From a Technical Point of View: The Intersection of Data Science and Anthropology

Upon joining my current company, Porini, almost two years ago, I was an enthusiastic young anthropologist with an academic background but limited work experience. After graduating I worked for a few companies in different sectors and different roles, always wishing to enter into the IT world. I was finally asked to work at Porini, an IT consulting company, which is the Microsoft competence centre of the DGS Group. I joined as a Project Manager and immediately oversaw projects in the Human Resources area, with a specific focus on the Microsoft Viva platform⁴ (Microsoft 2021a).

When I first learned about Viva, I was enthusiastic because it became increasingly clear that tools like this are crucial for improving employees’ wellbeing and ways of working in the modern workplace. These platforms analyse the values of an organisation through data and insights. Since data has become the way to read employees’ needs, wellbeing, and work styles, it is also essential to understand how to interpret that data, not only quantitatively but also in its compositional, multi-faceted, and heterogeneous aspects, which are shaped by human practices and culture. The potential of this technology immediately seemed immense to me in terms of developing a hybrid, socio-technical approach, but I was not immediately able to find a channel to bring the IT world into dialogue with anthropology.

Initially, the approach with the IT world and my colleagues made me feel disoriented, because I was already projected in ‘analysing with them how data science was a sociocultural object, both in its artifacts and its practices’ (Dourish and Bell 2011: 47), when even before we reached this depth of analysis there was a much more urgent and obvious difficulty to deal with: I did not have a computing background.

At first, from every perspective—technical, commercial, and so on—the computing world and language sounded exactly like Indonesian when I went to Indonesia for a fieldwork experience: an incomprehensible set of sounds, something far from my set of ‘meanings, values, beliefs, languages, and social practices that I use it to give meaning to my experiences’ (Geertz 1973) which are difficult to comprehend.

The same was true for the practices related to this language. As my colleagues know well, I didn't even know what a JSON⁵ file was when I first joined the company. It felt as if I was facing a different culture within my own culture: while my colleagues and I spoke the same language, it was as if we were operating on different epistemic foundations.

In order to begin a dialogue so that we could understand each other's approaches I had to relearn, not just a new language but also an entire world and the practices associated with it. This was not just a matter of learning how to translate technical details but also how those technicalities shaped that world and how they fit into the context of the company and the IT consulting industry as a whole (Bell 2014; Briggs 2017; Moeran 2005; Papa 1999). As time went by, and with my progressive ‘impregnation’ (De Sardan 2008) of the field, I have become increasingly more aware of the reality I was in. From their side, my colleagues, after their initial surprise and curiosity about having an anthropologist in the company, began to go beyond the anecdotes I was telling and became interested in my approach.

One day, as I walked past a data scientist's desk, I noticed a colourful, complex image displayed on her computer screen. Curiosity got the better of me, so I inquired about her work. I learned that the image was being processed by an algorithm for analysis, and she was attempting to decode its patterns and data, its technical protocols (Dourish and Bell 2011: 50). The mere idea that my colleague was trying to understand how algorithms think was illuminating to me. This was because I had found a significant overlap in the methodologies and working practices of both anthropologists and data scientists and, until then, I had not yet come across such an immediately concrete finding, despite its simplicity. What made the difference in this case was that she told me about something I was already familiar with: behaviour analysis (Countee 2015; Latour 2005). She spoke to me in a familiar language. Not only was I getting used to the way in which my colleagues see the world, through normal socialisation, but now I also had a common ground that gave the computing language a different significance. Watching my colleague try to understand the behaviour of algorithms conveyed by her technical protocols made me realise that I was approaching her, my colleagues, and the technologies I saw every day with my technical protocols.

Me and my colleagues simply belonged to two different groups of technicians who had distinct methodological backgrounds and languages which did the same thing and had the same goals. In my colleagues’ and my work, we encounter data and talk about data all day, every day and the growing synergy between our approaches and languages allows us to build together an innovative approach to both using and improving the technologies we use, develop, and sell every day.

In the following paragraphs I will detail two concrete examples of socio-technical approaches from daily work in contact with data scientists. The first example is about how we built standard packages for contextualised data analysis using Viva Insights technology, while the second is about how we mixed quantitative and qualitative approaches for creating a people-centred medical bot.

