On the Contesting Conceptualisation of the Human Body: Between ‘Homo-Microbis’ and ‘Homo-Algorithmicus’

Abstract

Microbiome science has highlighted human and microbial interdependency, offering a radical epistemic shift from the individualistic view of the human body and self. Research has accordingly offered to see humans as ‘homo-microbis’ – complex biomolecular networks composed of humans and their associated microbes. While social scientists have begun to address microbiome science, the proliferation and commodification of the homo-microbial episteme have largely been overlooked. Based on an ethnographic account of a research project that offered microbiome-based personalised nutrition and the successful start-up that emerged from it, this article examines the emergence, proliferation, and commodification of the homo-microbial body. We show that this episteme necessarily depends on opaque machine learning algorithms; that the microbiome is paradoxically seen as a data-driven individuating marker; and that homo-microbis is, in fact, also a homo-algorithmicus – a being that can only access its non-human sub-parts by blindly following opaque algorithmic recommendations in an app.

Introduction

Scientific knowledge has always influenced how humans conceptualise themselves as a species. From the Copernican and Darwinian revolutions (Ruse, 2019) to psychology and psychotherapy (Illouz, 2008), from the Human Genome Project (Nelson, 2016) to Immunology and Brain Science (Pradeu, 2011), for centuries, scientists have been remodelling the ways humans see themselves as a species, how human subjects are understood and constituted (Jasanoff, 2019), and how human bodies, health, and self are conceptualised and experienced (Mol, 2002). Recently, however, the study of the human microbiome – the vast microbial communities that live in, around, and on the human body (Lederberg, 2001) – offered a dramatic shift from previous conceptualisations of humanity. Research has shown that the number of microbial cells in the human body is at least equal to the number of human cells (Sender et al., 2016); that human evolution always included symbiotic fusions with various kinds of microorganisms (Margulis and Fester, 1991); and what is commonly seen as intrinsically human faculties (like human cognition, emotion, or behaviour), in fact, stem from complex interdependencies between humans and microbes (Rees et al., 2018). As Helmreich argued, instead of defining humans according to their creations (Homo Faber) or their cerebral capacities (Homo Sapiens), we should see humans in light of their intimate ties with the microbial world: As ‘Homo-Microbis’ (Helmreich, 2014: 53) or ‘Holobiont’ – biomolecular networks composed of the host and its associated microbes (Bordenstein and Theis, 2015: 1).

These findings transcend some of the fundamental dichotomies that guided Western thought throughout modernity, such as self/non-self, subject/object (Gilbert et al., 2012, 2018; Ironstone, 2019), body/environment (King and Weedon, 2020), and life/death (Schrader, 2017). As Palsson (2013) argued: ‘[I]ncluding the microbiome as a fundamental part of the body, and not just as an appendix to a “host”, suggests that we need to revise and expand our notions of humans’ (p. 240). Microbiome research opens a new ‘space of possibilities’ (Rees, 2016: 197), calling into question humans’ long-standing view of themselves as individuals (Gilbert et al., 2012, 2018). However, while humans are increasingly conceptualised by their relationships with microbes (Parke, 2021), the latter are mostly ‘seen’ and understood via machine learning algorithms (Benezra, 2016). Accordingly, start-up companies have begun to offer popularised, microbiome-based personalised products, paving the way for a microbiome-based biotech industry (Rajan, 2006; Sharon, 2018). But what happens to this epistemic shift when it gets commodified? What happens when the homo-microbial conceptualisation is sold to consumers in an app? How do biology, data science, and start-up culture intersect in making knowledge (Knorr-Cetina, 1999) about the human body? And how is this episteme being used, distributed, and implemented in affluent societies (cf. Meloni, 2018)?

While social scientists recently began to focus on microbiome science from various perspectives (see, for example, Beck, 2021; Benezra, 2020; Ironstone, 2019; Lorimer, 2020; Nading, 2016), the preoccupation with the epistemic turn this science brings about remains largely theoretical. Moreover, the proliferation of the homo-microbial episteme and its commodified applications have largely been overlooked (with the exception of Raffaetà, 2022). This article aims to fill these gaps by offering a sociological, empirical account of the commodification of the homo-microbial self – from its inception in a university lab to its commodification by a successful start-up company. We show that in its commodification, the microbiome is paradoxically seen as an individuating marker that does not question humans’ individuality, but rather highlights it. We also point to the centrality of machine learning algorithms in constructing the homo-microbial self, and argue that this episteme, in fact, reinforces a heightened preoccupation with an individualised and singularised human body.

