The Fertility Fix: the Boom in Facial-matching Algorithms for Donor Selection in Assisted Reproduction in Spain

Abstract

This article reads the uptake of facial-matching algorithms by fertility clinics in Spain through the lens of ‘the fertility fix’: a software fix to the social reconfiguration of kinship and a fixed capital investment made by competing fertility companies and firms. ‘The fertility fix’ is proposed as a critical, ethical lens through which to situate algorithmic facial-matching in assisted reproduction in the context of the racial politics of the face and phenotype and the spatial politics of market expansion. While an ‘infertility crisis’ is often mentioned when explaining the growth of the assisted reproductive technologies (ARTs) industry, the use of donated reproductive cells is tied up in societal, ecological and economic shifts. Combining Software Studies analysis with Marxist Feminist and trans*feminist perspectives on shifting re/production dynamics, the article details the role of computational technologies in promoting certain ideas and beliefs about family and fixing certain territories of capital flow.

KEYWORDS:

• Facial-matching algorithms
• assisted reproduction
• IVF add-on
• fixed capital
• software studies
• reproductive politics

The face and assisted reproduction

Around 2019, and increasingly throughout 2020, many fertility clinics in Spain began promoting their use of facial-matching algorithms for the selection of egg and sperm donors on their website homepages. Facial-matching algorithms compare a human face from a digital image or video frame with a database of faces and potentially select two similar images. In the context of the Assisted Reproductive Technologies (ARTs)¹ industry, algorithmic facial-matching is implemented by fertility clinics to help select donors, with the promise that the use of this software will ensure that the recipient’s donor-assisted child will resemble them or their partner. While facial matching algorithms have also appeared in the U.S (Fairfax Inc. has developed its own trade-marked Fairfax Facematch® algorithm), in Spain the facial-matching algorithm is currently experiencing a commercial boom. In a national context in which only anonymous donation is permitted, and where all donor selection is performed by the clinics, the facial-matching algorithm allows the clinics to make certain claims regarding the accuracy of their choices, beyond the typical evaluation of physical characteristics, doctor-patient communication, blood type and genetic matching. There are many issues at stake in the use of facial-matching algorithms in assisted reproduction, which this article seeks to address in a multi-disciplinary way. These issues include the impact of algorithmic automation of previous practices of matching faces in assisted reproduction; the racial dimension of the science and technology claims made regarding the phenotype by the promoters of facial matching algorithms, including whether such claims can be considered accurate; and how new Artificial Intelligence (AI)-based reproductive software slots into and transforms the commodification of reproduction and the expansion of fertility markets, already well underway.

The feminist approach to the world of reproductive technologies since the 1980s has long addressed the social relations and social forms that reproductive technologies themselves reproduce. It has often been observed in scholarship, for example, that ARTs are capable of ‘biologizing’ kinship precisely in the context in which this biological understanding of family is under threat through the use of donated reproductive cells (Strathern 1992, Franklin 2013). Broadly, the facial-matching algorithm promotes this biological understanding of kinship. Specifically, where a biological connection between parent and child today may designate the sharing of bodily fluids during pregnancy, a genetic connection points to the transmission of genetic information to offspring. And where biogenetics is promoted in contemporary egg freezing practices (as cryopreserving egg cells promises to preserve people’s capacity to give birth to genetically related children), the facial-matching algorithm rather emphasizes a phenotypical connection and promises to reproduce this even in the absence of the transmission of the genetic information necessary to unfold phenotypical relatedness. The concept of the phenotype was first introduced by Swedish botanist Wilhelm Johannsen in 1911 as part of his phenotype/genotype relation, which distinguished between an organism’s hereditary material and what that hereditary material produces in terms of observable traits (‘pheno’ in Greek means observe). The phenotype has long retained a conceptual flexibility within the field of biology, sometimes referring to observable entities, such as the skin, eye colour, hair, blood type, facial features and dis/abilities, and sometimes referring to the non-observable aspects of genes. This has enabled its mobilization by commercial and scientific projects with widely diverse claims and goals. The point is not to reject all current bio/sciences of the phenotype-genotype relation, but to investigate how shifting conceptions of the phenotype are, as anthropologists M’Charek and Schramm (2020, p. 322) suggest, ‘inevitably wrapped up in European or Western societies as they produce and are the product of race and racism’. They argue for an urgent analysis of the ‘work the face does’, including the social relations it generates, commenting that,

while scholarship on the body—for example, in anthropology, science and technology studies (STS), or feminist and postcolonial studies—has taken firm shape […] the critical turn to the face has not yet been made (p.321).