Viva Insights Standard Packages for Contextualised Analysis of Data

I mentioned at the beginning of the article that my first approaches to Microsoft technologies in Porini were with the Microsoft Viva suite. It is within the context of this technology that the collaboration between me and my colleagues resulted in a product.

Microsoft Viva can be considered a technology that falls under the human resources umbrella. In daily conversations with clients regarding Human Resources needs, there are themes that arose frequently, such as an overwhelming amount of data, dynamic and ever-changing work contexts, talent retention, quiet quitting, training, employee wellbeing, employee engagement, onboarding, data aggregation from different sources, data privacy, person-centredness, burnout, and too many meetings, among others. All these needs are all part of a cultural transformation in work habits known as the New Normal (Gullikseen et al. 2022).

To address them, the Insights module of the Microsoft Viva Suite can be used as a tool. Below is a brief overview of the features of this Viva platform module, which is useful in explaining the benefits of the multidisciplinary approach I advocate in this article.

Viva Insights is designed to identify patterns of collaboration among members of the corporate population (Microsoft 2021b). This identification has multiple objectives including determining how collaboration models impact workforce productivity and effectiveness, employee engagement, and employee wellbeing, understanding employee engagement, and analysing engagement data.

To achieve these objectives, Viva Insights collects statistics and data⁶ that are produced during the workday through multiple Office 365⁷ tools. This information is processed and presented in the form of a dashboard, which allows for the creation of recommendations to support members from various groups within the organisation, to make the most effective use of their working time (Microsoft 2021b).

Specifically, Viva Insights allows you to identify if there are business units within a company that have an excessive number of working hours or if employees are working too much after hours. Collecting this data is a fundamental starting point for taking concrete actions. For example, when we worked with a client based in France, he informed us that the right to disconnect has become part of the law, and companies with employees who work overtime can now be fined. Having data on post-work collaboration hours can lead to corrective actions in such cases. They therefore needed a tool that could provide evidence of the existence of these hours so that they could implement interventions and understand what the extra hours could be due to and whether they could affect employee wellbeing. Cases like this give evidence of how in the relationship between technology and society, social phenomena do not take place independently and in layers on a predefined technological substrate (Kling 1994).

Data collected by tools like Viva Insights provides very important indicators, which allow for the detection of issues such as those that emerged from the dialogue with the French client, but they do not offer an understanding of the motivations behind these quantitative measures. It is possible to see that a business unit has many hours of meetings, though I am unable to understand why this is the case. I often have ‘big data,’ but not ‘thick data’ (Cioffi-Rivella 2010; Geertz 1973). This is where an anthropological approach, in collaboration with a data analysis methodology, can make an impact. In this way integrating the nature of big data to discover insights by isolating variables to identify patterns, and the nature of thick data to take smaller samples to unearth human-centred patterns in depth, allows for building a more holistic view of the data.

While my colleagues and I have found that the needs are similar, we have also found that the circumstances of each customer are unique. When we talk to clients and explain this technology, we always emphasise that the data they see must be related to their specific context and possibly be explored through internal investigations aimed at understanding the reasons behind the data. The idea supporting this process is that data should be considered as unstable objects (Seaver 2017), because they are like other aspects of culture. They are manifold consequences of a variety of human practices and reflect our own flaws, interpretations, and selective processes. Additionally, another point we frequently stress is that, based on the analyses they choose to perform, there is always a bias present and, as a result, they may see certain aspects but miss others simply because they have already selected what they are looking for. Based on this concept, we can look at data as something not strictly technical or isolated, but as if they are enacted by practices that do not make a strong distinction between technical and non-technical concerns, but rather blend them together (Seaver, 2017). An example of dialogue between technical and non-technical, very similar to the case history on Viva Insights, is Paff's work (2023), which is reported in the article ‘Designing Machine Learning Products Anthropologically: Building Relatable Machine Learning’. The author describes working with a company that uses machine learning algorithms to suggest new products for its users to buy. The anthropological approach was used to better understand how users interact with these suggestions and their expectations of the recommendation system. Specifically, the author focused on understanding the human experiences and perspectives involved in using machine learning products. As a result of this analysis, it was found that users were often confused about why certain recommendations were being made and they expressed the desire for a more detailed explanation of the suggested product selection process.

Utilising these results as input, the company decided to integrate an explanation function that showed users why certain products were selected for them. As a result, product suggestions became more understandable and closer to users’ expectations, thus improving the overall product user experience.

