Abstract:
Prehistoric stone tools are one of the most important types of evidence for the study of hominin evolution, dating back at least 2.6 million years. These tools were most commonly made through the (repeated) fracture of a stone to detach from it a flake, creating—for example—useful cutting edges on the flake surface. This process of flake removal is known as knapping or lithic reduction. One method used to study lithic reduction is its experimental replication by modern humans, making inferences from the process on the various behavioural, cognitive, or even social learning requirements needed to make these artefacts. Historically, many researchers have suggested that the earliest stone tools require—and thus are evidence for—copying of know-how information, which is intrinsic to modern human culture, but which is also not present in any extant non-human ape. On the other hand, recent research has advanced an alternative hypothesis: suggesting many of the earliest stone tools did not require know-how copying, but that they were manufactured through a simple set of rules and with a base set of skills closer to those of non-human apes than modern humans (the ‘Zone of Latent Solutions’ hypothesis). Nevertheless, in order to robustly test how changing certain variables (e.g. maximising for flake length or area) can affect the products of knapping, many replicable large-scale lithic replication experiments are necessary. These experiments generally require considerable amounts of time, material, and funds to undertake, from the lithic reduction itself, to the cataloguing, storage, measuring, and analysis of the products at each step of the reduction sequence, and are inevitably subject to knapper-derived biases (e.g. experience, motivation, and fatigue during knapping). With replication experiments that can last months, a fully computer-based analogue to rules-based lithic reduction could prove a powerful tool for the study of human behavioural, cognitive, and cultural evolution. Such software could be effective for testing various hypotheses on the factors that affect stone tool manufacture, such as the necessity of know-how copying, helping advance evolutionary science more broadly or as a tool for teaching and outreach, as well as serving as a base for the development of additional tools. This dissertation presents a proof of concept for a computer program based on a machine learning framework able to perform lithic reduction at a fraction of the time and cost, and without knapper-derived biases: a Virtual Knapper. In addition, this work also discusses the results of a proof of concept for another application based on the same machine learning framework: a Virtual Refitter.