life cycle of data and models, including validation, monitoring, cybersecurity, technical documentation, updates and incident management. Added to this is algorithmic governance, dedicated to the continuous observation of model performance through indicators of accuracy, robustness, explainability, fairness, data drift and operational reliability. Finally, institutional governance ensures coordination between central administrations, Regions, local authorities, universities, standardisation bodies, businesses and independent authorities, recognising that Artificial Intelligence is a cross-cutting infrastructure that no single actor can govern alone. Governance must also be understood as a continuous process rather than a static set of procedures. Every AI system goes through a life cycle that includes identifying the need, preliminary assessment, risk analysis, model selection, validation, implementation, monitoring, updating, periodic audits and, where necessary, decommissioning. Each of these phases generates new evidence, new risks and new decisions. The function of governance is to keep this process under control, ensuring its transparency, traceability and adaptability.

Building national governance therefore requires permanent collaboration between institutions, local areas, universities, businesses, standardisation bodies and independent authorities. Central administrations set strategic direction and inter-institutional coordination; the Regions translate this direction into territorial policies; public administrations design organisational models; academia contributes research and advanced training; businesses develop applied skills and industrial capabilities; standardisation bodies define standards and assessment methodologies; independent authorities protect rights and ensure the right balance between innovation and the public interest. In this context, Fondazione ENIA can act as a methodological link between the various actors in the ecosystem, promoting shared languages, common methodologies, experimentation pathways, interdisciplinary training and ongoing dialogue between research, industry and institutions.

The Congress therefore proposes setting up a National Control Room for the Governance of Artificial Intelligence, conceived not as a new regulatory body, but as a stable coordination platform between administrations, local areas, universities, businesses and civil society. Its purpose would be to foster the convergence of organisational models, assessment methodologies, maturity indicators, operational standards and good practice, helping to build a coherent and interoperable national ecosystem. The Congress session on this theme will be the centrepiece of the whole programme, focusing on the shift from regulatory compliance to institutional capacity to govern. The aim is not only to discuss rules, but to outline an organisational model that allows Italy to govern Artificial Intelligence as a strategic national infrastructure. If the first three pillars build meaning, computing capacity and cognitive capital respectively, governance makes the coordinated functioning of the whole system possible. It is the organisational infrastructure that turns technologies, skills and data into a permanent capacity for innovation, ensuring that the development of Artificial Intelligence proceeds reliably, responsibly and in line with public-interest objectives.

If the first four pillars build meaning, computing capacity, cognitive capital and the organisational infrastructure of Artificial Intelligence, the fifth introduces a further dimension: the institutional system’s ability to understand how AI concretely transforms the lives of people and local areas. Every major innovation changes the relationship between citizens, institutions and the economy; Artificial Intelligence adds further complexity, since a growing share of decisions affecting everyday life is progressively supported, steered or automated by algorithmic systems whose functioning often remains barely visible to end users. For this reason, the AI infrastructure cannot be limited to its technological, organisational and regulatory dimensions. It must be accompanied by a democratic infrastructure capable of observing, interpreting and reporting back to institutions the social effects of adopting Artificial Intelligence.

Community Juries meet this need. They are not bodies of political control, nor deliberative assemblies meant to replace the competent institutions. Rather, they are a permanent infrastructure for social observation, aimed at systematically gathering evidence, analysing emerging needs and producing knowledge that can guide public policy and accompany the evolution of the AI ecosystem. Artificial Intelligence produces effects that can rarely be observed while an algorithm is being designed. The real consequences emerge when technologies enter administrative processes, health services, the labour market, credit, education, public services and citizens’ everyday lives. No research laboratory, technical audit or testing environment can fully capture these transformations. Many impacts concern trust, the comprehensibility of systems, social acceptability, perceived fairness, the emergence of new needs or organisational problems that only become apparent when technologies are actually used. Community Juries make this dimension observable. Citizens are no longer mere recipients of innovation but become qualified observers of its effects, contributing to a collective knowledge that integrates and completes that produced by technical monitoring systems.

Methodologically, the model works through a continuous process of evidence gathering, contextualisation, interdisciplinary analysis and institutional reporting. Usage experiences, problems, opportunities, emerging needs and weak signals are collected through structured procedures and then interpreted in light of the application domain, the organisational context, current legislation and the characteristics of the local area. Dialogue between citizens, experts and institutions then turns this information into reports, territorial indicators, operational recommendations and improvement proposals, feeding a permanent learning process for public administrations, businesses, the research community and governance bodies. In this perspective, Community Juries can be considered a true civic intelligence infrastructure. While Artificial Intelligence systems mainly process structured data, the Juries collect highly contextualised qualitative information that rarely finds room in algorithmic models: trust, risk perception, the comprehensibility of automated decisions, the quality of user experience, unexpected effects, emerging vulnerabilities and new community needs. Integrating this information completes AI governance with a social dimension that is still little explored today.

