AI enters archeology! Scientists used deep learning algorithms to discover evidence of human use of fire nearly 1 million years ago, published in PNAS

The use of fire was a key factor in the evolution of Homo sapiens. Fire was not only used to create more complex tools, but also made food safer, thus helping to brain development.

To date, only five sites with evidence of fire use dating back 500,000 years have been found worldwide, including Wonderwerk Cave and Swartkrans in South Africa, Chesowanja in Kenya, and Israel Gesher Benot Ya'aqov, Spain's Cueva Negra.

Now, an Israeli research team has used artificial intelligence algorithms to discover a sixth site that shows signs of human use of fire! This study reveals evidence of human use of fire at an Upper Paleolithic site in Israel. The research results have been published in the journal PNAS.

Paper address: https://www.pnas.org/doi/epdf/10.1073/pnas.2123439119

1 AI marches into archeology

Traditional archaeological methods for identifying fire sources at early ancient human sites mainly rely on altered sediments , visual assessment of lithic debris and bones, such as soil reddening, discoloration, warping, cracking, shrinkage, darkening, etc., may underestimate the prevalence of human fire use at the time.

In this study, the author's team developed a spectral "thermometer" based on Raman spectroscopy and deep learning algorithms, used to estimate chert pseudo Thermal exposure of shadows detects extreme temperatures that distort the atomic structure of materials, thus compensating for the possible lack of visual signatures of traces of fire.

Research shows that the Early Paleolithic open-air site (Evron Quarry) in Israel contains the remains of fire-burned animals and rock debris, ranging from 1 million to 800,000 years ago. .

##Caption: From left to right: Filipe Natalio, Ido Azuri, Zane Stepka

The research team first studied material excavated at Evron Quarry in 1976-1977 and found no obvious visual evidence of heat-related features, such as soil reddening, flint tool discoloration or Cracking, shrinkage or discoloration of animal remains, etc.

Caption: Archaeological excavation site of Evron Quarry site

Team test Many methods are used, including traditional data analysis methods, machine learning modeling and more advanced deep learning models. Popular deep learning models have specific architectures that outperform other models. The benefit of using AI technology is that it can analyze the chemical composition of materials and use this to estimate their thermal exposure.

AI technology can reliably distinguish whether modern flint has been burned, and can also reveal the temperature at which it burned. The heat of a fire can cause changes in nearby stones, and burning can change bone structure at the atomic level, with corresponding changes in the infrared spectrum.

In this study, the team used a deep learning model (a one-dimensional convolutional neural network) to learn the Raman spectral patterns of flint artifacts to estimate the temperature of the stone tools. The model performed better than a fully connected artificial neural network (FC-ANN), being able to reduce the mean absolute error between true and estimated temperatures from 118 °C to 103 °C.

First, the team pre-trained on modern flint collected from different locations in Israel and heated to known temperatures under controlled laboratory conditions. Second, the trained model was applied to an unknown sample (i.e., stone tools collected from the Evron Quarry site). The team used a supervised deep learning approach to correlate Raman spectra with the heating temperature of the chert. This method relies on irreversible thermally induced structural changes in the organic and inorganic components of chert while overcoming its inherent variability. The advantage of using a deep learning model for temperature estimation is that it can approximate any nonlinear decision boundaries between heat and spectral changes due to heat in alpha-quartz, moganite, and D and G band spectral regions.

In the picture below, the stone does not visually show any traces of being burned by fire. By using a deep learning model, the ultraviolet Raman spectrum collected from the stone is The thermal exposure was estimated and found that they had been heated to between 200°C and 600°C. This suggests that ancient humans had the ability to control fire rather than just use natural wildfire.

2 Follow-up discussion

Regarding the excavated bones, the research team also experimentally confirmed that they had been burned by fire. Chazan, one of the authors, said: "Without the flint results verified by artificial intelligence, no one would bother to test the thermal exposure of these bones. Condition".

This study, however, cannot determine whether the tools at the site were burned by natural or artificial fire. Spatial changes caused by burn marks can be interpreted as evidence of human intervention, as natural fires often cause homogeneous thermal changes throughout the burned area.

The authors acknowledge that wildfires and patchy vegetation can also cause uneven temperature distribution across an area, and that temperature is not a reliable distinguishing criterion between using wildfires and artificial fires. But despite this, the estimated temperatures of Stone Age tools and the presence of burned fauna suggest the possibility that fire was used by ancient humans at the site.

In the future, the methods used in this study can be extended to other Upper Paleolithic sites, which will have the potential to expand our understanding of the relationship between early hominins and fire. Understand, opening a window into early human life.

The above is the detailed content of AI enters archeology! Scientists used deep learning algorithms to discover evidence of human use of fire nearly 1 million years ago, published in PNAS. For more information, please follow other related articles on the PHP Chinese website!