Such critical turns to the face and face-making have so far focused on forensic DNA phenotyping (M’charek 2020); archaeological reconstructions of the face using DNA (Nieves Delgado 2020); taxonomies of the face within nineteenth century police and psychiatry practices, when photographs of people’s faces were presented as evidence of madness, sexual deviance and criminality (Sekula 1986, Didi-Huberman and Charcot 2003); and contemporary art practices that have reappropriated the self-portrait form as a site of dissident physiognomy (Preciado 2014). While the racial politics of selecting donors in assisted reproduction based on explicit racial categories of identification has been amply addressed (Russell 2018), the racial politics of matching faces, algorithmic or otherwise, in assisted reproduction remains understudied.

Marxist Feminism has long attended to the commodification of reproduction, with one strand of scholarship addressing the incorporation of reproductive processes into the realm of direct accumulation under neoliberalism since the 1970s (Waldby and Cooper 2014) and another strand emphasizing how capitalist production always depends on the expropriation of myriad forms of waged and unwaged reproductive labour and life (Davis 1983, Vogel 1983, Gimenez 2018, Federici 2020). Marxist Feminist scholarship on ARTs has combined these two strands to attend to both the productive and reproductive dimensions of fertility and has detailed how gender and racial oppressions intersect with labour exploitation across a globalized market today (Vertommen and Barbagallo 2022). Recent scholarship has stressed the fertility market’s ‘financialization’ and has explained how the commodification of reproduction increasingly intersects with the more volatile dynamics of financial capital (van de Wiel 2020). This article attends to the fertility industry as a financialized industry and details the local and global actors participating in the facial-matching algorithm boom. Combining these insights on the business of kinship this research article asks both, what ideas and beliefs about reproduction does the facial-matching algorithm promote, and what is the symbolic and material impact of its uptake by fertility clinics?

National assisted reproduction legislation, local experimental cultures of biomedical innovation and European ‘reproductive tourism’ patterns have conformed Spain as a leader in the European ARTs market. In this competitive context, investing in new computational technologies and networked services can be seen as a market strategy. I make an argument for analysing fertility company’s incorporation of reproductive software as fixed capital investments. That is, to properly explain and situate the development of algorithmic phenotypical selection in assisted reproduction, it is vital to consider the role of the algorithm as an example of fixed capital: a digital machine that lies at the centre of the structural territorial logics of capitalist expansion. This is not a new approach but one amply addressed in contemporary scholarship on fixed capital, a political economy concept that today encompasses, on the one hand, the immoveable built environment of the city, including social infrastructures such as research and education (Jessop 2006, Eckers and Prudham 2018), and, on the other hand, what the late post-Workerist thinker Antonio Negri called the ‘digital machine’ (2019): the software-hardware of algorithms, cables, computers and servers (Greene and Joseph 2015). Fixed capital has firmly left the factory and fertilizer of Marx’s original writings on fixed capital and moved into PCs, smartphones, domestic appliances and city centres. David Harvey (2001) explains how capitalism could not survive without being geographically expansionary and perpetually seeking out what he calls ‘spatial fixes’ for its problems, with the development of new transport and communication technologies constituting such fixes. The global fertility industry is exemplary in this respect as it not only depends on spatialized fixed capital arrangements to function (the clustering of fertility clinics and banks in the highest rent paying areas of capital cities; the fashioning of entire countries such as Spain as destinations for fertility treatments; the development of transport networks necessary to safely move biological tissue across borders; and the design of baroque fertility and donor catalogue websites to capture and manage consumers’ attention) but the creation of new fixed capital technologies (new pharma, new diagnostics and IVF extras) is central to the business model of competing companies.

The promotion of the phenotype as a software solution for individuals to ‘opt in for’ plays into social anxieties that are profoundly racial in two ways. Firstly, this purely biological conception of family has been widely addressed in scholarship as a hallmark of a Western and European racial sensibility and scientific gaze (Strathern 1992, Haraway 1997, Franklin 2013). As sociology of knowledge and Black Feminism scholar Patricia Hill Collins summarizes, the eugenics movements ‘typically transformed social issues such as crime or poverty into medical or biological issues’ (2000, p. 269). In this fundamental sense, race stands at the centre of the facial-matching algorithm’s proposal that some aspect of ‘biology’ may be considered a ‘solution’ to societal issues such as the reconfiguration of kinship. The use of ARTs and donated reproductive cells is tied up in societal shifts, such as the legalizing of LGBTQIA++ parenting in some countries and delayed parenthood trends. Due to diverse factors, including job insecurity and economic uncertainty, many people are waiting longer to start families. With age, the quantity and quality of reproductive cells can decline, a main motivation for many to turn to reproductive cell donation. In this ‘infertility crisis’ context, as much social as biological, the facial-matching algorithm mitigates against the symbolic threat of the participation of third parties in the reproductive process and allows the reproduction of the dual-parent led heteronormative family model to be a main outcome of the use of ARTs. By detailing how the facial-matching algorithm insists on the ‘phenotype as fix’, this article shows how the historical combination of biological determinism with the belief in biology as a technical fix persists well into the present.