Homo-Microbis: The Symbiotic View of Human Individuality

Amid the COVID-19 pandemic, in which a coronavirus strain (SARS-CoV-2) disrupts people’s lives on a global scale, the suggestion that microorganisms profoundly affect both the human body and the broader social order might seem trivial. However, while microbes’ effects on the human body have been known for over three centuries, the diversity and complexity of microbial communities and the relationships that microbes have with their human ‘hosts’ only recently began to receive scientific, medical, and cultural recognition. Previous perceptions of microbe–human relations were either adversarial, as reflected in the early modernist (Pasteurian) scientific attempts to eradicate microbes (Latour, 1993), or companion, as reflected in artisan cheese making, wine production, sourdough bread making, and the like (Dupré and O’Malley, 2007; Paxson, 2008). Nevertheless, both the adversarial and companion perceptions of human–microbe relations maintain that, ontologically, ‘they’ – microbes – are different and separate from ‘us’ – humans. But by the beginning of the 21st century, these two views were replaced by a sheer fascination with microbial life and a considerable reorganisation of our ideas about human–microbe relations.

Starting from the 1970s, researchers began to argue that the human body consists of an overwhelming number of microbial cells (Sender et al., 2016). Researchers also highlighted the remarkable variety of microbial cells, estimating that the human body consists of over 1000 different microbial species that collectively offer significantly more diversity and flexibility than the human genome alone (Gilbert et al., 2018: 392). Accordingly, it was claimed that ‘human’ cells have rich microbial histories, as human evolution is seen to have always included symbiotic fusions with a variety of microorganisms (Margulis and Sagan, 1986), and as much as 50% of our DNA is reported to be acquired exogenously (Lozupone et al., 2012). Moreover, what was previously seen as intrinsically human faculties were shown to stem from complex interdependencies between humans and microbes: the microbiome was described as essential for human metabolism and immune system (Sekirov and Finlay, 2006); changes in the human microbiome were seen to correlate with a wide array of illnesses (Eisenstein, 2020); and the microbiota were even seen to affect human cerebral functions, such as mood (Cryan and Dinan, 2012), personality, behaviour, and cognition (Vuong et al., 2017). These findings suggest that humans are far from being bounded entities, detached and essentially different from the ‘outside world’, but are equal parts in a buzzing and ever-changing community of human and microbial creatures (Rees et al., 2018). As Benezra (2018) succinctly described it: ‘microbes are in and out, human and nonhuman, simultaneously environment and body’ (p. 287).

Microbiome research offers a considerable reconceptualisation of human individuality as it is commonly viewed (Gilbert et al., 2012; Margulis and Sagan, 1986: 326). ¹ It envisages humans as collective entities – superorganisms that include multiple species and billions of individual organisms (Juengst, 2009: 142). And if according to this view, the human body has always been a cooperative multi-organism, then the world was never populated by discrete, autonomous ‘individuals’ (Hird, 2009). Instead, the human body, from its very onset, has been comprised of an array of companion species (Haraway, 2016) engulfed in a perpetual process of ‘mutual becoming’ (Beck, 2021; Lorimer, 2016: 60).

This reconceptualisation of the human body and self closely coincides with the ontological turn in social sciences – the methodological and analytical openness to other ways of being parallel to, and intertwined with, human existence (Kirksey and Helmreich, 2010; Kohn, 2015). This view similarly resonates with feminist and post-humanistic views of the human body, self, and identity (Haraway, 2003, 2016; Wolfe, 2010). While these intellectual streams have gained considerable popularity in some corners of academia, the symbiotic view of humans goes far beyond the predictable auspices of postmodern thought. This ‘style of thought’ (Rose, 2007: 12) is widely shared among scientists, science writers, and popular thinkers alike, and is repeatedly revibrated by some of the loudest media outlets and most powerful medical and scientific institutions today (Sangodeyi, 2014; Paxson and Helmreich, 2014; Morar and Bohannan, 2019). Accordingly, the human body is increasingly described by models of cohabitation, cooperation, community, and symbiosis (Ironstone, 2019: 33). But while humans are increasingly conceptualised by their microbial allies, the latter are mostly ‘seen’ and understood via machine learning algorithms.

Datafied Microbes

The move towards a homo-microbial view of the human self has been enabled by a broader development in the life sciences towards computational methods, and particularly by the turn from biology to a datafied ‘Big Biology’ (Stevens, 2013: 44; Tempini and Leonelli, 2016). The ambitious Human Genome Project is perhaps the first and the most emblematic example of this turn, and with it, the meaning of the very objects of biology began to change. Biological objects became virtual ones, and tangible organisms and cells transformed into computer code. Thus, one can only ‘see’ a genome using automated sequencers, vast databases, and visualisation software (Stevens, 2013). However, the move towards a symbiotic view of humans stems from a more diverse methodology, one that looks beyond a single gene or genome to identify the various organisms that symbiotically constitute the superorganism previously seen as ‘the human individual’. That methodology is metagenomics.

Metagenomics involves the computational sequencing of mixed samples of DNA from multiple species to identify all of the organisms therein and study their molecular interactions (Juengst, 2009: 132). Unlike the previous focus on individual human genomes, metagenomics explores exceptionally large communal gene pools collected from diverse microbial communities (Dupré and O’Malley, 2007: 836). This methodology is necessarily based on vast databases and complex machine learning algorithms that find patterns in them. Such algorithms render microbes visible and offer to link between human states and microbial states (Benezra, 2016).

