Scientific Archaelogy - Artificial Intelligence

We are interested in developing and implementing artificial intelligence algorithms to retrieve hidden information from archaeological contexts. These include:

1. Fire use by hominins. Hominins’ controlled use of fire is one of the most revolutionary transformations in cultural and biological evolution of the genus Homo. It provided our ancestors with heat, light, protection against predators, the ability to manipulate food products, both plants and meat, so as to ease their digestion and increase their nutritional value. The controlled used of fire, which differs from opportunistic use of naturally occurring fire, is associated with a specific set of cognitive skills and advanced behaviors, such as planning, collecting raw materials, fire handling, cooking and fabricating sharper stone tools. Investigating the origin of controlled use of fire has, however, been impeded by the fact that organic remains from fire activities do not survive for long periods of time. The identification of fire in Lower Paleolithic (LP) sites relies primarily on an initial visual assessment of artifacts' physical alterations, resulting in an underestimation of its prevalence in the archaeological record.

We have explored the organic and inorganic composition of flint/chert (1). The intercrystalline amorphous carbon material orients toward aligned stacking (through pi bonding) according to heating. Also, α-quartz changes the crystal structure as function of temperature, i.e., alpha to beta transitions at lower temperatures, or other polymorphs at higher temperatures. We have created a UV-Raman database using flint collected from different regions across Israel heated at different temperatures. We then, created an AI model, capable to estimate the temperatures to which the flint stone tools were heated. We applied this methodology to Qesem Cave (Israel) and discovered a differential heating (behavior) between flakes and blades (2) (Figure 1).

Figure 1. Schematic workflow.

 

Collaborators:

Dr. Liora Kolska Horwitz - National Natural History Collections, The Hebrew University, Israel

Prof. Michael Chazan - Department of Anthropology, University of Toronto, Canada

 

References:

1. Natalio, F., Corrales, T.P., Pierantoni, M., Rosenhek-Goldian, I., Cernescu, A., Raguin, E., Maria, R. and Cohen, S.R., 2021. Characterization of Eocene flint. Chemical Geology, 582, p.120427.

2. Agam, A., Azuri, I., Pinkas, I., Gopher, A. and Natalio, F., 2021. Estimating temperatures of heated Lower Palaeolithic flint artefacts. Nature Human Behaviour, 5(2), pp.221-228.

2. Technological evolution based on material culture. Deep learning is a powerful tool for exploring large datasets and discovering new patterns. Our initial work presented an account of a metric learning-based deep convolutional neural network (CNN) applied to an archaeological dataset (3). We trained a AI model (CNN) with several thousand artefact images, ranging from the Lower Palaeolithic period (1.4 million years ago) to the Late Islamic period (fourteenth century AD), were used to train the model (i.e., ) to discern artefacts by site and period. With this AI trained model, we constructed a website where researchers can upload their images and have the 5 closest hits (https://github.com/aviresler/antique-gen) based on our databse.

Also, we used this trained model to discern meaningful connections using Levantine Natufian artefacts as a case-study (3). Following up, we focus our current research to design toy-models by combining AI and network analysis graphs toward understanding technological evolution, transmission and dispersal of artififacts throughout time.

Collaborators:

Prof. Raja Gyres - School of Electrical Engineering, Tel Aviv University, Israel

 

References:

3. Resler, A., Yeshurun, R., Natalio, F. and Giryes, R., 2021. A deep-learning model for predictive archaeology and archaeological community detection. Humanities and Social Sciences Communications, 8(1), pp.1-10.