Secondly, this rhetorical denial of the participation of third parties in the reproductive process reproduces, while also disguising, the racial and class hierarchies that the business of assisted reproduction relies on. As further elaborated on below, the availability of donated egg cells in Spain depends on a number of structural income disparities and racialised class dynamics. The facial-matching algorithm’s emphasis on phenotypical continuity distracts not only from the role of donors in creating families but the exploitation that underlies the availability of their cells on the Spanish fertility market. Furthermore, this rhetorical denial is implicated in a broader social denial of familial configurations that deviate from the dual-parent led model. As trans*feminist² and Reproductive Justice scholars have noted, prevalent among working class families battling economic precarity, and/or the pressures of migratory control system protocols and state racism, are networked familial bonds not based on biology but grounded in resource sharing across households (Ross and Solinger 2017, María García 2023). Throughout this article I bring up this critical comparison between the ‘consumer opt-ins’ promoted by the commercial worlds of fertility, and the more abundant feminist and antiracist notions of ‘reproductive choice’. This comparison does not pretend to be empirically evidenced but rather serves as a possible ethical provocation for critiquing the hierarchies of value and visibility that symbolically or materially award some familial configurations over others.³

My concept of ‘the fertility fix’ in this article attends to the facial-matching algorithm as both a software fix that socially reproduces a particular racial sensibility regarding family as a biological entity and a fixed capital investment made by clinics and companies looking to expand and survive in a market context. At stake ultimately is not the dubiousness of the technology or necessarily that companies are morally wrong to innovate in fixed capital. Rather, the fertility fix as a lens emphasizes the territorial power-orientated politics of both fixed capital formation and social reproduction trends and pursues a critical, ethical reading of emergent computational and networked landscapes of reproduction.

The web continues to be a grey area for copyright, as well as for researchers using big data, digital data or informational material sourced online. The question of what counts as public versus private (Lester 2020) and related concerns regarding the discursive power of framing information as ‘data’ in machine learning practices (Ciston 2023) remain at the centre of contemporary discussions around researching ethically after and on the web. This article references commercial entities who have been informed of their being under study. Not assuming commercial websites or advertised software belong to the public because they are in the public realm, I make descriptions. The below descriptions result from the combined analysis of software, fertility clinic and bank websites, assisted reproduction policy, law and recent scientific publications on the DNA-face relation. I also make use of web archiving sites like WayBack Machine, which offer a uniform mode for preserving screenshots of online content. Both web archiving and web research are conditioned by partiality, as the sources cannot be too far separated from the ways in which they have been reassembled by search engines (Brügger and Milligan 2019). The gathering of this article’s research materials in and through the online research process aligns with an important hypothesis of this article: that websites and algorithms are sites of historical and material struggle conditioned as much by economic forces and human motivations as by the automated arbitrary.

The fertility fix: biology as fix

Scholars of assisted reproduction have already addressed the way in which the drop down ‘biogenetic menu’ of race and ethnicity in assisted reproduction simultaneously reproduces and depoliticizes race (Russell 2018); and have analysed identity production at the level of the personal computer interface (Gajjala et al. 2008) or how ‘race happens on the web in ways that are unique to the medium of hypertext and web menus’ (Nakamura 2002, p.111). This article contributes to these critiques of the racial politics of the Internet interface by focusing on a case where racial classifications are not serialized or even mentioned, but where the racial politics of biology as fix is nevertheless at play. Fenomatch, a company registered in 2019 by the Spanish trade office, describes how their algorithm ‘scans more than 12,000 data points in order to help find the donor with the greatest biometric similarity’, comparing passport-sized photographs translated into mapped data-points (Fenomatch 2022). The company’s website is populated by images of women’s faces segmented into geometric shapes traced by shining white lines. In one a woman looks down at her phone as white lines pour out of her eyes, cheeks and nose onto the phone’s screen. In another, a woman looks at a child, their faces connected in a constellation of white lines and dots. The Fenomatch logo bears the twisted braid of the double helix structure of deoxyribonucleic acid (DNA), discovered in the 1950s giving rise to modern molecular biology. DNA is composed of four bases (known as A, C, G and T) whose unique sequence determines the production of different molecules in the body. DNA constitutes the chromosomes present in the nucleus of every cell and so is fundamental for prescribing what phenotypical traits are passed on to the next generation. The link that the Fenomatch logo establishes between DNA and phenotype is both logical and vague: logical because different phenotypes may be the expression of differences at the level of DNA sequencing, but vague because it is widely accepted that phenotypes are equally, if not more greatly, influenced by environmental and social conditions. As DNA is understood to constitute genetic, and thus inherited, material, it is not surprising it appears symbolized as part of the logo of this company, which promises to align the DNA-related phenotypical expressions of gamete donors with those of the prospective parents. At the same time, the fact that phenotype can rarely be linked only to DNA, makes any promise to be able to select phenotypical expressions unreliable. The extent to which DNA actively makes the face, and the extent to which the face can be considered a reliable aspect of phenotypical expression, as eye colour is thought to be, is a site of considerable speculation well into the present. Fenomatch (2021) recognize the ambiguities of linking DNA to face. Their website, papers and talks correctly situate the phenotype in relation to extra-genetic social conditioning and epigenetics, which are the heritable changes in gene expression that do not respond to the information coded in the DNA sequence. The DNA-face relation that the company’s marketing visuals and logo establish is then not so much unscientific as a particular exploitation of the flexibility of the post-DNA reproductive imaginary of the genotype/phenotype relation/distinction. It is through the conceptual flexibility of the phenotype that social kinship configurations too are imagined to be biological and controllable through practices of selection.