However, algorithms do not simply mirror the biological ‘reality’. As Stevens explains: ‘[Algorithms] bring the biological and the computational into new relationships, as biological materialities are restructured into databases’ (Stevens, 2013: 8–10). As we will demonstrate below, such relationships extend beyond the scientific circles as they pave the way towards the popularisation and commodification of the microbiome.

Microbiome and Society

While microbiome research recently received enormous scientific and public attention, only a handful of social scientists address it. In line with anthropological and philosophical ventures into interspecies and multi-species research (Haraway, 2003; Kirksey and Helmreich, 2010) and into the ties between microbes and society (Latour, 1993; Paxson, 2008), researchers began offering sociocultural accounts of the human microbiome. Nading (2016), for example, explored the sociality of microbes, arguing that ‘microbes become social when people draw them into explanations about behavior, health, politics, and economics’ (p. 565). Focusing on apprehensions of microbiome science in the United States and Nicaragua, he argued that the circulation of this category has so far been limited to the Global North. Benezra (2016) has similarly highlighted the globalised ramifications of microbiome research by focusing on a microbiome study on malnutrition in Bangladeshi women and children. She showed that our microbes are made visible using big data analyses and that the causal relationships between microbial populations and malnutrition are produced through datafication (Benezra, 2016). More recently, Benezra (2018) highlighted the post-racial aspirations of microbiome research in its attempts to characterise human states microbially. Lorimer (2020) recently described the emergence of ‘probiotic turn’ and explored the promises and limitations of probiotic cultural practices today, and others characterised the immense popularisation of microbiome research in the United States (Beck, 2021; Sangodeyi, 2014).

While social scientists’ interest in the microbiome is rising, sociologists are slow to join this discussion. While scholars offered various perspectives on the emergence of the homo-microbial self (Helmreich, 2014; Ironstone, 2019), these have been almost entirely theoretical accounts that focus on ontological questions (who we are), rather than epistemological or phenomenological ones (regarding the creation, circulation, and acceptance of this new episteme). As Greenhough and her colleagues recently argued (2020), the commodification of the microbiome remains largely underexplored. This article aims to fill these gaps by offering a sociological, empirical account of the production and commodification of the homo-microbial self.

Empirically, this article focuses on the case study of the Personalized Nutrition Project (PNP), a recent and prominent example of the study, commodification, and dissemination of the human microbiome and a prominent illustration of the wider probiotic turn (Lorimer, 2020) in affluent societies. Launched in 2012 in one of Israel’s top scientific institutions, the PNP aimed to test the effectiveness of personalised microbiome-based nutrition. Drawing on a series of large-scale bioinformatic and clinical studies, the PNP aimed to explore human–microbe relations and their impacts on health and disease, focusing on participant nutrition based on blood glucose level. Findings from the PNP were first published in 2015 in a top scientific journal, and the project was immediately lauded as one of the first research projects to demonstrate that people’s responses to food significantly vary and that this variation strongly depends on their gut microbiome. These findings raised tremendous media attention, and soon thereafter, a microbiome-based start-up was established – NutriAI.

Based on this case study, we explore the constitution and circulation of ‘homo-microbis’ in three interrelated fields: in the PNP university lab, in the popularisation of the PNP project, and in the emergence of a microbiome-based start-up company – NutriAI. We conducted 20 semi-structured interviews with microbiologists, data scientists, nutritionists, and entrepreneurs who were involved in the PNP and NutriAI. ² We also critically analysed popular literature on the PNP, newspaper articles, online videos, and a wide range of sources on NutriAI, including online videos, newspaper articles, and lectures by their CEO.

The data were analysed using a thematic approach (Strauss and Corbin, 1997); we read and reread the data, identified recurrent themes and major concepts, and clustered similar segments together. In the following, we analyse the implications of the PNP through four interrelated stages: first, we focus on the PNP scientists’ personal views on human–microbe relations; then, we follow the immense popularisation of this research and its introduction to a broader public; next, we discuss the prominence of machine learning algorithms in producing microbiome-based personalised predictions; and finally, we focus on NutriAI and the commodification of the microbial-self in an app.

The Scientific Construction of the Homo-Microbial Self

When asked about their views on the microbiome, the PNP scientists seem to have embraced the symbiotic view and see humans as homo-microbis. As Dov, a physician and investigator in the study, told us in an interview:

For years we have been treating these microbes as inferior, or worse, as physicians, we used to see them as our enemies. Now, if you look at a single cell, it might look inferior, but as communities, they are just incredible! [. . .] So, with time, I increasingly learn to appreciate this ecosystem that’s inside us. [. . .] I no longer see it as an ‘internal ecosystem’, but as an ecosystem that [. . .] constitutes us as a holobiont.

Dov describes the conceptual change he went through, as a physician and a researcher, regarding the microbiome – from seeing microbes as inferior enemies, acknowledging their ‘incredible’ power, and finally, accepting that they, in fact, ‘constitute us’ as a species. According to him, the microbiome is not ‘inside us’, but it is us. Shira, a PNP bioinformatics scientist, similarly said: ‘I think the microbiome has a very significant role in our existence, in who we are’. Echoing the literature cited above, Dov and Shira break away from the modernistic, atomistic view of the human body that guided human thought for centuries – a self-contained package with clearly defined boundaries – and instead, they echo the view of humans as homo-microbis (Helmreich, 2014) – a symbiotic, multi-organismic, non-individualistic being.

