The way we relate to our bodies and their representation is heavily influenced by a diversity of elements, including the visual arts.14 While the female body has played a central role in art throughout history, its portrayal has largely been shaped by male artists. Thus, women have often been reduced to symbols of male privilege and power rather than recognized as the creators of culture.15 Consequently, women have been objectified in creative expressions, leading to male-centered interpretations of the history of art. Female artists have been systematically denied proper recognition for their work—e.g., Marietta Robusti (1560-1590)—, de-valued and underestimated by their contemporaries—e.g., Judith Leyster (1609-1660)—, exoticized as exceptions—e.g., Edmonia Lewis (1844-1907)—, constantly forced into a “feminine” lens through their art—e.g., Camille Claudel (1864-1943)—or, conversely associated with masculine adjectives as a way to appreciate and justify their success—e.g., Suzanne Valadon (1865-1938) and Paula Modersohn-Becker (1876-1907). The representations and interpretations of art have always been strongly gendered, with the very ideal of the femininity of the Renaissance woman starting as early as Vasari’s work16 and continuing later in the history of visual arts.17

In 1975, Mulvey analyzed the concept of the “male gaze” in Hollywood movies,18 according to which women are depicted as objects of sexual desire from a heterosexual male perspective. This perspective has become even more pervasive with the rise of social media19 and the broad use of algorithms. In fact, scholars have argued that the amplification of the male gaze is a result of the gender inequality in the environments where the algorithms are conceived, developed, deployed, and legislated.20 Such amplification is evident in the censorship of female artistic nudity, where female bodies are often perceived as solely objects of pleasure and pornographic content.21 Consequently, artists who work with female nudity to challenge the patriarchal sexualization of women face stronger restrictions on their freedom of expression, both because more people report them and because the algorithms are biased.

This systematic misinterpretation and silencing of de-sexualized representations of female bodies is an important phenomenon to reflect upon, study, and address when considering the impact of machine learning algorithms on our cultural and artistic ecosystems.

Relying on algorithms for content moderation undoubtedly allows the platforms to obtain significant economic gains in the form of savings,22 as automation allows them to scale-up to levels that would not be possible if they only relied on human moderators.23 In fact, with billions of users and millions of pieces of content uploaded every day, it is impossible for human moderators to perform this task alone.

Beyond this economic impact, we need to consider that social media and related online platforms are owned by private companies with a business model that largely depends on the monetization of the content that they host. Their ultimate goal is evidently to create an economic gain. Thus, it should not be surprising that social media companies keep whitelists of users for whom content moderation is not applied, or is applied differently.24 Such users are typically high-value members of the network (e.g., influencers), with hundreds of thousands or millions of followers so that their uploaded content generates significant traffic and revenue for the platform owners. This differential treatment has a direct impact on the type of content that is likely to be uploaded as its visibility and popularity will depend on how well it adheres to the mainstream aesthetics accepted on the platform, to the detriment of minority voices and perspectives which are less popular and “economically interesting.”25 Profit-driven content moderation decisions tend to prioritize content that aligns with dominant cultural norms, reinforcing existing power structures and contributing to perpetuating gender stereotypes and biases. Such a reinforcement is detrimental to the techno-feminist goal of challenging and dismantling patriarchal structures and amplifying diverse perspectives.

Furthermore, the lack of transparency regarding the use and inner workings of proprietary content moderation algorithms in social platforms hinders the users’ understanding of the reasons behind the censorship they experience. Thus, engaging in a public discourse about the legitimacy of this censorship becomes challenging. According to Riccio et al. (2022), content moderation algorithms play the role of an invisible critique in today’s online cultural ecosystem as they not only moderate content but also determine what is seen (and not seen), boosted, promoted, or de-prioritized. In essence, algorithms dictate what aesthetic qualities are deemed acceptable or unacceptable within our socio-cultural context by pervasively influencing all the elements of the creative chain,26 with the market acting as the ultimate driving force in the decision-making process. Consequently, social media platforms establish distinct—albeit invisible—boundaries between what can be posted and what cannot. Within such an opaque framework, it is hard to understand what the rules of artistic expression are nowadays. This opacity is both intentional from the side of the social platforms and technical due to the inherent complexity of the deep neural networks that are at the core of content moderation algorithms. Furthermore, the lack of transparency leads to a lack of accountability as there are no clear mechanisms for challenging or appealing content moderation decisions.

Examining the consequences of algorithmic censorship demands acknowledging the weaknesses of these algorithms, which are susceptible to errors and biases. Many of today’s social media platforms do not explicitly forbid artistic nudity in their community guidelines.27 However, the discrepancy between these guidelines and the actual experience of censorship by many artists28 suggests that the algorithms might not be sophisticated enough to properly discern pornography from art,29 failing to understand the non-explicit intent and context related to the uploaded images.30 Contextualization is a prominent focus of research in the field of Computer Vision:31 Even basic tasks like object detection require explicit cues to contextualize the image.32 Humans, in contrast, rely on their past experiences and diverse memories when engaging with the real world.33 The ability to contextualize a naked body plays an important role in distinguishing between pornographic and artistic images. As a consequence, detecting porn, nudity, or NSFW (not-safe-for-work) content still represents a technical challenge.34 Existing research on algorithms developed for this purpose often overlooks the nuanced distinction between artistic nudes and pornography.35 Although the Computer Vision literature has explored various techniques for context integration,36 there is a lack of art-centric perspectives in the methods developed to date for automatic porn or adult content detection.

In addition, any decision-making process supported or performed by machine learning algorithms requires addressing potential algorithmic biases.37 Algorithmic biases or discrimination refers to the systematic and unfair treatment (discrimination) that can occur when developing, training, or deploying machine learning-based algorithms.38 These biases emerge at different points of the machine learning cycle, from the data used to train them to the design choices made during their development. As a consequence, the biases may be inadvertently embedded in the algorithms and influence their decision-making processes, leading to unequal or unfair treatment of certain individuals, groups or content (e.g., women or portrayal of female bodies). The machine learning algorithms that permeate today’s social media platforms thus automate culture through technology, replicating and possibly amplifying human biases.39 In the context of algorithmic censorship, these biases could imply that certain representations of the body are more likely to be accepted than others (exacerbating the economic incentives previously discussed), or that artists representing minorities get censored more often, as Nayyar (2023), Shapiro (2022), Faust (2017) claim.