The human face as a locus of selection and control in assisted reproduction is not exactly new. A quick look at 1990s donor catalogue websites and one observes that the promise to match faces – the use of photos to select donors based on facial features – is not specific to machine learning: California Cryobank stated on their website as early as 1997 that they offered a ‘photographic donor matching consultation to help with donor selection’ (California Cryobank 1997). By the mid-2000s many donor catalogues were offering childhood photos of the faces of their donors. The use of facial-matching algorithms today thus provides a moment of visibility for the processes through which the face was already at work in assisted reproduction. Perhaps this desire to select faces is to be expected: the fact that children should look like their parents is widely appreciated and looking alike is an understandable site of emotional investment for families. On the other hand, today, as ever, societies are upheld by familial networks whose members may look nothing like each other. The facial-matching algorithms’ promotion of the family as a biological entity presents a significant delimiting of kinship connections as they have been elaborated across the distinct geographies of the feminist reproductive rights and justice movements, including the historically marginalized forms of ‘queer and trans* social reproduction’ theorized by Nat Raha (2021); the ‘trans non-monogamies’ that challenge Spain’s bureaucratic structures for legally recognizing households theorized by Rosa María García (2023); the LGBTQIA++ ‘chosen families’ empirically documented by Kath Weston (1991); and the Reproductive Justice movement’s abundant and strategically anti-moral definition of reproductive choice as ‘the human right to maintain personal bodily autonomy, have children, not have children, and parent the children we have in safe and sustainable communities’ (Ross and Solinger 2017, Price 2021). While race and racial identifications are key sites of politicization, self- and collective identity formation and affective connection across these movements, also central is the cultivation, in theory and practice, of reproductive and care networks that may include but do not depend on biologized kinship.

Fenomatch are keen to distinguish their algorithm from others on the Spanish market, such as Ovo clinic’s ovomatch, which is based on the Amazon Rekognition API. Indeed, facial-matching technology, as opposed to recognition, emerged uniquely through the fusing of facial recognition and image similarity research within the field of computer vision. Face recognition dates to the 1960s (Bledsoe 1966) and consists of two main tasks: face identification (locating the identity of a person using a photograph of their face) and face verification (checking if two images of faces refer to the same person). By the 1990s researchers began to approach the face less as a geometric entity defined by isolated features – the nose, mouth and eyes – and more as a two-dimensional surface: a ‘projection’ onto a feature space, a process standardized with the use of ‘eigenvalues’, derived from a set of training faces called the EigenFaces (Turk and Pentland 1991). The arrival of neural networks and deep learning combined with the availability of big data sets towards the late-2000s marked a new age for facial recognition. Image similarity research is, on the other hand, rooted in image retrieval practices and grew exponentially in scale with the release of the Mosaic Internet-browser and the availability of digital vision sensors. Today most facial recognition algorithms use a low-dimensional space into which high-dimensional vectors are translated, called an embedding space (Koehrsen 2018). Facial recognition plots a face’s data points into the embedding space in such a way that the distance between different faces is great. Matching algorithms, on the other hand, plot into an embedding space in such a way that the space between similar pairs is small. Fenomatch (Meléndez et al. 2021, p. 2) explain this in a recent article, stating how facial recognition optimizes characteristics that differentiate one from the other. A birthmark or a tattoo, for example, will give a high weight to these traits to differentiate. The optimization process of the Fenomatch algorithm is supposedly weighted towards matching traits to achieve, as they say, ‘maximum similarity’. Yet, to create a set of statistical calculations to measure similarity, required for a facial-matching algorithm trial and error optimization process, one must be able to define the rules of similarity. As many researchers have noted, the geometry of the human face is far from straightforward and human-specific neural processing for faces differs from that of other objects (Hadders-Algra 2022). Furthermore, any matches would be impossible to verify as they are based on highly subjective markers. Today, the algorithms used for facial recognition tasks are often trained with normative identity as categories, for example, gender, race and sexuality. This contradiction is clear when Fenomatch state proudly, despite the constant appraisal of the accuracy and efficiency of their algorithm, that the final selection of the donor is ‘still made by a human’ (Fenomatch 2021). In the end, it turns out, evaluating facial similarity is a problem not for computer-vision but human-vision. The double helix of the Fenomatch logo as it combines with the datafied faces of mother and child promises biological security in a context in which this understanding of family is under threat. This is the first sense in which the facial-matching algorithm acts as a ‘fertility fix’: a biological fix to the severing of phenotypical links that the use of ARTs may represent for some fertility patients.

