Introduction: A New Frontier in Digital Archaeology
Researchers at Griffith’s Centre for Social and Cultural Research have piloted a digital archaeology framework to explore who created ancient finger flutings in cave walls. Finger fluting, marks made by fingers through a soft mineral film called moonmilk, is one of the oldest forms of rock art known from Europe and Australia. By combining tactile experiments with virtual reality (VR), the team sought to determine whether image-recognition methods could infer the sex of the fluting makers from the marks they left behind.
Why Finger Flutings Matter
Finger flutings appear in pitch-dark caves dating back hundreds of thousands of years. The oldest known examples in France have been attributed to Neanderthals, around 300,000 years ago. Understanding who made these marks can illuminate ancient social structures, access rights to sacred sites, and the cognitive practices of early humans. Yet traditional methods for identifying the maker—such as measuring finger sizes or hand proportions—have proven unreliable. Variability in pressure, surface texture, pigment distortion, and overlapping size ranges often blurred the signal investigators hoped to detect.
The Griffith Digital Archaeology Approach
The project was led by Dr. Andrea Jalandoni, a digital archaeologist, who championed a data-driven path to replace brittle measurements with machine learning. The team designed two controlled experiments with 96 adult participants. Each participant produced nine flutings twice: once on a moonmilk-clay substitute designed to mimic real cave surfaces, and once in VR using the Meta Quest 3 headset. The intent was to collect robust datasets that could reveal generalizable patterns rather than artifacts of a single experimental setup.
From Hands-on to Headset: The Two Experimental Tracks
The tactile experiments yielded the most promising results. Images captured from the physical flutings were then used to train two common image-recognition models. In this pragmatic pipeline, the researchers also stressed the importance of guarding against overfitting—ensuring that models learned generative features of genuine fluting rather than idiosyncrasies of the specific experiment environment. In contrast, the VR-generated data struggled to produce reliable sex classification. While some moments showed acceptable accuracy, overall discrimination and balance remained weak, underscoring the challenge of translating in-situ fluting cues into VR imagery.
What the Findings Suggest About Machine Learning in Archaeology
Dr. Gervase Tuxworth notes that the tactile data offered clearer signals about potential sexual dimorphism in flutings, achieving around 84 percent accuracy under certain training conditions. Yet even with encouraging metrics, the team emphasized that the models learned patterns tied to their own dataset—artefacts of the experimental setup rather than universal features of finger flutings. This outcome is not a failure but a crucial learning point: robust, generalizable models require diverse data and transparent, replicable workflows.
A Transparent, Reproducible Blueprint for Future Research
A standout aspect of the Griffith study is its commitment to openness. The researchers released the computational pipeline, the tactile representation, and the VR capture environment as open resources. Dr. Robert Haubt describes the project as a proof-of-concept that others can replicate, critique, and improve. The open framework invites archaeologists, computer scientists, and cognitive researchers to test hypotheses across species, sites, and cultural contexts, paving the way for more rigorous interdisciplinary work.
Implications for Archaeology, Forensics, and Beyond
Beyond deciphering who made ancient marks, the study has implications for forensics and psychology, where understanding how people interact with textured surfaces and how those interactions translate into digital signatures is increasingly relevant. The framework could extend to other early human art forms or even to non-artifact domains where subtle motor patterns hold clues to behavior and identity. In doing so, Griffith researchers emphasize a broader question: what can machine learning tell us about cognitive and cultural practices from the deep past, and how can we ensure those signals are genuinely informative rather than artefacts of experimental design?
Conclusion: A Roadmap for Interdisciplinary Discovery
The finger-fluting project marks an important step in modern archaeology: a transparent, scalable, and testable method to study ancient creators through digital means. While more work is needed to extract robust, generalizable indicators of the maker’s identity, the team’s open data, open code, and rigorous evaluation set a high standard for future interdisciplinary investigations across archaeology, forensics, psychology, and human-computer interaction. The ultimate aim is not only to identify who made the marks but to enrich our understanding of early human behavior and cultural practice, using technology as a careful, collaborative instrument.
