Artificial Intelligence meets extinct carnivores

A paper published in Science Bulletin, a multidisciplinary academic journal supervised by the Chinese Academy of Sciences and co-sponsored by the CAS and the National Natural Science Foundation of China, shows that Artificial Intelligence is a highly effective tool to know which species of extinct carnivores acted on the fossil bones found at paleontological sites. This research has focused on the site of Venta Micena (Orce, Granada, Spain).

Why is Orce relevant?

The Archaeological Zone known as 'Cuenca de Orce' is located in the southeast of the Iberian Peninsula, in the province of Granada (Spain). Orce is included in the Guadix-Baza basin, one of the areas with the greatest archaeopalentological potential of the Early Pleistocene together with the Nihewan basin (China). Research carried out since 1982 has yielded a large record of large vertebrates from around 1.5 Ma ago. Among the twenty sites with vertebrate assemblages, three stand out: Venta Micena, the oldest (1.6 Ma) with no human presence, Barranco León (1.46 Ma) where a deciduous human tooth has been found together with a large amount of knapped lithic tools, and Fuente Nueva 3 (1.2 Ma) where human tools are also very abundant.

The fauna of Orce

Among the species found in Orce, and in particular in Venta Micena, we can highlight the following herbivores taxa: a proboscidean, a hippopotamus, a rhino, a horse, two cervids, six bovids. Our protagonists, the carnivores, include two saber-toothed cats (Homotherium latidens and Megantereon cultidrens/whitei), a jaguar ancestor (Panthera cf. gombaszoegensis) and a lynx ancestor (Lynx sp.), three species of canids (a fox ancestor [Vulpes alopecoides], a Lycaon relative [Xenocyon lycaonoides] and a wolf ancestor [Canis mosbachensis], an extinct bear [Ursus etruscus] and our protagonist, the extinct hyena Pachycrocuta brevirostris. This mammal assemblage is representative of the Eurasian Early Pleistocene.

The Pleistocene landscape of Orce

The Orce landscape during the Early Pleistocene was very different from the present one. First, a large saline lake dominated the northeastern part of the Guadix-Baza basin. However, periods of lake regression, which led to the appearance of multiple groundwater lagoons, ponds and pools, were fundamental for attracting large vertebrates to the site.  Venta Micena was one of those freshwater ponds 1.6 Ma ago. This is where the rich fauna that inhabited this area came to quench their thirst and hunger, making Venta Micena a place where a game of life and death took place. The vegetation was also very different, with the presence of larger stands of trees in a wetter Mediterranean climate.

More than meets the eye

One of the characteristics of certain carnivores is that during the process of consuming the carcasses on which they feed, they accidentally leave tooth marks. These will be more present the more the carnivores try to finish off the meat present or if they attempt to fracture the bones to obtain the bone marrow. The problem has traditionally been how to distinguish which carnivore has bitten which bone.

“To solve this problem the first thing to do is to create a database of tooth marks from current species. in the words of Lloyd Courtenay (University of Salamanca), “one of lead authors of this research”, Specifically, of those species that have extinct analogues. In our case, lions, wild dogs, foxes, wolves, leopards, jaguars and hyenas. However, despite the obvious differences between the species mentioned, it is difficult to distinguish visually to which species each of them belongs”.

The novelty of our study is our ability to characterize for the first time the tooth marks of an extinct carnivore using a robust computational protocol based on 3D Geometric Morphometrics and Artificial Intelligence. The first approach, 3D Geometric Morphometrics is based on locating points in the same position for all tooth marks. This makes it possible to characterize the shape of each and every one of them. “Where the eye cannot reach, morphometric characterization arrives”, explains José Yravedra (Complutense University of Madrid) and another of the main authors of this paper. The second one, particularly the Deep Learning, focuses on teaching computers to learn. In our case, to distinguish between the tooth marks of current species and to find which of them most resemble those found at the Venta Micena site. “Machine Learning involves complex mathematical calculations and a high degree of programming knowledge”, says Courtenay. “But big claims require Big Data. And this cutting-edge technology”, notes Juan Manuel Jiménez-Arenas, director of the Orce Research Project. Indeed, the use of these methodologies yields very high match sucess  rates of over 90%.

Tooth marks, the barcode to carnivores of the past

The use of these technologies allows us to open a new window to the past. 3D Geometric Morphometrics would function as the barcode that we find on many products in any supermarket and Artificial Intelligence as the barcode reader with which to recognize which carnivores intervened in the sites.

“As an example, we have used Venta Micena, a site whose main agent that generated the great accumulation of bones present in this place was the great hyena Pachycrocuta brevirostris. And our analysis confirms this because most of tooth marks belongs to that species. But what will be really interesting will be to apply it to other sites where the protagonists are not so clear, opening a window to interpret the behaviour of extinct species”, claims Yravedra. “Our challenge is to apply it to sites where humans and carnivores competed for access to herbivore carcasses, such as Barranco León and Fuente Nueva 3, both in Orce”, confirms Jiménez-Arenas.

This study has been possible thanks to the collaboration of researchers from the Complutense University of Madrid (Spain), University of Salamanca (Spain), University of Granada (Spain) and University of Oxford (United Kingdom). This research has been possible thanks to the support and authorization of the Regional Government of Andalusia (Junta de Andalucía) (Spain).

---

---