Digital Archaeology

Between Innovation and Theoretical Continuity

Martin Hinz

Institute of Archaeological Sciences, University of Bern

2025-03-26

Take off

Story Time

Vacant wall in the Louvre’s Salon Carré after the painting was stolen in 1911. Source: Wikipedia Measurement system of the Bertillonage. Source: Wikipedia Vincenco Peruggia Source: Wikipedia

On August 21, 1911, the Mona Lisa was stolen from the Louvre. Paris was in shock. The police, determined to find the thief, turned to what was considered the most advanced forensic science of the time: Bertillonage.

Developed by Alphonse Bertillon, this system used detailed body measurements to identify suspects. It was rigorous, mathematical, and seen as infallible. But it had a fatal flaw: it was nearly impossible to search efficiently.

Ironically, the thief—Vincenzo Peruggia—had left his fingerprints behind. His prints were already on file. But fingerprinting, a newer and simpler technique, was dismissed by Bertillon as unreliable. So the case remained unsolved for over two years.

What’s the lesson here? Bertillonage wasn’t just flawed—it was a product of its time. It created a technological path dependency: a commitment to a method long after its usefulness had passed.

This story is a parable for us in digital archaeology. We work with powerful digital tools—databases, models, algorithms—that feel objective and advanced. But we must ask: are we shaping these tools, or are they shaping us? Are we leaving better options unexplored because we’re invested in the systems we already know?

Introduction

Digital archaeology is more than technical aid—it’s reshaping how we think.
Tools influence methods, but also our questions, categories, and interpretations.
We must ask:
- Are we shaping our tools—or are they shaping us?
- Can we ensure that theory keeps pace with technology?

Inherited Structures

The Legacy of Early Digital Archaeology

Digital Foundations (1960s–1990s)

Statistical models (seriation, clustering)
Databases (standardised catalogues, structured forms)
GIS (site-to-landscape shift)

Lasting Influence:

Prioritised structure over ambiguity
Data formats still influence interpretation
Created deep methodological path dependencies

Hodson 1970, Cluster Analysis in Archaeology Fig. 7.

From Ground to Grid

Impact:

Quantification over narrative
Codified what was worth recording
Landscapes over places

Risks Today:

Are we still locked into old categories?
Can we break the grid without losing rigour?

Punch card – Humboldt field research, Sudan, after 1967. HU Sudanarchaeological Collection

Abstraction and Scale

From Excavation to Extrapolation

Archaeology Without Digging

Remote sensing, predictive models, 3D scanning
From trench to terrain: new perspectives on place
Rise of abstraction and pattern-based reasoning

Challenge: - Does data distance us from material experience?

SYMAP plot – Arnold 1982

When Data Takes Over

Big data enables scale: SPDs, ML, regional models
Patterns emerge—but context may fade
We scale up—but also step away from the trench

Questions:

What gets lost when we generalise?
Are we prioritising what’s computable over what’s meaningful?

Digital Logic, Analogue World

Scaling up brings distance.

But the past isn’t neat.

It’s contradictory, ambiguous, lived
Regularity is tempting—but can erase meaning

Theory not optional any longer:

the scaffold when granularity is lost.
the lens that questions objectivity itself

As we move deeper into pattern and simulation, we risk losing the interpretive friction of the material record. Trenches hold more than data—they hold ambiguity, contradiction, sensory context. These aren’t noise; they’re the texture of the past.

Models, by contrast, reward regularity. They prefer clean categories and clear boundaries. That’s where theory becomes essential—not as ornament, but as scaffolding.

The further we move from the trench, the more we need theory to hold our abstractions in check. Abstraction isn’t the enemy—it can reveal structure—but only if we remain aware of what it flattens.

Theory doesn’t slow us down; it keeps us accountable. It reminds us that what appears objective often reflects choices we’ve already made. We mustn’t forget: behind every machine is a modeller.

