X Close

Heritage Science and Engineering Network Blog

Home

Presenting the latest research and ideas in Heritage Science and Engineering

Menu

UCL Heritage Science and Engineering Network Blog

By aliciacolson, on 17 April 2020

Rock Art and AI

by

Dr Alicia Colson FSA, FRGS

 

Rock Art is another name for those troublesome elusive images, sometimes labelled pictographs, rock paintings, petroglyphs, engravings or rock paintings. The label shifts depending on the researcher. They are definitely in the domain of archaeology but rock image sites cannot be studied using the same techniques as might be applied to other archaeological sites. The theoretical approaches used and the questions asked may be the same but the data sources are not only radically different and but generally far more limited in number. These images cannot be excavated using the standard techniques for recovering, cataloguing, and analysing data archaeologists apply to materials in ‘conventional’ archaeological sites. The area surrounding such images may be excavated but the physical context of the site itself often provides little or no information about the meaning(s) of the images themselves. I propose that these representations should be termed rock images, or petroglyphs and pictographs (Colson 2007:8). This term is problematic (see Colson 2018: 80-82) but this is not the place for such discussion. A rich literature attests to the elusive search for meaning using a variety of theoretical frameworks (for: culture- history, contextual, analogical, homological and intuitive – see Colson 2019: 38-39). Supposition is as in any discipline, rampant. Suppositions, for example, about the physical location of the pictograph sites in the Canadian Shield, as always located on the vertical faces of granite cliff fundamentally influences their discovery, and how and by whom they are interpreted (see Brown and Brightman 1988: 55-56; Dewdney and Kidd 1962: 13-14; Kohl 1985 [1860]: 415; Rajnovich 1989: 185-186; Wheeler 1977a, 1977b, 1979).

Given the difficulties in obtaining absolute dates and finding those elusive meaning(s) the question exists can ‘machine vision’ provide a way in which these images can yield some of their ‘secrets’? Computer scientists have long considered this possibility. They have  developed and utilised algorithms for image retrieval for large databases of large collections of images from the archaeological record as well and applied machine learning (for example Wang et al. 2010; Wang et al. 2019; Zeppelzauer et al. 2016; Zhu et al. 2009). Zhu and his colleagues observed that computer scientists noticed that though image processing, information retrieval and consequent data mining has “had large impacts on many human endeavors, they have had essentially zero impact on the study of rock art” (Zhu et al. 2009:1057). Yet digital image processing, as it’s termed by archaeologists, is been considered an integral component for many in heritage and archaeology (see: Clogg, et. al. 2000; Domingo et al. 2013; Jaladoni et. al. con 2018; Tomaskova, Silvia 2015; Williams and Shee Twobig 2015).

Naturally the question stands of whose perception is under discussion here, where ‘seeing’ and ‘perception’ are crucial. In this case reference should be made to Berger’s book “Ways of Seeing’ (1973). Attributing them automatically implies they were made by someone, even might be owned by someone: person or corporate entity. Computing tools that utilize AI, which is a cover-all term, apparently offer a way forward not merely  to assist attribution, but to offer the  entrancing potential – of linking Rock Art firmly a large story, finding out what paintings ‘mean’, even debating the origins of human  species – so the stakes are high. It is hardly surprising that researchers in the rapidly-growing fields of digital humanities, digital heritage, digital archaeology, and humanities computing have pitched in with a range of tools AI tools, RTI (reflectance transformation imaging) (e.g. Diaz-Andreu et al. 2015), combining them with photogrammetry (e. g. Cerrillo Cuenca et. al. 2013; Cortón Noya et al. 2015) as well as image enhancing software such as D-Stretch (e.g. .Le Quellec et al. 2013, Le Quellec, et, al. 2015; Quesada & Harman 2019) and Adobe Photoshop (e.g. Brady 2006). All in the race to record such images in the hope of extracting something more definitive. It’s even worth checking Grau et al. 2017; Robin 2015 and Vincent et al. 2017.