Like in Paff's case, I have created together with my colleagues a product, which consists of standard packages that examine both the customer experience and the technical capabilities of the tool when activating Viva Insights and other Viva modules. These packages take into account the most commonly occurring needs expressed by clients, which have led to changes in previous design experiences over the course of work. They provide a starting point, with suggested analyses that customers may not have initially considered, since they may not have realised that other needs exist.

This also happened in the case of the French client already mentioned. The client's initial request was to examine this type of collaboration during company working days, Monday through Friday. As the customers tested the Viva Insights platform and saw the analyses, they realised they needed to include Saturdays and Sundays as well. This way, they would have a complete overview of after-hours collaboration for the whole week. This together with Paff's case are concrete examples of how collaboration between two different technical approaches can yield significant results. In the next paragraph, I will explore this theme more deeply and highlight how this awareness has been strengthened through a study of computing culture, which I previously mentioned.

Applying a Socio-Technical Lens: Anthropology and AI in Health

Data science is one of the new disciplines emerging in order to adequately make use of all the data that is being generated from our computers, mobile devices, and increasingly networked societies (Countee 2015; Latour 2005). Data is progressively becoming a ubiquitous cultural practice, shaping our everyday world in steadily more obvious ways. The information being collected and analysed has become essential to business decisions, urban planning, scientific research, media content production, among others: just think of the rules related to data and GDPR⁸, which are changing entire legal systems, or ChatGPT⁹, initially banned in Italy. Data, as understood in this way, cannot be analysed as isolated objects. If we understand data as enacted by the practices used to engage with them, then the stakes of our own method and interpretation can change. The reason is we are not remote observers but rather active enactors, producing data as particular kinds of objects through our practices and our research (Seaver 2017; Latour 2005).

A concrete example to illustrate my point can be drawn directly from a recent incident that occurred during my daily work routine. One day, a colleague, with whom I typically do not collaborate professionally because his area of expertise does not overlap with mine, thought of involving me in the early stages of devising a bot designed to assist people who had undergone a specific trauma and medical experience. He asked me to partake because he believed my background could be important in designing a solution that would be as people-friendly, responsive, and non-robotic as possible.

Many reports and research papers, such as those by Countee (2015), Engler (2015), Wheeler (2018), Bell (2011; 2017) state that a qualitative approach to data analysis is increasingly necessary to fully make sense of the data. Qualitative analysis not only provides context for the data but also helps understand what the data implies and enables the creation of a strategic plan to use those implications to move forward.

This was a case where I felt compelled to accept the proposal enthusiastically. I was put in touch with some members of my colleague's team, including data scientists, one of whom was studying the behaviour of the algorithm mentioned earlier.

What my colleagues and I did was to concretely apply the socio-technical approach mentioned by Seaver (2017). The first step was defining the problem. The data scientists and I put in place a kind of enhanced requirement gathering¹⁰ (Mordenti 2022), where they gave a careful and clear definition of the problem, the need, and the objective of the analysis itself (Mordenti, 2022). I, on the other hand, paid special attention to the context in which the final product was and should be dropped (Engler 2015). In this situation, the challenge was figuring out how to train the algorithms that would give life to the bot mentioned above to make it as people-friendly as possible. My approach, once I understood the objective, was fundamentally qualitative. I asked a few questions to better discern and define the context such as: Do we know anything about patient pathways? How did they experience them? What meanings do they give them? Approximately, what are their family and social situations? Are there any interviews?¹¹.

I justified my requests by explaining how this type of detail had helped me in a previous field experience on medical anthropology in Indonesia to better comprehend my patients and their approach to reality, as well as, the way in which their practices were defined regarding treatment and the use (or non-use) of the national health assurance.

From a previous fieldwork experience in hospitals in Indonesia and previous experience with cancer patients during a medical anthropology course at the University of Milan-Bicocca, I learned that patients facing a major illness need to talk about what they are going through. However, they do not always do so immediately, especially if they lack confidence. Therefore, they tend not to speak about their condition right away, and questions like, ‘How are you?’ could be counterproductive. Instead, I learned that it is more effective to start a conversation by asking non-directive questions or things to distract them from their condition, such as, ‘What did you eat today?’.

Based on this experience, it became clear that it would make sense to educate the algorithm to be a non-directive assistant, allowing the patient to take the necessary time to start talking about their illness.

