Literally clever components are now also enriching the archeology toolbox, as two current studies make clear: Researchers have used techniques from the field of artificial intelligence to assign stone artifacts and thus solve archaeological puzzles.
“Computer science of a special kind” has meanwhile found its way into our everyday lives, industry, medicine and research. The techniques from the field of artificial intelligence (AI) are based on simulations of intelligent behavior and so-called machine learning. The AI systems can capture certain principles in given examples and, after a learning phase, develop models on the basis of generalizations. These can then be used to assess unknown data.
A team led by Michael Thrun from the Philipps University of Marburg, together with Mexican archaeologists, has now shown the potential of AI for determining the origin of archaeological finds. These were workpieces made of obsidian stone glass that were found during excavations in Xalasco in western Mexico. There are indications that the inhabitants of this pre-Columbian site maintained lively contacts with other regions of Mexico. This probably also included the exchange of raw materials and products. So the question arose whether the source material for the numerous obsidian artifacts discovered in Xalasco came from local sources or was obtained from remote sites.
AI clarifies the origin
“We examined the chemical composition of the obsidian artifacts spectroscopically and analyzed these measurements with the help of artificial intelligence in order to compare the workpieces with samples that came from possible natural sources,” says Thrun. For chemical analysis, the researchers used a portable X-ray fluorescence spectrometer, with which they examined 256 obsidian finds from Xalasco. In order to assign the workpieces to the natural sources of obsidian, the team obtained samples from several Central American sites of the material.
As they explain, conventional computer programs for data analysis cannot clearly differentiate between the various archaeological sources. Instead, the authors used the new approach that makes use of machine learning. The process is based on a swarm of independent software units that interact with each other and thus form an intelligent system. The method was published earlier this year in the journal “Artificial Intelligence” released. “Our results show that this approach is suitable for an unbiased analysis of archaeological material,” says Thrun: With the help of AI, the obsidian workpieces could be classified into separate groups based on the origin of their raw material.
As far as the archaeological result is concerned, the findings now suggest that the Xalasco culture actually had a certain degree of exchange with more distant settlements. Most obsidian workpieces from Xalasco, however, are made of a material that was mined near the site, the archaeologists discovered with the help of artificial intelligence.
Artificial neural network differentiates between tool sets
The second example of a successful application of a technique from the field of AI is a method for differentiating between tool sets from the Middle and Late Stone Age. Researchers from the University of Liverpool and the Max Planck Institute for the History of Human History in Jena (MPI-SHH) have used an artificial neural network whose function is based on machine learning.
As they explain, the transition from the Middle to the Late Stone Age marked a decisive change in the culture of our ancestors who were still hunter-gatherers. Nevertheless, it is often difficult to assign archaeological finds to one period or another. Because the oldest tool sets from the time of the origin of our species were sometimes used up to around 30,000 years ago. However, 67,000 years ago there were changes in the stone tool making process that indicate a significant change in behavior.
As part of the study, the researchers have now examined to what extent the presence or absence of certain types of tools in tool sets enables them to be assigned to the Middle or Late Stone Age. For this purpose, 16 different tool types within 92 tool sets from Africa were analyzed. Instead of focusing on each individual tool, the focus was on capturing certain constellations of tool shapes that often appear together in the sets, the researchers explain. “We used the artificial neural network for this,” says co-author Matt Grove from the University of Liverpool. It was fed the data on the characteristics of the complex finds and was then able to uncover similarities and differences. “This technique enabled us to see how the various patterns in the composition of the tool kits differed between the Middle and Late Stone Age,” says Grove.
His colleague Jimbob Blinkhorn from the MPI-SHH explains: “If tools with blunt backs but sharp tips, blades and bipolar technologies do not work together with core tools such as hand axes, the Levallois technique and scraping tools, this reliably means a set from the late Stone Age conclude. The other way around is a set from the Middle Stone Age, ”says Blinkhorn.
With the help of their method, the team hopes to be able to better examine the regional differences in cultural change during the African Stone Age.
“We also hope that the study shows that our method could find a much wider application in archaeological research,” concludes Grove.
Source: Philipps University of Marburg, specialist article: IEEE Access, doi: 10.1109 / ACCESS.2020.3016244; Max Planck Institute for the History of Human History, specialist article: PLOS ONE, doi: 10.1371 / journal.pone.0237528