Fenomatch are not alone in bringing up the genotype/phenotype distinction/relation through new algorithmic technologies. A flurry of papers on the use of AI tools that link DNA to face have been published in the last decade. These articles include studies of ‘the genetics of the human face’ in an article that quotes the Victorian forensic scientist and eugenicist Francis Galton as a ‘pioneer’ (Crouch et al. 2018, p. 676); algorithms that ‘identify genetic syndromes using deep learning such as DeepGesalt’, presented as a ‘next generation phenotyping technology’ (Gurovich et al. 2018) and experiments that claim to be able to predict people’s sexuality by algorithmic evaluation of photographs of their faces (Skorska et al. 2015). Many of these studies result from problematic practices of data gathering, from the use of racialized data cohorts to train the machine-learning algorithms to the possible lack of informed consent from participants (Nieves Delgado 2020). In the study by Skorska et al. (2015, p.1377) of the face and sexuality, the authors used a ‘facial modelling program trained on photographs of white faces’, which they gathered by ‘approaching people at pride events’. The results of this experiment include the ‘findings’ that ‘lesbian women had noses that were more turned up, mouths that were more puckered, and […]’ (p.1378). Another experiment conducted at Stanford University used deep learning neural networks to ‘predict sexual orientation’ using a data cohort of 35,326 facial images (Wang and Kosinski 2018, p. 246), concluding that:

Consistent with the prenatal hormone theory of sexual orientation, gay men and women tended to have gender-atypical facial morphology, expression, and grooming styles … Given a single facial image, a classifier could correctly distinguish between gay and heterosexual men in 81% of cases, and in 74% of cases for women.

Across these studies, the face works to establish links between antiquated and contemporary theories of racial, sexual, gender and ability difference: here morphological, there genetic, here neurological, there hormonal. By exploiting the historically elastic relations between genotype and phenotype, the facial-matching algorithm relies on the same ‘eclectic’ scientific research framework that drives the homophobic, racist and ablest ‘studies’ mentioned above.

According to Fenomatch, over one hundred clinics worldwide subscribe to their software, (Egg Donation Friends 2021), which can be found advertised on the clinic’s respective websites. It is at this level of the Internet interface where the ‘opt-in’ logic of biological control in fertility meets the computational logic of ‘direct manipulation’. The term ‘direct manipulation’ was coined in the early 1980s and refers to the interaction style for graphical user interfaces (GUIs) exemplified in the desktop metaphor: one organizes digital files as one organizes one’s desk, by directly moving things around (Shneiderman 1983). This visual-based interaction replaced the previous instruction-based command line. Direct manipulation as a term describes well the astounding subjective effects of causing ‘immediate’ change to the interface via the most motion-restricted bodily activities: finger tapping and thumb wriggling extend in heroic narratives that while virtual are very much real to those clicking (Manovich 2002). In a national context conditioned by a certain patient powerlessness with regards to donor choice, opting in for the facial-matching algorithm at the level of the interface enables the powerful subjective experience of ‘directly manipulating’ kinship connections as a biological entity. In this specifically software sense, the facial-matching algorithm fits well current porous definitions of ‘add-on’ IVF technologies that ‘craft hope’ in fertility contexts (Perrotta and Hamper 2021). That the facial-matching algorithm does not explicitly mention racial categories of identification does not mean that race and racial politics are not at stake in these choices. As Russell (2018 p.154) notes, the persistent focus of assisted reproductive technologies on ‘matching physical appearance participates in depoliticizing racial identity, […] severing contemporary decisions from their historical contexts’. Add-on technologies like the facial-matching algorithm excite the commercial possibilities of an ingrained racial sensibility regarding both the value of phenotypical lineage to kinship and the broader framing of biology as a technical fix to societal transformations.