Story time

Windisch, Karl Gottlieb von: Karl Gottlieb von Windisch’s Briefe über den Schachspieler des Hrn. von Kempelen. Source: Bayerische Staatsbibliothek, CC0.

Toward a Reflective Digital Archaeology

The Interpreter Inside

Tools as Theories in Disguise

Digital tools seem neutral.
But behind every model, there’s a modeller.

Bayesian models → reflect assumptions about time
ML classifiers → mirror training biases
Databases → encode structured categories and choices

Just like the Mechanical Turk, our tools hide
a human mind—shaping outcomes from behind the curtain.

When Assumptions Become Infrastructure

What makes assumptions powerful?

Not that they exist — but that they go unseen.

They hide in:

“Default” categories and clean data
Proxies standing in for complex phenomena
Standards that fossilise early decisions

⸻

If we don’t surface these assumptions, they become infrastructure — invisible, inherited, hard to question.

Photo by Susan Q Yin at Unsplash

Now here’s where it gets even trickier.

The issue isn’t just that digital tools embed assumptions — it’s that over time, those assumptions become part of our infrastructure. They turn into “defaults,” standards, data formats. And when that happens, they stop being questioned.

Think of categories that were once temporary simplifications — like using a single function label for complex artefacts, or a particular proxy for population. Over time, they solidify. They become part of the system — and the system starts shaping research silently.

This is where responsibility shifts from individual interpretation to collective practice. Because once assumptions become infrastructure, they’re harder to notice — and even harder to change.

And that’s why ethics must come early — not after the tool is built, but before the infrastructure becomes invisible.

Ethics of Infrastructure

Digital infrastructures are not neutral.

They shape:

What becomes visible
Whose knowledge is counted
Who gets access—and who is excluded

Key questions:

Who sets the defaults?
Whose perspectives are built in?
How do we include contested or local voices?

→ Open science is not just technical — it’s an ethical and political commitment.

Photo by Giammarco Boscaro at Unsplash

Infrastructure sounds neutral—but it never is. Behind every “default,” every metadata field, every platform governance decision lie embedded assumptions. And with that comes power: power to shape what is seen, what is valued, and who is heard. If we don’t reflect on these structures, we risk building exclusion into the foundations of our digital practice. That’s why open science isn’t just about data access—it’s a political and ethical commitment to inclusion, transparency, and pluralism. Who defines the defaults? Who gets to participate? And who is left out?

These are not questions we can fully answer in a single presentation — or a single project. But we can try to approach them reflectively in practice. The next two examples aren’t presented as perfect solutions. But they do represent steps toward a more reflective digital archaeology — one where theory and transparency are built into the process.

The Shape of the Past?

Looks like population change.
But is it?

Do SPDs reflect the past—or our models of it?
Are curves shaped by theory, or by data availability?
What’s invisible: sampling bias, taphonomic filters, research intensity?

Models must make theory explicit, not hide it
Represent uncertainty — don’t flatten it

ESTER logo ESTER kicks off in Kiel, Sept 2019

Background: Hinz et al. 2019

Summed probability distributions—SPDs—are now common tools to explore past population trends. We see elegant curves suggesting demographic booms and busts across prehistory.

But how do we interpret them?

Do these curves actually represent the past—or do they reflect how we’ve chosen to model it?

There’s a risk that SPDs are driven less by demographic reality, and more by data availability, sampling strategies, or theoretical assumptions we’ve built into the process without fully acknowledging them.

This is where the ESTER project comes in. It doesn’t just generate curves—it asks how theoretical frameworks shape demographic modelling. It explores how to build population models that are not only computationally robust, but interpretively transparent and empirically grounded.

In short: don’t let the curve do the thinking. Let theory guide the modelling — explicitly, reflexively, and critically.

The past of the shape

Neural networks classify pottery (eg. ArchAIDE) and lithics
But… what do these models really “see”?

