The use of artificial intelligence in medicine has long been a factor of science fiction. The notion that the body’s functioning can be understood in a standard fashion – that despite our individual natures, every human heart, hip or hand work on the same principals. It therefore is a tangible idea that symptoms of illness can be programatised. Skin cancer is one of the most common human malignancies (Andre Esteva et al 2017). It is primarily diagnosed visually by a dermatologist but in January 2017, computer scientists, dermatologists, and engineers at Stanford University published a study documenting the successful development and deployment of an algorithm to diagnose skin cancer.
Drawing on a database of 129,000 medical images of melanomas and carcinomas the algorithm is utilised deep convolutional neural networks, engineered to reflect the neural network patterns of the human brain to recognise signs and patterns. After training the algorithm on this database, the software was tested against 21 clinicians in three areas: keratinocyte carcinoma classification, melanoma classification and melanoma classification using dermoscopy and results were 91% consistent (Andre Esteva et al 2017).
Can the use of algorithms of this nature enable us to approach the body through an infrastructural lense? Previously, automated classification of this kind was not possible due to the highly fine-grained variability of individual skin (Kubota 2017). The use of an algorithm trained on hundreds of thousands of other examples shifts the focus from the individual person and melanoma or carcinoma to the augmented mass of skin cancer.
In their ethnography of information structure, Star and Ruhleder posit that ‘infrastructure is a fundamentally relational concept. It becomes infrastructure in relation to organized practices’ (Star & Ruhleder 1996).
If one understands infrastructure not as a ‘thing’ but as a way of relating, this diagnostic tool can be understood as infrastructural in two ways – firstly through its use of neural networks. ‘ANNs (Artificial Neural Network), like people, learn by example. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process. Learning in biological systems involves adjustments to the synaptic connections that exist between the neurones. This is true of ANNs as well (Stergiou et al. 2011)’. This demonstrates how the fundamental architecture of such an algorithm is relational, and therefore infrastructural.
The second way an algorithm of this nature could be understood as an infrastructure is through its use as a mobile application, which is an ambition of the engineers. If a technology of this kind were made available the ability to diagnose skin cancer without a medical professional could become ubiquitous. Star and Ruhleder draw on Yrjo Engestrom’s question “When Is a Tool?” (Engestrom 1990), which he answers with the suggestion that ‘a thing becomes a tool in practice, for someone, when connected to some particular activity’ (Star & Ruhleder 1996). Thus, the mobile application that can diagnose skin cancer can be seen as an infrastructure, created by the relations of practice represented by its use.
For anthropology, this presents an opportunity to understand disease in a totally new way, understanding the global through the individual use of such an infrastructure. The development and use of the algorithm provides an infrastructure of relation that has the potential to be understood by an infrastructural analytic.
Andre Esteva et al., 2017. Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), p.115.
Engestrom. Y, 1990. “When Is a Tool? Multiple Meanings of Artifacts in Human Activity,” in his Learning, Working and Imagining, Orienta Konsultit Oy, Helsinki
Kubota, T, 2017. Deep learning algorithm does as well as dermatologists in identifying skin cancer. Stanford News. http://news.stanford.edu/2017/01/25/artificial-intelligence-used-identify-skin-cancer/
Star. L.S, Ruhleder. K, 1996. Steps Toward an Ecology of Infrastructure: Design and Access for Large Information Spaces. Information Systems Research, 7(1), 111-134.
Stergiou. C, Siganos. D, 2011. Neural Networks. Imperial University. https://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html (accessed 19.04.17)
(Image credit: Matt Young)