There is a catch. For once recorded those pesky images, photographed in situ, previously seen only by archaeologists in laboratories, conferences or worse still in academic articles are now; irretrievably ‘out there’. Now archaeologists can’t hide from the questions. The public asks: How many images are there? What do they mean? What is their vocabulary? What’s the range of images used? Are there combinations of images? What combinations are common? These questions raise two challenges. Firstly, the tools which utilize AR, 3D printing, photogrammetry etc are mere lenses which permit us all to ‘see’ the images. They cannot tell us how to see them, still less how others see them. They do not involve ‘machine learning’. Scholars from other disciplines have considered how to describe images (see for example: (Jaritz and Schuh 1992; Sawyer 2010), some have even discussed computer vision (Brassard 1999). Secondly, since research on these images has often been funded by the general public, there is pressure to apply path-breaking AI tools successful in other endeavours to data which has been to hand for almost two decades (see Jaritz 1993; Martinez 1991; Martinez et al. 2002; Thaller 1991 & 2017).

But deploying AI (usually in machine learning) carries an ‘unacceptable risk’ for those involved– it is extremely time consuming. My decades of research and conversations with colleagues in computer science, indicate that relatively few researchers (archaeologists) have drawn on machine vision and applied it to a dataset of such images, even though the originals were all carefully, and officially, recorded some time ago. It appears that researchers suspect that the results of such an exercise might turn out to be negative – that there is no vocabulary of images: that the images themselves have little in common and comparisons not easily made which would allow for a comparison between the Rock Art in (say) Northern Ontario and that found in Norway or Sweden. Too perilous by half for junior researchers, and not the stuff others wish to hear.

That negative result is especially irritating if the work is going in other directions, and emphasizes the uniqueness of each painting, its local setting rather than larger continuities. So relatively few researchers from archaeology, digital humanities, heritage utilise AI on a large scale, undertake complex pattern matching studies, let alone experiment with machine learning. Pseudo accountability of the research funding organizations, skews against the potentially forbidding outcome of a negative result making the prospect ever more risky for potential researchers.

But Rock Art continues to fascinate the general public and, the tantalizing notion that these apparently ancient images might just throw some light on that age-old quest for the origins of humans as ‘thinking’ species continues to nag (for example: (Power, Finnegan, and Callan 2017). The saga continues – in 2002, Henshilwood, an archaeologist, and his team discovered images (on the surface of the rocks) at Blombos, an archaeological site in South Africa (Henshilwood et al. 2011). They argued the images were created by peoples who had relatively ‘modern’ cognition and behaviour. He is quoted to as arguing in ‘Nature’, “that 77,000-year-old etchings were examples of symbolic behaviour and represented the earliest known evidence of abstract thought[.]” (Tollefson 2012). But can he really say that? Are there any other visual examples in i.e. rock art, of abstract thought and what do these images really ‘mean’? The appearance ochre pigment (ochre) in the archaeological record has consequences in the debate of human origins especially if it suddenly exists, becomes continuous, irregular and then regular. It can cause some to speculate that this is possibly culturally significant.

If they are so powerful why are some AI tools not being used to consider the existence of a ‘vocabulary’ of images, and detect their potential range? It’s a worthwhile endeavour (Colson 2006 & 2011). But let’s push the boundaries even further. Why are some tools not being utilised to enable us to envisage combinations of images from the evidence provided by many rock art sites? Since archaeologists search for continuities the use of AI would be the natural approach.