The fertility fix: fixed capital

While Europe constitutes the global fertility sector’s biggest continental market, the legislation, affordability and norms surrounding family-making vary widely across the continent. This ‘patchwork of permissivity and prohibition’ has resulted in Spain becoming the ‘most active country in assisted reproduction’ (De Geyter et al. 2018). In the last two decades, Spain’s assisted reproduction market has come to be associated with cheap rounds of IVF and a commercial landscape populated by experimental culture of commercial biomedical innovation. This context is key for situating the facial-matching boom. Assisted reproduction is regulated in Spain by a 2005 law, which allows single women and LGBTQIA++ people to access ARTs either privately or through the state funded public health system. Spain is also a priority destination for individuals or couples seeking fertility treatments that require donated eggs, because over 40% of all eggs used in ART procedures in Europe are donated in Spain (Lafuente-funes et al. 2023). While the law is keen to link egg donation to ‘altruism’ – with clinic marketing material also asserting ‘the special empathy of Spanish women’ (IVI Fertility Clinic 2021) – many studies have rather situated the high numbers of egg donation in relation to the 2008 financial crisis, when people turned to gamete donation as a form of ‘informal service labour’ (Alkorta 2021 p. 223). Many ART procedures performed in Spanish clinics are performed on patients who travel to Spain for treatment, which explains why the majority of Spanish fertility websites are multilingual. While many use the term ‘cross-border care’ (Pennings 2005) to describe these journeys because it is both more ‘descriptive and neutral’ (Präg and Mills 2017), the term ‘reproductive tourism’ more accurately portrays the way in which these journeys map atop already existing structures of re/production that order North–South relations in Europe. Countries like Spain and Greece were encouraged to replace their manufacturing-centred economy with construction and services, such as tourism, as part of their entry requirements into the European Union in the mid-1980s. In the context of the European ART market, these North–South and North-East relations within the European Union have contributed to cementing Spain and Ukraine as gamete and surrogacy provider countries. The term ‘reproductive tourism’ better visibilizes these continental market dynamics where certain countries are fashioned as sites of production and manufacturing and others as sites of leisure and consumption. These conditions lie behind Spain becoming a ‘reproductive tourism hub’ since the mid-2000s, the ideal environment for fertility add-on start-ups such as Fenomatch. Clinics like Instituto Marquez, in Barcelona, exemplify this experimental culture of biomedical innovation at the intersection of reproductive science, health, medicine, research and software development. Their website displays adverts for add-on fertility technologies, including a smart-phone accessible live embryo surveillance (a tool called Embryoscope), options for algorithmic embryo selection, a baby pod (an ipod like speaker that plays music into the womb) and a multimedia high-definition screen in a masturbation booth for sperm donors (Instituto Marquez 2021a). The Director of the clinic, Dr. Marisa López-Teijón, was awarded the Ig Nobel Prize for Medicine presented in 2017 at MIT, for experiments in ‘discovering how foetal hearing works’ and for organizing ‘live concerts for embryos’, including a recent visit to the lab by pop-rock band The Corrs (Instituto Marquez 2021b). This article approaches these emergent computational and networked physical and social fertility infrastructures as the result of fixed capital investments.

The concept of fixed capital first appeared in the eighteenth century writings of David Ricardo, for whom fixed capital meant any physical asset not used up in the production process. Marx’s writings on fixed capital were first introduced in the Grundrisse: Foundations of the critique of political economy (1973) and elaborated on across Chapters 7–11 of Capital, Volume II (1978). There Marx amends what he considers to be a major flaw of Ricardo and Adam Smith’s political economy theories, adding a distinction between fixed and fluid capital to the constant/variable and fixed/circulating distinctions. In the case of both variable capital (labour power) and constant capital (raw or auxiliary materials), value is mostly passed on to the product in ways that are used up ‘immediately’ (Marx 1978, p. 237). Yet there are some cases of constant capital where a portion of the means of labour, though ‘fixed’ in the productive sphere, also distributes its value ‘over a longer time’ (p. 238). This complicates any easy distinction between ‘circulating’ labour power and ‘fixed’ raw or auxiliary materials, as physical infrastructure, Marx points out here, also appears to flow. The examples of fixed capital given in these passages are machinery, an ox and fertilizer. The distinction between fluid and fixed emphasizes the central but temporally complex role that biology (eg. the ox) and technology (eg. the machinery) – and their many intersections (eg. the fertilizer) – can play in accumulation.