Omer, a PNP computational biologist, said:

Learning about the microbiome completely changed how I understand our place in the world. [. . .] I used to look at this organism, the human being, as an entity that has autonomous thought, but that changed with [my awareness of the] microbiome because [I realized that] many other organisms play a role here, and they all have their own interests. [. . .] So, I think . . . I mostly feel . . . I understand that what’s ‘running this system’ is more than my genome or my lifestyle, it’s also other, selfish creatures, and that’s why there’s another level of randomness here, beyond that of our genes, beyond that of our subjectivity.

Omer describes how a growing awareness to the microbiome made him forsake his previous, individualistic view of humans and recognise that ‘many other organisms’ are, in fact, involved in this symbiotic being. According to him, this is not only a logical, factual argument but also a deep, embodied understanding of a newly realised ontological state (‘I think . . . I mostly feel . . . I understand’). Interestingly, Omer’s narrative seems slightly less harmonious than Dov’s or Shira’s. According to him, microbes may be symbiotically attached to humans, but they are also separate ‘selfish creatures’ who adhere to ‘their own interests’. The homo-microbis Omer envisages, is one that still differentiates between humans and microbes, one that is not entirely detached from previous, more belligerent views of microbial life.According to this perspective, humans and microbes may be different, even adversarial entities, but they are inextricably linked – they are different parts that constitute a whole. Omer’s choice of words is telling: he describes the human body as ‘a system’ and the microbiome as the factors that add ‘randomness’ to this system, alongside other, more human-centred elements (the human genome, lifestyle, or human subjectivity). As a computational biologist, he examines the human body as a stochastic information system whose different elements merely make it more or less random (‘there’s another level of randomness here’). According to him, microbes may co-constitute the human body and self, but that primarily means that they add ‘noise’ to our efforts to make sense of ‘our place in the world’ and that the human ‘core’ is becoming harder to decipher. Hence, at first glance, it seems like the PNP scientists have deeply internalised a homo-microbial, symbiotic view of humans, but a closer look at the project offers a more nuanced account.

From Symbiotic Allies to Markers of Individuality

By 2015, the PNP researchers had published several research papers in top scientific journals, offering to use the gut microbiome to predict individualised blood glucose responses to particular foods. They argued that because people react differently to different foods, their bodily responses to foods can be predicted by their microbiome. The PNP line of research is heavily based on a machine learning algorithm that integrates participants’ gut microbiota with other parameters (such as dietary habits and physical activity) and offers them personalised diets that promise to lower their blood glucose levels.

The PNP studies quickly became a media sensation, featured in thousands of articles in major media outlets (BBC, CNN, The New York Times, and more). The project’s head scientists gave countless public lectures, participated in prime-time radio and TV shows, and co-authored a popular book about their project. In these public engagements, they repeatedly described the ‘seismic shift’ their research brought about. As they write in their book,

Our observations changed everything we used to know about nutrition [. . .] The implications of our findings are wide, and they demonstrate that generalized nutritional advice is limited because it focuses on the food rather than the person eating it. We believe that we are entering a new era in nutrition study. We are moving away from standardized diets and dietary advice and toward a new frontier of personalization in its many guises.

These statements describe a dramatic shift that directly stems from the PNP research – one that promises to forever change the relations between people and the food they eat. This shift offers a move from previous, standardised diets (e.g. Paleo Diet, Atkins Diet) – towards a tailor-made, personalised diet, customised for individualised subjects to perfectly fit their nutritional needs and bodily properties.

This view is based on the now popular trope of personalisation – heralded in the early 2000s and found in a swath of online, consumerist, and medical services. The personalisation paradigm sees people as singular individuals who are innately differentiated and should accordingly enjoy technological services that will fit their unique profiles and needs (Even Chorev, 2020; Kant, 2020; Lury and Day, 2019; Prainsack, 2018). In this case, personalisation indeed ‘focuses on the person’, as the PNP scientists describe, but it is also based on the person’s microbial allies. As their book explains:

Our research enabled us to develop a microbiome-based algorithm that precisely predicts people’s personal reactions to specific foods even before trying them [. . .]. Personalized nutrition is a data-driven approach with the potential of creating individually tailored diets. It is a view that considers the many different variables that affect the symbiotic ties between humans – the host body – their microbiota, their dietary responses, as well as other changing clinical states.

According to this narrative, the microbiome’s role in this paradigmatic nutritional shift is highlighted, but the ‘microbial self’ – homo-microbis – is nowhere to be seen. Instead of a symbiotic, multi-organismic being, the PNP scientists see the microbiome as a useful data point for algorithmically informed, personalised nutrition. According to this narrative, the microbiome is an important marker that does not question or diffuse humans’ individuality or singularity but instead highlights it, presenting each person as a unique human creation that requires (and is worthy of) personalised algorithmically tailored nutrition. The fact that humans are also comprised of other organisms and are dependent on the symbiotic ties with them does not escape the PNP scientists, but for them, this is no reason to distance themselves from an individualistic view of humans. On the contrary – it is an opportunity to zoom in a little closer to a micro-individuated, human-centred, personalised view of the human subject.

