AI and 3D Tech Decode Ancient Bone Marks

Determining whether early hominins actively hunted prey or merely scavenged leftovers is a foundational question in paleoanthropology, often answered by analyzing microscopic marks left on fossilized animal bones. Historically, distinguishing between stone-tool cut marks and carnivore tooth marks relied on subjective visual inspections by experts, leading to heated debates over humanity's earliest behaviors. Today, zooarchaeologists are resolving these disputes by combining 3D geometric morphometrics with artificial intelligence. By capturing ultra-high-resolution three-dimensional models of bone surface modifications and feeding this data into machine learning algorithms, researchers can now objectively and accurately classify marks created millions of years ago, fundamentally reshaping our understanding of early human diets and tool use.

The Problem with Traditional Bone Analysis

For decades, zooarchaeologists used hand lenses and standard microscopes to look at Bone Surface Modifications (BSMs). The general rule was simple: stone tools leave V-shaped grooves with fine parallel micro-striations inside them, while carnivore teeth leave U-shaped grooves with crushed bone at the bottom.

However, nature rarely leaves pristine evidence. Weathering, sediment abrasion, and the specific angles at which a tool or tooth hits the bone can make a tooth mark look like a cut mark, and vice versa. Because traditional methods relied heavily on the individual researcher's experience, two experts looking at the same two-million-year-old bone often reached completely opposite conclusions.

Capturing the Data: 3D Geometric Morphometrics

To remove human subjectivity, researchers turned to 3D geometric morphometrics—the quantitative analysis of form. Instead of just looking at a mark, scientists use structured-light scanners, laser scanners, or confocal microscopes to create a digital 3D topography of the bone fragment down to the micron level.

Once the 3D model of a mark is generated, researchers place "landmarks" along the groove. These landmarks are specific mathematical points that measure the mark's exact geometry, including:

• Depth and width ratios
• The angle of the walls
• The curvature of the trajectory
• The roughness of the groove's floor

Next, a mathematical technique called Procrustes superimposition is applied. This process strips away non-shape variables like size, orientation, and position, leaving behind pure shape data. This turns a physical scratch on a bone into a complex dataset of three-dimensional coordinates.

Training the Artificial Intelligence

Before the AI can analyze mystery fossils, it must be taught what known marks look like. Zooarchaeologists create an experimental baseline—or training dataset—by producing modern marks under controlled conditions. They butcher fresh animal bones using replicas of ancient stone tools (like Oldowan flakes) and feed other bones to captive carnivores like lions, hyenas, and crocodiles.

These modern marks are scanned and converted into 3D morphometric data, or processed as high-resolution 2D depth-map images. This data is then fed into machine learning models, such as Support Vector Machines (SVMs), Random Forests, or Deep Convolutional Neural Networks (CNNs).

The AI analyzes thousands of these known marks, learning the subtle, microscopic patterns that the human eye cannot perceive. It learns, for example, how the microscopic flaking of a flint tool edge alters a groove differently than the enamel ridges of a hyena's premolar.

Distinguishing Butchery from Scavenging

Once the AI model is trained and achieves an accuracy rate of over 90% on test samples, it is ready to analyze actual fossilized bone fragments.

When a fossil mark is scanned and input into the system, the AI compares its 3D geometry and microscopic texture against its training data. It does not just provide a "yes" or "no" answer; it outputs a statistical probability. For instance, the AI might classify a specific mark as having a 96% probability of being made by a stone tool and a 4% probability of being a carnivore tooth score.

This objective classification allows zooarchaeologists to reconstruct the sequence of events at ancient sites. If a bone shows high-probability stone-tool marks on meat-bearing shafts, overlaid by carnivore tooth marks, it proves hominins had primary access to the carcass (hunting or aggressive scavenging) and carnivores ate the scraps. Conversely, if tooth marks dominate the prime meat areas and stone-tool marks are only found on bones broken for marrow, it indicates hominins were passive scavengers, arriving only after the predators had finished.