Working Paper No. 15 Towards a Theory of Cognitive Archives Archives as Institutional Ecosystems of Learning, Memory, Trust and Governance

 

PDS & Ged/A | Open Research Series

Working Paper No. 15

Towards a Theory of Cognitive Archives

Archives as Institutional Ecosystems of Learning, Memory, Trust and Governance

10 August 2026

---

🌎 Open Science

This Working Paper marks an important milestone in the research programme developed by the PDS & Ged/A Research Group. Rather than introducing a single new concept, it integrates the theoretical foundations progressively developed throughout the previous fourteen Research Notes into a broader framework for understanding archives in the age of Artificial Intelligence, Digital Preservation and Institutional Digital Transformation. As with every publication in this Open Research Series, the ideas presented here remain provisional, open to debate and intended to stimulate collaborative scientific discussion.

---

Returning to the laboratory

The first research seminar of the second academic semester, held on 3 August 2026, proved to be considerably more significant than we initially expected.

After several weeks without formal meetings, researchers returned bringing new readings, conference experiences, software experiments, methodological reflections and, perhaps most importantly, entirely new questions.

One observation quickly became evident.

Research groups do not stop researching simply because meetings temporarily stop.

Questions continue developing.

Ideas mature.

Hypotheses evolve.

Sometimes they are strengthened.

Sometimes they are abandoned.

The summer break had not interrupted our research programme.

Instead, it had expanded it.

Each participant returned carrying different perspectives shaped by independent study, international discussions and the extraordinarily rapid evolution of Artificial Intelligence during recent months.

Perhaps this was the first conclusion emerging from our meeting.

Scientific research never truly pauses.

Only its conversations temporarily become silent.

---

Looking back before moving forward

A considerable portion of our discussion was devoted to reviewing the intellectual path that had brought us to this point.

Initially our research focused on Systemic Digital Preservation, proposing that digital preservation should be understood not merely as technological maintenance but as the continuous preservation of authenticity, provenance, documentary context and institutional continuity.

Subsequently we introduced the concept of Digital Archival Ecosystems, recognising that archival records never exist independently but always within complex networks of institutions, functions, people, technologies and documentary relationships.

Later we revisited the OAIS Reference Model from an archival perspective.

We expanded the concept of the Chain of Archival Digital Custody.

We explored the convergence between the European archival tradition, Information Science, the iSchools movement and Computational Archival Science.

Our attention then shifted toward Artificial Intelligence, Archival Algorithmic Governance, Intelligent Archival Ecosystems, Intelligent Archival Records and Intelligent Archival Metadata.

Looking retrospectively at those fourteen Working Papers, something gradually became clear.

None of them had been isolated investigations.

Each represented one stage of a much larger research programme.

Each question naturally generated the next.

Together they seemed to converge upon a single overarching question.

How do institutions learn without losing their documentary memory?

---

Perhaps we were asking the wrong question

For many months our discussions revolved around a familiar theme.

How will Artificial Intelligence transform archives?

Returning from the academic break, we realised that this question might no longer be sufficient.

A different question gradually emerged.

How can Archival Science guide the development of intelligent institutional ecosystems without sacrificing authenticity, provenance, accountability and public trust?

This subtle reformulation changes everything.

Artificial Intelligence no longer occupies the centre of our theoretical framework.

Archival Science does.

Artificial Intelligence becomes one component within a broader institutional architecture designed to preserve documentary evidence, organisational memory and democratic accountability.

Technology no longer defines the archive.

Archival principles define how technology should operate.

---

A discussion that reshaped our thinking

One particular comment generated almost an hour of discussion during the seminar.

"Perhaps archives have never been merely places where memory is stored. Perhaps they have always been environments where institutions learn."

Initially this statement appeared provocative.

Yet the discussion gradually revealed its significance.

Organisations continuously produce records documenting their activities.

Those records preserve institutional experience.

Future decisions are frequently based upon previous documentary evidence.

New records then document those decisions.

Institutional learning therefore becomes a continuous documentary cycle.

Artificial Intelligence did not create this process.

Digital transformation merely makes it more visible, faster and computationally accessible.

---

Towards a Theory of Cognitive Archives

The central conceptual proposal emerging from our discussion may be summarised as follows.

We propose understanding Cognitive Archives as institutional ecosystems in which archival records, metadata, organisational policies, technologies, intelligent agents, administrative processes and human expertise continuously interact to preserve, interpret, reuse and expand institutional memory.

The cognition does not belong to the archive itself.

Nor does it belong to Artificial Intelligence.

It belongs to the institution.

Archives create the conditions through which institutions are able to learn responsibly from their own documentary memory.