That is, in its popularisation, the PNP acknowledges the microbiome, but rather than offering a posthuman perspective, or multi-species one, it serves to highlight each person’s idiosyncratic uniqueness. This perspective is closely tied to other non-human factors through which such individuated personalisation is achieved – machine learning algorithms.

On the Algorithmisation of the Microbial Body

In an interview with Ziv, a PNP scientist, he recounted:

Ever since my days as the first PI in Israel to study the microbiome, we have been dealing with some VERY big medical data [. . .]. In the beginning, we used to see [the microbiome] as a long-neglected organ, [but now] we have data on 3 million genes, compared to only 30,000 human genes we previously dealt with! So, this is crazy big data. [. . . A]nd in fact, big data was what made the creation of this field possible, to begin with.

Ziv refers to the microbiome as ‘an organ’, a mere part of the human whole. It is a view that echoes previous, individualistic conceptualisations of the human body. At the same time, Ziv also acknowledges the immense complexity of this ‘organ’, with its countless metagenomic, multi-species’ sub-parts. Echoing the computational turn in biology (Markowetz, 2017), he contrasts his previous work in human-centred genomic research and states that studying the microbiome necessarily entails analysing vast swaths of information – ‘crazy big data’. He further explains that big data is, in fact, the bedrock upon which the study of microbiome became possible. As Benezra has shown, data-driven bio-computational methodologies make visible previously unknown microorganisms and their complex interconnections (Benezra, 2016: 342). Hence, scientists can only ‘see’ our incredibly diverse microscopic allies through algorithmic eyes. But what does such an algorithmic vision entail? And how is the homo-microbial self seen and constructed through algorithms?

Yariv, a PNP researcher, described:

Some discoveries are just too complex for the human brain because they are based on big data, and we just have to accept them as they are. And then, some discoveries are simple and easy to grasp, like a [study of a] single bacteria that secrets a single metabolite which improves or worsens an illness – that, we can get our heads around. The PNP belongs to the first type, because of its size, because of the amount of data [it is based on], because of the complex interconnections that combinatorically exist between the microbes and their hosts, and between the different algorithm features. So, in the meantime, we are at this level [of understanding], and it’s OK because thanks to that, we can provide people with nutritional recommendations.

Yariv describes their data-driven methodology as something that offers access to things that are beyond human grasp. Such analyses provide results that can be acted upon (e.g. by giving ‘nutritional recommendations’) but not entirely comprehended, and people are left to ‘accept them as they are’. After all, machine learning algorithms afford what Andrejevic described as a ‘post-comprehension strategy of information use’, one that addresses the challenges of information overload (Andrejevic, 2013: 87). It is a case in which artificial intelligence is allegedly needed to complement human intelligence, and in this case, it is needed to cross the interspecies divide between humans and their microbial allies.

Such algorithmically infused methods offer more than a methodological shift, but an epistemological one – a shift that profoundly affects how, what, and whom can be seen and acted upon. Microbes may be an indivisible part of the human body, and metagenomic methods might afford ways of working with these microbes, but it is a way that largely escapes human cognition, a way that is only partially understood. In other words, like other algorithmic methodologies, PNP’s algorithms are black-boxed (Pasquale, 2015). As Hila, a PNP dietician, explained:

We’re not telling [participants]: ‘you have this bacterium, and therefore you have this or that [glycemic] response to it’. [. . .] We only enter the parameters, and the algorithm gives us a certain score.

Shira, a bioinformatics scientist, said:

The thing is, that this algorithm is black boxed. You don’t really know why it chose this particular prediction or that specific model.

While the PNP algorithm is said to predict which food would fit which participant, the exact reason behind this ‘suitability’ remains largely unknown (or ‘post-comprehended’, to use Andrejevic’s terminology (2013)) – even to the scientists who created these tools. Similarly, while the gut microbiome is constituted by billions of individual microbes and thousands of types of bacteria, viruses, and fungi, these algorithms are not designed to identify the specific effects specific microbes have on specific humans. These algorithms may shed light on the homo-microbial body and self (e.g. by teaching participants they should refrain from eating a particular food), but this light is incredibly opaque, as people remain in the dark regarding the exact meaning of the algorithmic outcome, its ‘score’, and the exact nature of the microbial communities that live in/with them.

Thus, people may be co-constituted by rich microbial communities, but in order to understand, use, or ‘communicate’ with these microbes, they are bound to use opaque algorithmic recommendations. Participants’ diets may stem from a deeper understanding of the homo-microbial allies, but these systems are not programmed to know people, nor their microbes, but to affect them, to algorithmically nudge them towards healthier ways of living. In this case, the datafied microbiome merely feeds a ‘choice-inducing algorithm’ (Kotliar, 2021) that triggers motivational feedback loops (Schüll, 2019) between participants and their apps. Thus, the homo-microbial self is almost necessarily also an algorithmic self – an identity that is constituted through and is acted upon by machine learning algorithms. It is a holobiont (homo-microbial) body that can only be accessed (but not entirely known) by data-intensive computerised tools.

