Unveiling Ancient Hands Through Digital Archaeology
Griffith researchers have taken a bold step into the deep past by marrying artificial intelligence with digital archaeology to study one of humanity’s oldest rock art traditions: finger fluting. These marks, created by fingers sliding across a soft mineral film known as moonmilk on cave walls, offer a doorway into the lives of ancient people. The study investigates whether modern image-recognition methods can learn enough from contemporary finger-fluting images to determine the sex of the creator, a question with profound cultural and access implications for ancient sites.
What is Finger Fluting and Why It Matters
Finger flutings are delicate grooves or marks produced when fingers traversed moonmilk layers on cave surfaces. Found in pitch-dark caves across Europe and Australia, some of the oldest known examples—located in France—have been attributed to Neanderthals, dating back around 300,000 years. Beyond art history, understanding who made these marks can influence who is allowed to access sacred or sensitive sites for cultural reasons. The Griffith project reframes this query through data-driven methods rather than traditional measurements.
From Finger Measurements to Digital Signals
In the past, researchers relied on physical attributes like finger size, pressure, or hand geometry to guess the artist’s identity. Those approaches proved inconsistent: finger pressure varied, surfaces differed, pigments distorted outlines, and feature ranges often overlapped between the sexes. The research team sought to move away from such assumptions by developing a digital pipeline that can generalize beyond a single dataset.
The Experimental Framework: Tactile and Virtual Reality Pathways
The study conducted two controlled experiments with 96 adult participants. Each participant created nine flutings twice: once on a moonmilk-like clay substitute designed to mimic real cave textures and a second time in virtual reality using the Meta Quest 3 headset. This dual approach offered a rich set of images for training and testing AI models under very different capture conditions.
Data Collection and Modeling
Images of all finger flutings were gathered and curated. Researchers trained two common image-recognition models on this dataset, then evaluated performance with standard metrics. A key focus was ensuring the models learned genuine patterns rather than memorizing idiosyncrasies of the experimental setup—a problem known as overfitting in machine learning.
What the Results Show
The VR-derived images were less reliable for sex classification, with inconsistent discrimination despite superficially decent accuracy. In contrast, tactile images—those produced via the moonmilk-substitute environment—showed stronger results. Under a particular training condition, the models reached approximately 84% accuracy, and one model demonstrated a notable discrimination score. Yet the researchers caution that the signals may be dataset-specific, and further work is needed to determine whether learned patterns would generalize to authentic cave contexts.
Interpreting the Findings
Dr. Andrea Jalandoni emphasized that the project’s aim was to pioneer digital archaeology rather than to deliver a ready-made method for sexing ancient artists. The findings illuminate the potential and the current limitations of using AI to infer biological attributes from art-like markings. Subtle artefacts from the experimental setup, rather than universal features of finger-fluting, sometimes dominated the models’ decisions. This underscores the need for broader datasets and rigorous validation before drawing firm conclusions about ancient behavior.
Open Science and the Road Ahead
One of the study’s hallmarks is transparency. The team released the computational pipeline, including open-source code and materials, to enable replication, critique, and scaling by others. Dr. Robert Haubt, a co-author, framed this as a crucial step in turning a proof of concept into a robust, reusable tool for archæology, forensics, psychology, and human-computer interaction. The researchers envision interdisciplinary applications that deepen our understanding of early human cognitive practices and cultural routines.
Implications for the Future of Archaeology
Beyond identifying potential gender-related patterns, the framework demonstrates how tactile and VR modalities can complement traditional artefact studies. As technology matures, this approach could contribute to broader questions about who created ancient art, how social roles manifested in early communities, and how access to culturally significant sites might be managed with better contextual knowledge.