The role of fixed capital in accumulation is further detailed in Marx’s discussion of simple and expanded reproduction in Chapters 20 and 21 of Capital Vol. II (1978). Simple capitalist reproduction, Marx (1978, p. 471) explains, is the effect of a dynamic interaction between two departments: Department I is the means of production (encompassing variable and constant capital) and Department II is the means of consumption, which includes all the subsistence products necessary to reproduce labour-power (from food to luxuries). For capitalist reproduction to proceed in a balanced way, workers must spend all their wages on the subsistence products produced in Department II, and capitalists must also spend their profits there too. Yet expanded reproduction requires hoards (money), and so the capitalists stop consuming, which tilts the circuit off balance. This leads to crisis: a crisis in hoarding in Department I and a crisis in over-production in Department II. David Harvey (2003, p.151) calls this crisis ‘overaccumulation’, which capitalists respond to by commodifying and devaluing social reproduction assets – from land to labour-power – that are ‘thrown out of circulation’ and then ‘lie fallow and dormant until surplus capital seizes upon them to breathe new life into capital accumulation’. One supposed fix for the capitalist in times of crisis is to expand the Departments by investing in productive infrastructure; new technologies of communication, transport, education or research. According to Marx’s theory of accumulation, and further observed in Harvey’s writings, fixed capital stands at the absolute centre of capitalism’s crisis tendencies. This is because while investing in new physical and social infrastructure may temporarily absorb hoards and support new production across the departments, this is only a quick fix. Harvey insists it is never long before the cycle repeats, and hoarding and devaluation return to the Departments. The point of Marx’s neat diagram in Volume II of Capital is to highlight that when expanded, the reproduction of capital is never neat or ‘simple’, never proceeds in a balanced way.

Capitalism’s crisis’ of (over)accumulation are evident in the fertility industry, especially vulnerable to capital flow blockages, value realization issues and, in the case of egg donation labour, wage stagnation and suppression. Between the unfounded pretensions of future-orientated marketing claims of biomedicine (Rajan 2006) and the reliance of the market on volatile private equity, fertility companies are finding it difficult to realize their steep projected profits. The speculatory business logic of financial capital further drives dispossession on the ground as debt dynamics push firms towards the extraction of the promissory value pre-assigned to investors and shareholders elsewhere, with disruptive consequences for local and global re/productive labour ecologies and habitats (Harvey 2003, Sassen 2016). These territorial stasis/crisis dynamics can be observed in some of the financial transactions that have sustained the facial-matching algorithm boom in Spain in the last few years.

Soon after emerging on the fertility market in 2019, Fenomatch was acquired and integrated into the services offered by the Spanish chain IVI Fertility (Escobar 2019). Two years previously, IVI had merged with U.S company RMANJI to create IVI-RMA. With more than 75 clinics and 28 labs in 9 countries, 35 of these clinics in Spain, IVI-RMA are the largest assisted reproduction group in the world. In 2022 the international fund KKR acquired IVI-RMA, beating other funds such as Amulet and Nordic Capital in a competitive bid presided over by Bank of America, Credit Suisse and Deutsche Bank, with Morgan Stanley advising the Spanish owners of IVI-RMA (Bayón 2023). In 2024, the Spain based IVI-RMA clinics are advertising their own branded Perfect Match Phenotype 360° facial-matching algorithm. Spain’s fertility sector first began to show signs of financialization in the early 2010s, when Luarmia S.L, who hold the global operations of Barcelona based fertility clinic chain Eugin, was sold to Spanish private equity company ProA Capital for 75 million euros. Luarmia was sold again in 2014 for nearly double, at 143 million euros, to NMC, a stock listed health corporate based in the United Arab Emirates. The ‘volatility’ of finance referred to throughout this article is evidenced in these kinds of rapid-fire fertility capital profit hikes and dips: in 2020 NMC was placed in administration by the Federal Credit Administration for misleading the market about its debts, but not before they could sell the share capital of Luarmia SL to Fresenius Helios, a German multinational. The stories of Luarmia and IVI-RMA come together in 2023, when KKR also acquired Luarmia and its operations of the Eugin clinics. The fact that KKR now owned both IVI-RMA and Eugin, while also owning GeneraLife, who operate the Sevillian fertility clinic chain Ginemed, caused alarm at Spain’s Comisión Nacional de los Mercados y la competencia, who regulate against the monopolizing of markets (Bayón 2023). Not only are fertility companies competing with monopoly firms, but fertility patients are also becoming directly ‘entangled with debt accruement institutions’ in new ways through IVF-related subscription plans (van de Wiel 2020). Fenomatch facial-matching algorithm, for example, forms part of a fertility treatment system offered by Spanish fertility company La Unidad de Reproducción Vistahermosa. Called ‘SecureFIV’, the patient pack includes IVF procedures, preimplantation genetic testing and the algorithmic donor selection, accessible via a financing plan of up to 60 months (UR Vistahermosa 2021). The facial-matching algorithm boom in Spain emerges at this intersection of the financialising of fertility capital and the financializing of fertility patient options. In this competitive context, clinics and companies increasingly secure fixed capital capabilities, investing in new tech to develop their competitive edge while warding off value realization issues. The ‘fertility fix’ is situated at this juncture of symbolic and material fixes.