While most of the scholarly discussion on algorithmic identities revolves around online social data (Cheney-Lippold, 2011; Kotliar, 2020), in this case, it is metagenomic data about vast communities of organisms that constitute the self; it is a data-driven self that does not entirely rely on its human parts or on online human behaviour, but on its microbial parts and on the complex interconnections between them. Like other online identities, this microbiome-based algorithmic identity is prone to biases and misrepresentations (Buolamwini and Gebru, 2018; Noble, 2018). After all, microbiome science tends to offer a universalising and post-racial view of people (Benezra, 2020), and personalisation algorithms similarly offer a post-demographic one (Rogers, 2009). In this case, people’s bodily states and nutritional recommendations are determined according to various algorithmic correlations, not their race, sex, or ethnicity. Moreover, while such data are logged into the PNP’s databases, participants’ demographic data are so deeply mixed with their microbial data, and the algorithmic results are so cryptically opaque, that people’s offline identities and their effects on their nutrition remain in the dark.

This datafication also paves the way for other data-driven applications. As Omer explained:

I have some background in biology, so I got into microbiology, but I have to tell you – it doesn’t really matter. A Waze engineer could just as easily have done this work. You can do it as an expert system, as in ‘I read about this or that bacteria, let’s add it to the algorithm features’, it might work. But letting the data speak is usually better.

Omer suggests that ‘letting the data speak’ promises better results than closely assembling the features that would constitute the algorithm. In this view, he highlights one of the main differences between ‘traditional’ biology and computational biology (Markowetz, 2017) and one of the main epistemological differences between ‘classic’ scientific discoveries and data-driven ones. According to the latter, knowledge about microbes stems from opaque algorithmic analyses, not a human-led, expert-based assembly of its algorithmic rules. This data-driven epistemology, alongside the datafication of microbes, affords an understanding of microbes that is not necessarily attached to ‘traditional’ biological knowledge, but to a computational, algorithmic one. That is why, according to Omer, an engineer from the Google-owned navigation app Waze ‘could just as easily have done this work’. In other words, the fact that microbes are ‘datafied’ makes their data highly commensurable. Coding them into databases not only expands the types of experts who can analyse and make use of such data but it also allows the data to expand into new markets, away from the somewhat limited epistemic culture (Knorr-Cetina, 1999) of the university lab. The case of the successful start-up NutriAI provides an excellent case in point.

Towards a Microbiome-based Diet in an App

NutriAI is a successful start-up company that offers microbiome-based personalised nutrition. Founded in 2015 following an agreement with the PNP’s university, the company’s technology is explicitly based on the PNP research, and the PNP PIs are listed among its co-founders. After a generous seed investment, the company raised tens of millions of dollars and started offering its product to consumers in the Global North.

NutriAI is a high-profile company, with regular radio commercials, several TV and radio shows that showcase their product, and their CEO regularly speaks at public events. Upon registering to the service, NutriAI customers receive a home test kit that includes a stool test and a personal health and lifestyle questionnaire. After the kit is sent back to the company, customers’ microbiome is characterised using metagenomic sequencing, and the microbial characterisation goes into the company’s personalisation algorithm, alongside data from the questionnaire. Customers then receive an invitation to install the company’s app and a report on their microbiome. The report is illustrated with diagrams, graphs, and generic images of bacteria, and includes detailed theoretical explanations about the gut microbiome and its taxonomic classification. It also includes microbial interpretations, namely, the identification of microbial DNA sequences from the consumer’s stool sample. The report ends with a brief explanation of metagenomics sequencing.

At first glance, NutriAI’s report seems to highlight consumers’ microbiomes, hence echoing a homo-microbial view of humans. However, this report seems to primarily emphasise the medical and scientific foundations of NutriAI’s product, in an attempt to highlight their scientific authority in a last promotional effort before the consumers focus on the central apparatus that would guide them in this diet – NutriAI’s app.

To use the app, NutriAI’s customers need to enter any item they plan to eat, and the app instantly generates an allegedly personalised score for this item on a scale of 1 to 10. Customers can then try entering other items or check if a combination of several items will change the score that they receive. They can also use the company’s ‘coaches’, namely, dieticians, who help them persist in using the app. Thus, NutriAI’s app offers customers a ‘unique’ perspective on the foods they eat and allegedly predicts their personal response to each nutritional item. Hence, it helps them choose which items they should eat or avoid. As Miri, a NutriAI’s executive, described:

[Using our app,] I discovered about myself that apples, which are usually seen as ‘an apple a day keeps the doctor away’, right? So, for me, that’s just wrong. Apples just raise my [blood] sugar levels, while other fruits are healthier for me. That’s why I took apples off my diet permanently.