Building this infrastructure requires stable collaboration between regional institutions, local authorities, universities, research centres, businesses and the Third Sector. The Regions can coordinate the territorial network of Juries, integrating it into policies on innovation, healthcare, welfare, economic development and civic participation; Municipalities are the natural level of proximity to citizens, associations and businesses; universities and research bodies ensure methodological rigour in gathering and interpreting evidence; businesses can use the knowledge generated to improve the transparency, usability and social impact of their solutions; the Third Sector facilitates the involvement of communities and groups most exposed to the risk of exclusion. In this context, Fondazione ENIA can provide national methodological coordination, defining operating criteria, training pathways, quality standards, evidence-gathering methodologies and shared analysis tools.

The Congress therefore proposes launching the National Network of Community Juries for Artificial Intelligence, conceived as a civic infrastructure distributed at municipal, provincial, regional and national level, able to systematically collect the evidence generated by the introduction of AI in different economic and social contexts. The information produced by the network could feed a National Observatory on the Social Impact of Artificial Intelligence, combining quantitative and qualitative data to support public policy, research and innovation strategies. The closing session of the Congress will be devoted precisely to this prospect, with the aim of presenting the methodological model developed by Fondazione ENIA, sharing the first local experiences and encouraging institutions to take part in building a permanent network of civic observation of Artificial Intelligence. If the first four pillars make the development of the AI ecosystem possible, the fifth guarantees its democratic legitimacy. A technologically advanced society is measured not only by the quality of its infrastructure or the power of its algorithms, but by its ability to listen to, understand and integrate people’s experience into innovation processes. In this perspective, Community Juries make civic participation a structural component of Artificial Intelligence governance, helping to build an ecosystem in which technological progress, institutional trust and the public interest can evolve harmoniously and reinforce one another.

Every technological transformation becomes a real driver of development only when the economic and financial system acquires the ability to understand its value and support its growth. This happened with the industrial revolution, electrification, telecommunications and later the Internet. Artificial Intelligence is no exception. For this reason, the banking, insurance and financial system should not be involved only at the final stage of innovation, when technologies are mature and ready to be financed. Its contribution is strategic from the very first stages of building the ecosystem, because that is precisely when the criteria are defined through which an algorithm, an AI model or a new technological solution acquires economic recognition.

Today the main difficulty is not only financing Artificial Intelligence projects, but building a common language between research, businesses, investors and institutions. Algorithms have characteristics that are profoundly different from traditional assets: they evolve over time, learn from data, improve through use, generate new knowledge and can be reused in different contexts. Their ability to create value depends simultaneously on data quality, model robustness, governance, scalability, interoperability and the trust the market places in how they work. This evolution makes it necessary to develop new tools for economic valuation, new due diligence methodologies, new insurability criteria and new models for financing innovation. It is equally important to define shared ways of qualifying algorithmic assets, estimating their prospective value, measuring their risk and understanding their contribution to business growth and national competitiveness.

The involvement of the financial system is therefore also decisive for technology transfer. Every year, universities, research centres and start-ups develop skills and technologies that struggle to become industrial initiatives because there is no shared framework enabling financial operators to interpret their value correctly. Closing this gap means accelerating the transition from research to market, encouraging the creation of new high-tech companies and supporting the growth of existing ones. Within the ENIA Congress, this topic could be addressed as a permanent forum between institutions, universities, the banking system, insurers, investment funds, innovative companies and research bodies, with the aim of building a shared vocabulary on the economics of Artificial Intelligence and identifying tools to finance innovation more effectively.

The proposal Fondazione ENIA intends to promote is the launch of a national pathway dedicated to the economic qualification of algorithmic assets and the definition of shared methodologies for their valuation, bankability and insurability. Such an initiative would connect the world of research with that of finance, helping to create the conditions for Artificial Intelligence to become not only a lever of technological innovation, but also a stable driver of economic and industrial development. In this perspective, the banking and insurance system is not simply a financier of the AI ecosystem. It becomes one of the actors that help define its value, make it understandable to investors and turn technological potential into economic growth for local areas and the country.