Archival records preserve evidence.

Metadata preserve relationships.

Policies preserve governance.

Archivists preserve authenticity.

Intelligent agents extend operational capacity.

Institutional learning emerges from the interaction among all these elements.

This distinction became one of the strongest consensuses reached during our seminar.

---

Intelligence remains fundamentally human

Throughout the meeting we repeatedly returned to one concern.

Contemporary discussions surrounding Artificial Intelligence frequently attribute human characteristics to computational systems.

Our research deliberately rejects that interpretation.

Archives do not think.

Records do not think.

Metadata do not think.

Algorithms possess neither institutional responsibility nor ethical accountability.

Human beings remain responsible for institutional judgement.

Artificial Intelligence may support decision-making.

It cannot replace institutional responsibility.

Perhaps this distinction represents one of the most important contributions Archival Science can offer to current international debates surrounding trustworthy Artificial Intelligence.

---

PREMIS beyond preservation

One particularly productive discussion concerned the future role of PREMIS.

Traditionally interpreted as preservation metadata, PREMIS increasingly appears capable of supporting much broader governance functions.

Imagine an archival record preserved within a trusted digital repository.

A PREMIS event records that a legal restriction period has expired.

An intelligent agent consults institutional archival policies.

Another evaluates whether personal or sensitive information requires continued protection under applicable privacy legislation.

A third updates access permissions.

AtoM automatically publishes the archival description.

A Retrieval-Augmented Generation environment incorporates the newly accessible documentary context.

Every computational action generates new PREMIS events.

Nothing occurs without documentary evidence.

Nothing interrupts the Chain of Archival Digital Custody.

Automation therefore becomes accountable rather than autonomous.

---

Digital sovereignty and open infrastructures

Another recurring theme throughout our discussion concerned technological sovereignty.

Public archives, universities and memory institutions increasingly require open, sustainable and transparent digital infrastructures.

Open-source environments such as Archivematica, AtoM, RiC-O, PREMIS, local language models executed through platforms such as Ollama, Retrieval-Augmented Generation architectures, Knowledge Graphs and interoperable APIs may all contribute to such infrastructures.

Yet one principle repeatedly emerged.

Technology should always remain subordinate to archival governance.

Technological innovation is inevitable.

Archival principles remain indispensable.

---

Revising one of our hypotheses

Perhaps the most important intellectual outcome of our meeting was recognising that one of our earlier assumptions required revision.

For several months we had argued that Artificial Intelligence would transform archives.

Today we increasingly suspect something different.

Artificial Intelligence primarily transforms how we understand the institutional role of archives.

Archives gradually cease being perceived merely as repositories preserving historical documentation.

Instead, they increasingly become institutional infrastructures supporting organisational learning, transparency, accountability, governance and public trust.

This conceptual shift significantly expands the scope of Archival Science.

---

Research agenda for the second semester

Before concluding our meeting we collectively defined several research priorities.

During the coming months we intend to investigate:

• Actionable PREMIS.

• Integration between Archivematica, AtoM and intelligent agents.

• Secure communication through the Model Context Protocol (MCP).

• Knowledge Graphs derived from RiC-O.

• Archival Retrieval-Augmented Generation.

• Local Artificial Intelligence models for public institutions.

• Indicators measuring the maturity of Intelligent Archival Ecosystems.

• Documentary trust metrics.

• Archival governance policies for intelligent agents.

• Educational frameworks preparing archivists to design and govern Cognitive Archives.

This agenda represents considerably more than a list of research projects.

It reflects our commitment to ensuring that digital transformation strengthens authenticity, documentary evidence, institutional memory and democratic accountability.

---

A broader reflection

As our meeting concluded, one observation seemed particularly meaningful.

Perhaps the greatest achievement of the day was not discovering new answers.

Perhaps it was discovering better questions.

Questions capable of expanding our research programme while positioning Archival Science within international discussions surrounding Artificial Intelligence, Digital Government, Data Governance, Open Science and Trustworthy AI.

If our current hypotheses prove sustainable,

Archival Science will continue playing an essential role within increasingly intelligent societies.

Not because archivists will develop the most sophisticated algorithms.

But because the discipline remains uniquely responsible for preserving something no computational system can independently produce:

authenticity.

documentary evidence.

institutional memory.

provenance.

context.

public accountability.

trust.

Perhaps that ultimately represents the true meaning of Cognitive Archives.

Not teaching machines how to think.

But enabling institutions to continue learning without forgetting who they are, how they arrived where they are today, and why trustworthy documentary memory remains indispensable for democracy, public governance and future generations.