According to this executive, NutriAI’s app helps people reach significant self-revelations, ones that question traditional dietary wisdoms (‘an apple a day keeps the doctor away’) and replaces them with personalised ‘facts’, based on the unique idiosyncrasy of every individual. Following the PNP logic, Miri imagines her company’s clients as unique individuals whose diet requires a much greater specificity than traditional, unified diets. But while these tailor-made diets are based on customers’ innermost entrails, they also require meticulous human labour: manually entering each food one desires to eat into the app, getting its personalised score, and repeatedly checking alternative nutritional combinations.

This is a clear expression of the cultural logic of ‘healthism’ (Biltekoff, 2013; Guthman, 2012), one that is focused on personal health attainment and on placing responsibility on individuals for their malnutrition and its consequences (Yates-Doerr, 2015). While the microbiome is at least partly directing the dietary recommendations shown on the NutriAI app, human individuality, singularity, and agency are extremely highlighted here.

While people can be seen as singular-yet-plural entities (e.g. in a view that highlights people’s uniqueness and their homo-microbial symbiotic nature), NutriAI’s discourses and messaging are far from such a message. Instead, they seem to approach customers with a more ‘traditional’ discourse, mirroring that of other tech companies about individual humans and their technological tools. In the case of NutriAI, customers are expected to reflexively and meticulously engage with the app, and make conscious, responsible decisions about every food they desire to eat. Such calculable logics are widespread in various diets (e.g. counting calories), as well as in other nutritional paradigms and attempts to attend to alimentary uncertainties and concerns (Sanabria and Yates-Doerr, 2015; Landecker, 2013). In this case, however, customers are not basing their calculus on the characteristics of the food itself but on an algorithmically generated personalised prediction. They are encouraged to follow the algorithmic scores as closely as possible, adhering to its recommendation by the authority of its scientific rigour, of its algorithmic power (Beer, 2009), and on the belief that this 1–10 scale is a distant proxy of their unique microbial attributes.

The company’s website paints a similar picture. In the upper right corner, there is a stock photo of four people dining together. They are sitting in a bright room, wearing bright, semi-formal attire, smiling and laughing with wine glasses in their hands, and plates full of fresh foods on a white marble dining table. Above each plate, there is an arrow with a number that allegedly reflects the personalised score each of the diners received for their dish from NutriAI’s app. Three of the scores are relatively high (7.2; 9.7; 8.2) and are underlined with a light green line, while one diner’s score is lower (6.4) and is marked with an orange underline.

This picture reflects an affluent, positive, consumerist ethos, reflecting the apparent ease of use and clarity that comes with using the company’s services, as well as the relatively privileged audiences NutriAI’s messages are directed towards. At the same time, it also reflects a scientific, quantified ethos, one that is represented by the precise numbers above the plates. Like other quantified metrics and other personalised self-tracking devices (Hamper, 2020; Lupton, 2016; Sanders, 2017), they simultaneously refer to the food and to the diners who eat it. While these metrics also allegedly mirror the wants and needs of people’s microbiome, they are presented as much more conventional evaluations of users-as-individual-humans and their (nutritional) actions. Thus, the algorithmic scores promise people an opportunity for healthier nutrition, but they also offer a way of conducting a self-quantifying ‘body work’ (Featherstone, 1999; Sanders, 2017; Schüll, 2019) by self-evaluating through an algorithmic interpretation of their nutritional choices. Before anything else, the self that emerges from this picture is a quantifiable one. Accordingly, a prominent title to the left of the website’s picture reads: ‘Blood Sugar Control Made Easy: The Algorithm Diet Personalized for You’. Rather than highlighting the microbiome, or human–microbe relations, NutriAI primarily presents its diet as an ‘algorithmic’ one – a diet that is based on quantified metrics, individual responsibility, and self-surveillance. Indeed, self-tracking is often presented alongside promises to improve people’s health and wellness (Neff and Nafus, 2016: 135). In this case, however, it is not only human life that is being reduced to numbers (Neff and Nafus, 2016: 6) but also the life of the microbes that co-constitutes it. By this reduction to numbers, and by the obligation to meticulously log foods into the app, people’s microbial parts are foregrounded by their strictly human characteristics. People’s humanity is paradoxically highlighted, just because it is overwhelmingly dependent on algorithmic recommendations. Thus, in the case of NutriAI, the homo-microbial self has given way to a distinctly human quantifiable self, while the microbiome is only represented by simplified algorithmic metrics – distant proxies to the interspecies’ drama in our gut.

Conclusion

The ways in which humans conceptualise themselves have always emanated from the epistemic devices they use, from the technologies of self at hand (Rose and Miller, 2009), and from various biopolitical mechanisms (Foucault, 1978). Microbiome research is perhaps the most recent link in a long epistemic chain, and the picture it presents posits that the human body and self are, in fact, not entirely human. According to this emerging episteme, the relations between humans and microbes are not only reciprocal or coevolutionary. They not only reflect the symbiotic co-dependencies between different species, but rather a new view of humanity, a homo-microbial one in which humans’ bodies and selves are also microbial, and vice versa (Helmreich, 2014).