Attempts to breakdown disciplinary boundaries will always burst the intellectual bubbles, which might be getting in the way. These attempts are the essence of scholarly endeavour. They exist anywhere and are everywhere. The Emperor may well be wearing nothing but fantasy, Oz had to be found out. Such activity might enable technologies still under development to surface, to get traction. Why not tackle that humanities equivalent to a Grand Challenge – when did people, human beings, actually begin to conceptualize? So could ‘machine vision’ serve as a way to meet this Grandest of Challenges? The task could not be more serious. It appears that those fields of digital humanities, digital heritage, and digital archaeology, and humanities computing involved in this particular challenge appear to define machine vision differently from the computer scientists who first developed and first deployed it. While some practitioners appeared to view machine vision as assisting them to find and provide meaning to the user, the viewer (the general public) and the researcher, computing scientists may well recoil. They know that they are being asked the impossible. Since questions of ‘meaning’ cannot be in the design specification, they cannot be tackled by this route. AI agents are trained to perform many different tasks (some in simulations and some in the real-world). But as yet, by themselves they cannot provide meaning(s). They will ‘find’ null ‘matches’ they will produce those very ‘negative’ results. Unfortunately the traditional perspective of a researcher in the humanities and social sciences who is concerned to isolate that ‘moment’ of ‘conceptual thought’ will find such results at the very least to be profoundly  problematic.

As an archaeologist and ethnohistorian who’s used computing science and intelligent agents to assess the composition and features of rock paintings I simply narrow the discussion to those images which are encountered on the surface of the rock – which are not those of my imaginings.

 

Cited References

Barnett, T., A. Chalmers, M. Díaz-Andreu, et al. (2004) Breaking Through Rock Art Recording: Three Dimensional Laser Scanning of Prehistoric Rock Art. In . K. Cain, Y. Chrysanthou, Franco Niccolucci, and N. Silberman, eds. Pp. 70–71. Goslar, Germany: Eurographics Association.

Berger, John (1973) Ways of Seeing. London: Penguin Books Ltd.

Brady, Liam M (2006) Documenting and Analyzing Rock Paintings from Torres Strait, NE Australia, with Digital Photography and Computer Image Enhancement. Journal of Field Archaeology 31(4): 363–379.

Brassard, Louis (1999) The Perception of the Image World. PhD, Simon Fraser University: Vancouver, Canada

Brown, Jennifer S. H., and Robert Brightman (1988) The Orders of the Dreamed”: George Nelson on the Cree and the Northern Ojibwa Religion and Myth, 1823. Winnipeg: University of Manitoba Press.

Cerrillo Cuenca, Enrique, Pedro Ortiz-Coder, and Jose Angel Martinez del Pozo (2013) Computer Vision Methods and Rock Art: Towards a Digital Detection of Pigments. Archaeological and Anthropological Sciences 6 (3).

Cortón Noya, Natalia, Ángela López García, and Fernando Carrera-Ramírez (2015) Combining Photogrammetry and Photographic Enhancement Techniques for the Recording of Megalithic Art in North-West Iberia. Digital Applications in Archaeology and Cultural Heritage 2(2–3): 89–101.

Colson, Alicia J. M. (2018) The Perpetual Dilemma of a Pictograph Site. VISUAL PAST A Journal for the Study of Past Visual Cultures 5: Special Issue “Bildsinne/Image Senses”: 75—98.

Colson, Alicia J. M. (2011) Do Not Make Snap Decisions about What You Are Seeing: How Digital Analysis of the Images from the Canadian Shield Highlights the Difficulties in Classifying Shapes. International Journal of Humanities and Arts Computing 5(1): 59–81.

Colson, Alicia J. M. (2006) An Obsession with Meaning: A Critical Examination of the Pictograph Sites of the Lake of the Woods. PhD, McGill University: Montreal, Canada

Clogg, Phil, Margarita Diaz-Andreu, and Brian Larkman (2000) Digital Image Processing and the Recording of Rock Art. Journal of Archaeological Science 27: 837–843.

Dewdney, Selwyn, and Kenneth E. Kidd (1962) Indian Paintings of the Great Lakes. 1st Edition. Toronto: University of Toronto Press.

