The use of fire is a key factor in the evolution of Homo sapiens. Fire can not only be used to create more complex tools, but also make food safer, thus contributing to the development of the brain. So far, only five sites with fire evidence dating back to 500000 years have been found worldwide, including wonderwerk cave and Swartkrans in South Africa, chesowanja in Kenya, Gesher Benot Ya'aqov in Israel and Cueva Negra in Spain. Now, an Israeli research team has found the sixth site that shows the traces of human fire using artificial intelligence algorithms! The study revealed evidence of human use of fire at a late Paleolithic site in Israel. The research results have been published in PNAS journal. Thesis address: https://www.pnas.org/doi / epdf / 10.1073 / pnas. two billion one hundred and twenty-three million four hundred and thirty-nine thousand one hundred and nineteen AI enters Archaeology The identification of fire sources used in early ancient human sites by traditional archaeological methods mainly depends on the visual evaluation of altered sediments, rock cuttings and bones, such as soil redness, discoloration, warpage, cracking, shrinkage, darkening, etc., which may underestimate the prevalence of human fire at that time. In this study, the author's team developed a spectral "thermometer" based on Raman spectroscopy and deep learning algorithm to estimate the thermal exposure of chert artifacts and detect the atomic structure of extremely high-temperature distorted materials, thus making up for the possible lack of visual features of fire traces. Research shows that there are remains of burned animals and rock debris in the early Paleolithic open-air site (evron quarry) in Israel, which is between 1million and 800000 years ago. (Note: Filipe natalio, Ido azuri and Zane stepka are from left to right) The research team first studied the materials excavated in evron quarry from 1976 to 1977, and found no obvious visual evidence of heat related features, such as red soil, discoloration or cracking of flint tools, shrinkage or discoloration of animal remains. (Note: archaeological excavation site of evron quarry site) The team tested many methods, including traditional data analysis methods, machine learning modeling and more advanced deep learning models. The popular deep learning model has a specific architecture superior to other models. The advantage of using AI technology is that it can analyze the chemical composition of materials and estimate their heat exposure. AI technology can reliably distinguish whether modern flint has been burned or not, and can also reveal its combustion temperature. The heat of fire can cause changes in nearby rocks. Combustion will change the bone structure at the atomic level, and the corresponding infrared spectrum will also change. In this study, the team used a deep learning model (one-dimensional convolutional neural network) to learn the Raman spectral patterns of chert artifacts, so as to estimate the temperature of stone tools. Compared with the fully connected artificial neural network (fc-ann), the model has better performance, and can reduce the average absolute error between the real temperature and the estimated temperature from 118? ° C down to 103? °C。 First, the team pre trained modern cherts collected from different parts of Israel and heated them to a known temperature under laboratory controlled conditions. Secondly, the trained model is applied to unknown samples (i.e. stone tools collected from the evron quarry site). The team used a supervised deep learning method to correlate Raman spectra with the heating temperature of chert. This method relies on irreversible thermally induced structural changes in the organic and inorganic components of chert and overcomes its inherent variability. The advantage of using deep learning model for temperature estimation is that it can approximate heat and α- Any non-linear decision boundary between quartz, moganite, and spectral changes due to heat in the D - and G-band spectral regions. In the figure below, the stones can't see any burning traces visually. However, by using the depth learning model to estimate the thermal exposure of the UV Raman spectra collected from the stones, it is found that they have 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 wildfires. Follow up discussion For the excavated bones, the research team also confirmed through experiments that they had been burned by fire. Chazan, one of the authors, said: "if there is no flint result verified by artificial intelligence, no one will bother to test the heat exposure of these bones.". Although the study is not yet able to determine whether the tools at the site were burned by natural fire or artificial fire. The spatial changes caused by burning traces can be explained as evidence of human intervention, because natural fires usually lead to homogeneous thermal changes in the whole burning area. The authors acknowledge that wildfires and uneven vegetation may also lead to uneven temperature distribution throughout the region, and that temperature is not a reliable criterion for distinguishing between wildfires and artificial fires. Nevertheless, the estimated temperature of stone age instruments and the existence of burned fauna can still indicate the possibility that ancient humans at the site used fire. In the future, the method used in this study can be extended to other sites in the late Paleolithic age, which may expand people's understanding of the space-time relationship between early ancient humans and fire, and open a window to understand early human life. (Xinhua News Agency)