As we have shown above, the popularisation and commodification of the microbial-self entail a return to a differentiating, distinctly humanistic view of humans. By offering a tailor-made, personalised diet, the PNP scientists envisaged a singular subject, and the microbiome was merely seen as a marker that paradoxically highlights humans’ individuality. Focusing on the algorithmic work at the heart of the PNP, we also showed that humans’ ability to ‘see’, conceptualise, and potentially communicate with their homo-microbial self is necessarily afforded by machine learning algorithms. We argue that the algorithms should be seen as more than methodological tools in studying microbes, but as important actants (Latour, 2005) in constructing and sustaining the homo-microbial self. Nevertheless, algorithms’ involvement in this homo-techno-microbial holobiont does not provide a clear understanding of the homo-microbis, nor an easy way to ‘see’ or act upon it. As described above, algorithmic analyses of the microbiome are almost necessarily based on a triple black box: first, a microbial black box that stems from limited knowledge about microbial life; second, a ‘nutritional black box’ (Paxson, 2016) that stems from our limited scientific understanding of what is ‘good’ or ‘healthy’ to eat (Yates-Doerr, 2015: 55) and from the changing reasoning about how food and the body interact in and through metabolism (Landecker, 2013); and third, an algorithmic black box that stems from opaque algorithmic structures and their innate inexplicability (Pasquale, 2015).

Algorithms are also crucial factors in the commodification of the microbial self, as the datafication of the microbiome paves the way for other algorithmic uses of these data. Focusing on the successful start-up NutriAI, we argue that it intensifies the interconnected processes of individuation and algorithmisation; while the company provides consumers with basic information about their microbiome, consumers are encouraged to reflexively and meticulously engage with the company’s app, perpetually and laboriously ‘feed’ its algorithms with more data about their meals, and meticulously follow its instructions. Thus, NutriAI’s operations may be based on the microbiome, but it envisages a quantified human individuality and singularity, one that is based on (human) self-quantification practices (Schüll, 2019) and relies on algorithmically generated personalised predictions. Instead of highlighting humans’ symbiotic relationships with non-human microbial others, this view emphasises the human self, with its responsibility, agency, control over nature, and incessant preoccupation with human health and well-being.

Microbiome research might offer a profound ontological turn, one propelled by new epistemic devices (like metagenomic sequencing and machine learning algorithms), but if we seek to understand this turn, focusing on its scientific foundations is hardly enough. Such a view risks ignoring the political, economic, and material contexts in which these new conceptualisations are embedded (cf. Parry and Greenhough, 2018), as well as ignoring the ties between these contexts and the epistemic devices through which microbial artefacts are produced. As the case of the PNP reveals, ontological shifts may stem from the university lab, but they are moulded, popularised, commodified, and disseminated in particular cultural and technological contexts.

New algorithmic methodologies may grant people an opportunity to ‘see’ and communicate with their innermost allies, but this view emanates from specific technologies, with their specific affordances, constraints, and epistemologies. After all, algorithms do not simply mirror the biological ‘reality’, but they bring the biological and the computational into new relationships, as biological materialities are restructured into databases (Stevens, 2013: 8–10). Accordingly, algorithmic outputs do not have inherent meanings – they must be reinterpreted back into the language of biology, and to previous, socially accepted categories (Stevens, 2013: 200). Thus, considering the fact that microbes are predominantly conceptualised by machine learning algorithms and are commodified through personalisation apps, we argue that the homo-microbis is necessarily also a homo-algorithmicus – one constructed from the algorithmic calculation of trillions of microbial and human data points. It is a being that can only access its non-human sub-parts through self-quantifying consumption by blindly following opaque algorithmic recommendations on an app. Thus, the microbial imaginary (Greenhough et al., 2020: 3; Lorimer, 2016: 71) is inextricably intertwined with an algorithmic imaginary (Bucher, 2016), and the homo-microbial subject is only constructed through a complex combination of both.

Kotliar recently described how algorithmic epistemologies get amalgamated with previous, modernistic epistemologies and create ‘epistemic amalgams’ (Kotliar, 2020: 1162). In the case before us, the PNP – and the subsequent commodification of the microbiome – does not necessarily challenge the long-standing individualistic view of humanity, but rather, it gets amalgamated with an algorithmic episteme (Gillespie, 2016), as well as with self-quantification (Lupton, 2016; Schüll, 2019) and personalisation (Lury and Day, 2019) tropes. This amalgam does not offer a radical post-humanistic ontological view of human–microbes’ holobiontic nature. Instead, it paradoxically reinforces preceding individualistic views of the human body and self. It seems like we are still more accustomed to a self that depends on an app than to one that depends on our gastrointestinal allies; we find it easier to quantify and discipline our human identities and bodies than accept our holobiontic nature. However, perhaps the homo-microbial episteme is still in its infancy; its imaginary, with its accompanying discourses and narratives, is still underdeveloped, and its dependency on other algorithmic epistemes is only for the time being.