The fertility fix: biology as fix and fixed capital

This article has situated the facial-matching algorithm boom in Spain as a phenomenon that intersects productive and reproductive, economic and non-economic spheres, stimulating intimate desires for kinship connections and capitalist drives towards profit and expansion. Post-Workerist conceptions of fixed capital have concentrated on Marx’s earlier reflections on fixed capital as the outcome of collective social processes and communal knowledge making. In the Grundrisse Marx (1973, p. 606) notes that while such ‘machines’ are the direct product of the waged labourers who built them in factories and warehouses, fixed capital is also the indirect ‘organ of the human brain, created by the human hand, the power of knowledge objectified’. This insight powers the political project of ‘reappropriating fixed capital’. For, ‘what is an algorithm?’, asks Negri (2019, p. 209), if not ‘fixed capital, being a machine born of cooperative social intelligence.’ The present article’s ‘fertility fix’ concept departed from an earlier moment in the Marxian fixed capital argument, drawing on the explanatory force of fixed capital as a concept useful for considering why the drive towards the creation of new physical and social infrastructures is so central to capital accumulation, and ultimately crisis. This was key to explaining the ‘innovation’ dynamics at play in the emerging computational and networked worlds of fertility and their geographic concentration in Spain as a fertility destination. As suggested, some of the assumptions implicit in the application of facial-matching algorithms for donor selection are not new at all, and rather represent a continuation of historical scientific practices that have combined biological determinism with the belief in biology as a technical fix. I further elaborated on the racial dimensions of the facial-matching algorithm’s science and technology claims regarding the phenotype while also attending to how such reproductive software slots into the commodification and financialization dynamics already at play in assisted reproduction. This article has highlighted in particular the central role that the screen, either fertility patient’s screen (where they may access the websites where facial-matching algorithms are advertised) or clinician’s computer screen (where the algorithm may be integrated) plays in automating not only donor selection processes but certain reproductive rationales. The outcomes of opting in for algorithmic facial-matching, this article suggests, is a kind of screen-based software produced ignorance that invests in biology as a technical fix to societal transformations to kinship; invisibilizes the labour conditions that underlie the availability of reproductive cells in assisted reproduction; and participates in the broader social and symbolic denial of myriad familial forms.

The visual cultures of contemporary fertility can trigger bad feelings. This article moves beyond, though, a straightforward dismissal of marketing representations as somehow missing the mark where another image or narrative could get it right. As software studies scholar Wendy Chun (2011 p.59) observes, rather than condemning the visual cultures of interfaces ‘as forms of deception, designed to induce false consciousness’, the task is to investigate software’s ‘paradoxical combination of visibility and invisibility, rational causation and profound ignorance’. This software studies (Fuller 2008) troubling of the binary distinction between intelligence and ignorance, visibility and invisibility, material and immaterial, coincides with a trans*feminist politics that conceives of the human body and human reproduction not as pre-existing ‘natural’ entities and processes awaiting ‘technological’ modification, but as the always already material-symbolic artificial-natural effect of vertically stacked and networked medical, social and political technologies of the body (Preciado 2017). At stake in this critique of the facial-matching algorithm boom ultimately is not that companies are intentionally duping clients, nor necessarily that firms are morally wrong to seek out fixed capital solutions to value realization issues. Rather, the fertility fix concept has emphasized the territorial power-orientated aspects of fertility capital accumulation and the computational and networked configuration of contemporary reproductive choice.

Disclosure statement

No potential conflict of interest was reported by the author(s). Funding has also been received from the Irish Research Council, the Health Research Board, the Environmental Protection Agency and the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 101034345.

Additional information

Funding

This work was supported by Koneen Säätiö: [grant number 202010758].

Notes

1 According to the International Committee for Monitoring Assisted Reproductive Technology and the World Health Organization, Assisted Reproductive Technologies (ARTs) ‘includes the in vitro handling of both human oocytes and sperm, or embryos for the purpose of establishing pregnancy’ (Zegers-Hochschild et al. 2009, p. 2685).

2 I use ‘transfeminism’ interchangeably throughout the article with trans*feminism and trans*, following the adoption of the asterisk in English speaking contexts by many activists and scholars since the early 2010s (Raha 2021, Cardenas 2022).

3 My doctoral thesis Post-Internet Queer Reproductive Work and the Fixed Capital of Fertility (2023) attends more amply to the structural relationship between historical forms of ‘queer reproductive choice’ and todays heteronormative global fertility market today. Though beyond the reach of the present article, I further elaborate the Queer Marxist perspective on reproductive software initially developed there.