Making data available from a case that was similar to the one for which the algorithm was designed, allowed us to start reflecting and designing an educational path for the algorithm based on a tangible and replicable case. Contextual clues, such as knowing that patients are initially untrusting but need to recount their illness story, that the perception of self-changes, that they have a detailed historical and terminological memory of their illness, and that they have different motives for finding a reason to continue living (children, friends, parents, new perspectives), among others, made it possible, for example, to start the bot's interaction journey with a wider approach (Master 2023; Pizza 2005). A similar case related to creating a tool with a user experience designed around users’ needs is the collaboration between Microsoft and the global team of trained anthropologists at the furniture company Steelcase (Briggs, 2017). This partnership aimed to develop a line of furniture products to improve the worker experience in meeting rooms and shared work areas. In particular, the alliance focused on designing the work environment to support the use of technologies like Microsoft's Surface Hub, an interactive screen that enables real-time collaboration in a meeting. Steelcase furniture is designed to enhance the user experience with Surface Hub, ensuring easy integration and optimal use of technology in modern work environments. As a result, the Steelcase product range for Microsoft Surface is now available worldwide.

The role of anthropologists in the project was to gain a deeper understanding of the needs and behaviours of workers in modern workplaces. Through an ethnographic study, the anthropologists aimed to understand the behaviours and challenges workers face in their work environments and identify their needs so that they could create furniture products that best meet them. Working closely with Steelcase designers, the anthropologists guided product development based on the specific requirements of workers, with the goal of creating furniture products that are designed for people and provide a comfortable and functional work environment, improving worker productivity and wellbeing.

It's All about Data: A Matter of Perspective

While those may seem like minor examples, and in the bot case I was asked to provide just simple consultation, they illustrate the immense potential of multidisciplinary collaboration. The implementation of promotional ethnographic analysis techniques by data science, such as social network analysis, text mining and machine learning, allows for new questions about human experience to be formulated and helps identify patterns and structuring relevant to anthropology. Similarly, anthropology can help provide context and meaning to data obtained through data science techniques, guiding the design of the algorithm itself, understanding ethical assumptions, social practices, power relations and cultural nuances which influence the decisions that machine learning algorithms make, allowing machine learning techniques to be used more accurately and reliably, and avoiding any unrealistic results (Engler 2015).

If both disciplines can apply their knowledge and skills jointly, it is possible to achieve surprising and innovative results, as demonstrated in many cases described in scientific literature.

Conclusions

This article is based on the idea that multidisciplinary collaboration between seemingly very distant fields such as anthropology and data science is the key to producing a more extensive understanding and interpretation of some of the most pressing social problems of our times (Raffaetà 2020; Engler 2015; Kitchin 2014). To support this position, I have cited some examples from my current work experience in IT consulting, connecting it with my experience in the field in Indonesia and the scientific literature.

I talked about how the initial impact with the IT world was difficult, in many ways like I was facing another culture and how my daily work with data scientists and the various, more or less, technical figures with whom I interact has allowed me to immerse myself further into the field. In addition, I was able to clearly see the depth of my technical background, which I used to interface with all of them, making it as cryptic as they were to me at the beginning. Once I gained this awareness, dialogues with colleagues, especially data scientists, allowed me to find concrete commonalities and complementarities (Engler 2015) between our methodologies.

Anthropologists and data scientists start from specific theoretical assumptions and rely on different ways of acquiring and analysing data. Data science seeks to identify regularities in the world through a quantitative and computational approach. Meanwhile, anthropology studies the complexity of the world through systematic observation and narrative description of human events.

On the one hand, an ethnographic approach helps to assess the naturalness of data, capture the complexity of human experience, and challenge the apparent immediacy of data science, helping to develop new research questions. On the other hand, the analysis process of ethnographic data could be helped with the use of machine learning techniques, like text mining, to extract relevant patterns and themes, making the analysis process more efficient and accurate (Engler 2015). An example is my use of ChatGPT (Moses 2023), to rationalise and establish a lot of notes that later gave rise to this article.

Based on the reasons I mentioned in this article, I am convinced that a socio-technical approach (Seaver 2017), can enable us to overcome the limitations of traditional research, bringing new insights into human experience, using innovative tools and an interdisciplinary approach (Engler, 2015; Kitchin, 2014). All this represents a great opportunity to push the boundaries of knowledge and test new approaches in the world of technological and scientific innovation.