Critical questions:

Are we encoding past typologies — or rethinking them?
Do models preserve nuance — or flatten diversity?
Is the theory behind the labels transparent?

The ideal? Archaeologists and algorithms in dialogue

Shape categories. Interpret results. Iterate.

MET & MOVE Projects use unsupervised learning and mixed methods to explore reflective artefact classification

Another rapidly advancing frontier in digital archaeology is machine learning, especially when it comes to artefact classification. Neural networks are now able to classify pottery or lithics based on thousands of images. Projects like ArchAIDE have shown how powerful these tools can be. I am not criticising this, these are excellent applications. But we need to ask: What are these models actually seeing? Are we genuinely rethinking our categories, or are we simply encoding old typologies in a new digital form? Many classification systems we now train machines on originated decades ago—are we teaching algorithms the past of our discipline, rather than the past we aim to uncover? There’s also the risk of flattening diversity. Machine learning often requires categories to be clean, stable, and distinct. But archaeological material is messy. Cultural boundaries are fluid. Typologies are interpretations, not ground truths. So how do we move forward? Ideally, machine learning becomes a heuristic process—not a replacement for expert judgment, but a tool archaeologists actively shape. We should be involved in designing training datasets, interpreting output, and questioning classifications. That’s precisely the approach we’re taking in the MET and the MOVE project. We’re exploring unsupervised learning techniques—letting the data speak first—and then using mixed methods to critically reflect on what the machine sees, what it misses, and how it might actually help us discover new patterns, rather than reproduce old ones. Ultimately, this isn’t about automating archaeology. It’s about expanding our interpretive range—with the help of well-understood, critically applied digital tools.

The Future of Digital Archaeology

From Methods to Mindsets

We’ve seen how digital archaeology can be reflective and theory-driven. But that requires a shift—not just in tools, but in how we train ourselves.

What’s next?

Digital archaeology ≠ button pushing
Reflexivity, theory, and interdisciplinarity are core skills
Let’s talk about education.

Beyond Code – Digital Literacy as Critical Thinking

Digital literacy ≠ just tech know-how. It’s understanding what tools do—and what they hide.

Key questions:

How do algorithms encode bias?
How does data structure influence interpretation?
What does it mean to “know” something computationally?

→ We don’t just need archaeologists who can code.
→ We need archaeologists who can think critically about code.

Toward Reflective Practice

Five Pillars for Responsible Digital Archaeology

Document assumptions - Make modelling and classification choices explicit.
Design for flexibility - Allow ambiguity, multiplicity, and change.
Think with theory - Let interpretation guide tool use—not the reverse.
Share openly - Build inclusive, transparent infrastructures and practices.
Train reflectively - Embed critical thinking and ethical awareness into every digital skill we teach.

Photo by Emma Henderson at Unsplash

To bring everything together, I want to end with five guiding practices for a more reflective digital archaeology.

They’re not meant as rigid rules—but as foundations we can build on: 1. Document assumptions – Make modelling choices explicit. Let others see the scaffolding behind your conclusions. 2. Design for flexibility – Allow room for uncertainty, ambiguity, and alternative interpretations. Don’t let your data structures lock in a single way of seeing. 3. Think with theory – Let theoretical reflection guide your use of tools—not the other way around. 4. Share openly – Transparency and accessibility are ethical choices. From code to collaboration, openness broadens participation. 5. Train reflectively – Digital literacy must include critical awareness. We need archaeologists who understand what tools do—and what they conceal.

These five aren’t the destination—but they offer a direction: toward a digital archaeology that’s not just efficient or impressive, but thoughtful, inclusive, and well grounded.

Final Thought

Are we building tools with vision—or just tools with convenience?

Our choices today shape the archaeology of tomorrow.

“Every tool is also a way of seeing— and a way of not seeing.” — Inspired by Langdon Winner

Thank you.

Find this presentation here: https://martinhinz.github.io/kielconference2025