Diaz-Guardamino, Marta, Leonardo Garcia Sanjuan, David Wheatley, and Victor Rodriguez Zamora (2015) RTI and the Study of Engraved Rock Art: A Re-Examination of the Iberian South-Western Stelae of Setefilla and Almadén de La Plata 2 (Seville, Spain). Journal of Field Archaeology 2 Special issue on Digital imaging techniques for the study of prehistoric rock art edited by Guillaume Robin(2–3): 41–54.

Domingo, Ines, Valentín Villaverde, Esther López-Montalvo, José Luis Lerma, and Miriam Cabrelles (2013) Latest Developments in Rock Art Recording: Towards an Integral Documentation of Levantine Rock Art Sites Combining 2D and 3D Recording Techniques. Journal of Archaeological Science 40: 1879–1889.

Henshilwood, Christopher S., Francesco d’Errico, Karen L. van Niekerk, Yvan Conquinot, Zenobia Jacobs, Stein-Erik Lauritzen, Michel Menu, Renata Garcia-Moreno (2011) A 100,000-Year-Old Ochre-Processing Workshop at Blombos Cave, South Africa. Science 334(6053): 219–222.

Jaladoni, A, I. Domingo, and Paul S. C. Tacon (2018) Testing the Value of Low-Cost Structure-from-Motion (SfM) Photogrammetry for Metric and Visual Analysis of Rock Art. Journal of Archaeological Science 17: 605–616.

Jaritz, G., and B. Schuh (1992) Describing the Indescribable. In Images and Manuscripts in Historical Computing Pp. 143–153. Halbgraue Reihe Zur Historischen Fachinformatik, Serie A: Historische Quellenkunden. St Katherinen: Max-Planck-Institut für Geschichte: In Kommission bei Scripta Mercaturae Verlag.

Kohl, Johann Georg (1985) Kitchi-Gami Life among the Lake Superior Ojibway. Originally published in 1860 London: Chapman & Hall. St Paul: Minnesota Historical Society.

Leymarie, Frederic F., David B. Cooper, Martha Sharp Joukowsky, Benjamin B. Kimia, David H. Laidlow, David Mumford, Eileen L. Vote (2000) The SHAPE Lab: New Technology and Software for Archaeologists. Providence, Rhode Island: The Shape Lab, Brown University.

Le Quellec, Jean-Loïc, Jon Harman, Claudia Defrasne, and Frederique Duquesnoy (2013) DStretch® et l’amélioration Des Images Numériques: Applications à l’archéologie Des Images Rupestres. Les Cahiers de l’AARS 16: 177–198.

Le Quellec, Jean-Loïc, Frederique Duquesnoy, and Claudia Defrasne (2015) Digital Image Enhancement with DStretch: Is Complexity Always Necessary for Efficiency? Digital Applications in Archaeology and Cultural Heritage 2 Special Issue: Digital Imagining techniques for the study of Rock Art (2–3): 55–67.

Martinez, Kirk (1991) High Resolution Digital Imaging of Paintings: The Vasari Project. Microcomputers for Information Management 8(4): 277–283.

Power, Camilla, Morna Finnegan, and Hilary Callan, eds. (2017) Human Origins: Contributions from Social Anthropology, vol.30. Methodology and History in Anthropology. New York and Oxford: Berghahn.

Rajnovich, Grace (1989) Visions in the Quest for Medicine: An Interpretation of the Indian Pictographs of the Canadian Shield. Midcontinental Journal of Archaeology 14(2): 179–225.

Robin, Guillaume (2015) Editorial. Digital Applications in Archaeology and Cultural Heritage 2: 35–40.

Sawyer, Andrew (2010) Catching up with the Past: Historians, Pictures and PCs. History and Computing 10(1–3): 120–123.

Seidl, Markus, Ewald Wieser, and Craig Alexander (2105) Automated Classification of Petroglyphs. Digital Applications in Archaeology and Cultural Heritage 2(2–3): 196–212.

