Hiking with Machine’s Eyes: A Computer Vision Exploration of Nature Photography in Instagram
Keywords: nature imagery, computer vision, machine learning, ethnography, social media studies, algorithmic cultures, cultural representation, visual culture
TL;DR: Combining computer vision with ethnographic insights, this paper examines how Finnish nature sites are represented in Instagram, highlighting the complex interplay between digital cultural practices and machine learning.
Abstract: This paper explores the methodological and epistemological implications of using computer vision to analyse visual representations of Finnish recreational nature sites in social media. The study focuses on the Instagram imagery of two nature sites in Finland, while also being informed by ethnographic walking interviews that focus on how the uses of digital media transform the representations and experiences of nature and its fragility. By combining and contrasting machine learning techniques with qualitative inquiry (cf. Maltezos et al., 2024), the study aims at making sense of how the complex interplay between algorithmic visual cultures and the quotidian uses of technology shapes our environmental relations.
Through scraping the Instagram API with a hashtag-based approach, a large dataset of images was collected about the two fieldwork sites: Patvinsuo National Park in Lieksa, Finland, and Viiankiaapa Mire Reserve in Sodankylä, Finland. The images were analysed using Google's Inception v3 API for image embeddings and further through unsupervised machine learning methods (hierarchical clustering, principal component analysis) in Orange data mining platform. These methods facilitated constructing a visual taxonomy of nature representations as well as a set of dichotomous factors that supposedly describe a part of the dataset's variance as captured by the embedding algorithm. This visual taxonomy highlights AI's proficiency in object detection, while the categorisations of landscapes images were harder to interpret in a cultural context. Notably, the process of hierarchical clustering creates pairings of which some are predictable but others very unexpected ("selfies and canoes"), challenging us to consider the embedded values, assumptions, and often invisible data labour that shape AI's understanding (cf. Denton et al., 2021; Carah et al., 2022).
The study asserts that the proliferation of digital photography on social media in combination with ethnographic approaches provides a rich basis for exploring how boundaries between virtual and on-site nature are currently being blurred, and how this entanglement transforms our relations with nature. Algorithms not only categorise but co-create our visual digital cultures, and thus there is a need to critically assess their underlying tendencies and biases. The research also underscores the AI's methodological limitations in visual content analysis. While AI offers an efficient method to manage and categorise large image datasets, the interpretative nuance of human analysis remains essential, particularly for contextually rich images. A mixed-method approach can thus yield a more holistic understanding of nature's digital representations.
References:
Carah, N., Angus, D., & Burgess, J. (2022). Tuning machines: an approach to exploring how Instagram’s machine vision operates on and through digital media’s participatory visual cultures. Cultural Studies, 36(3), 456-478. https://doi.org/10.1080/09502386.2022.2042578
Denton, E., Hanna, A., Amironesei, R., Smart, A., & Nicole, H. (2021). On the genealogy of machine learning datasets: A critical history of ImageNet. Big Data & Society, 8(1), 1-15. https://doi.org/10.1177/20539517211035955
Maltezos, V., Luhtakallio, E., & Meriluoto, T. (2024). Bridging ethnography and AI: a reciprocal methodology for studying visual political action. International Journal of Social Research Methodology, 27(2), 234-249. https://doi.org/10.1080/13645579.2024.2330057
Submission Number: 63
Loading