Tollefson, Jeff (2012) Cultural Roots. Nature 482: 290–292.

Tomaskova, Silvia (2015) Digital Technologies in Context: Prehistoric Engravings in the Northern Cape, South Africa. Digital Applications in Archaeology and Cultural Heritage (2): 222–232.

Vincent, Matthew L., Victor Manuel Lopez-Menchero Bendicho, Marinos Ioannides, and Thomes E. Levy (2017) Heritage and Archaeology in the Digital Age: Acquisition, Curation, and Dissemination of Spatial Cultural Heritage Data. Springer.

Wang, Haiyan, Zhongshi He, Dingding Chen, Yongwen Huang, and Yiman He (2019) Virtual Inpainting for Dazu Rock Carvings Based on a Sample Dataset. ACM Journal on Computing and Cultural Heritage 12(3): 20:1-20:17.

Wang, James Z, Ge Weina, Dean R. Snow, Prasenjit Mitra, and C. Lee Giles (2010) Determining the Sexual Identities of Prehistoric Cave Artists Using Digitized Handprints: A Machine Learning Approach. In Proceedings of the ACM Multimedia 2010 International Conference. Firenze, Italy: DBLP.

Wheeler, Clint (1977a) The Historic Assiniboine: A Territorial Dispute in the Ethnohistoric Literature. In Actes Du Huitième Congrès Des Algonquinistes. William Cowan, ed. Pp. 115–123. Ottawa: Carleton University.

Wheeler, Clint (1977b) The Oxford House Pictograph or May May Quah Sao Are Alive and Well at Oxford House. In Proceedings of the Second Congress. Jim Freedman and Jerome H. Barkow, eds. Pp. 701–713. Canadian Ethnology Service, National Museum of Man, Mercury Series. Ottawa: National Museums of Canada.

Wheeler, Clint (1979) The Oxford House Pictograph. Canadian Rock Art Research Association Newsletter: 2–9.

Williams, Ken, and Elizabeth Shee Twobig (2015) From Sketchbook to Structure from Motion: Recording Prehistoric Carvings in Ireland. Digital Applications in Archaeology and Cultural Heritage 2(2–3): 120–131.

Zeppelzauer, Matthias, Georg Poier, Markus Seidl, Christian Reinbacheret, Samuel Schulter, and Horst Bischof (2016) Interactive 3D Segmentation of Rock-Art by Enhanced Depth Maps and Gradient Preserving Regularization. Journal on Computing and Cultural Heritage 9 (4): 19:1 -19:30

Zhu, Qiang, Xiaoyue Wang, Eamonn Keogh, and San-Hee Lee (2009) Augmenting the Generalized Hough Transform to Enable the Mining of Petroglyphs. In KDD’09 Proceedings of the 15th ACM SIGKDD Internation

 

Dr Alicia Colson FSA, FRGS is an archaeologist and an ethnohistorian with a PhD (McGill) and BA (Hons.) Institute of Archaeology, UCL. She’s a Fellow of the Royal Geographical Society, Society of Antiquaries and of the Explorers Club. She’s a Patron of the Great Britain and Ireland Chapter of the Explorers Club. She trained in academic, government, and governmental organizations in the non-profit sector in the UK, Canada and the US. She is fluent in four languages and has undertaken extensive fieldwork in Canada, the UK, US, and Antigua. She has worked as an archaeologist on rock art sites, habitation sites, human remains as well as burial mounds. She’s worked for the British Exploring Society several times in Iceland, Namibia and SV Tenacious (a modern wooden sail training ship). Her research interests include: hunter-gatherers of the Boreal Forest, digital humanities, archaeological theory, history of archaeology, and sub-Saharan Africa. She is currently developing projects in various places globall with colleagues and is engaged in series of research and publishing projects in cognate fields.

Photo of Dr Alicia Colson

Dr Alicia Colson London, UK