AI and Gender
Body And Gender
Gendered Knowledge And AI
Gender Bias
Artificial Intelligence has a profound influence on the way individuals are perceived and treated in society. However, AI technologies embody and replicate limiting and confining ideas about gender and race, which are repeatedly reinforced. As technology embodies gender relations, the meanings and characteristics of masculinity and femininity, in turn, derive significance from their involvement and integration into working machines.
For decades, feminist scholars in science and technology have explored the interconnectedness of gender and technology. Experts have drawn a parallel between Alan Turing's assertion that a computer operates through imitation and the ability to learn----and the idea of gender as a learned and imitative behaviour so well processed that it appears natural.
The relationship between gender and technology can be further expanded by adopting a more intersectional approach. In essence, AI, the gender binary, and colonialism all share the common goal of essentializing, controlling, fixing, and establishing a hierarchy of identity. Gender itself plays a role in producing racial distinctions, functioning as a racializing apparatus. Both gender and race focus on socio-political relationships to perpetuate power structures and exert control over bodies.
Research in the field illustrates how AI technology functions as a tool for identity control. Using automated facial recognition, these systems claim to determine a person's gender by analysing photographs. These technologies operate based on gender and race recognition algorithms, relying on fixed notions of gender and race as systemic frameworks, ultimately leading to discriminatory outcomes.
Body And Gender
Currently, AI systems’ persistent reproduction of constrictive notions of gender is generally observed, as mentioned before, in terms of the embodiment of gender in artificial intelligence, the concept of gendered knowledge, and ultimately, gender bias.
When it comes to body and gender, the body is generally regarded as the site upon which the physiological gender norms and stereotypes are imprinted. The surface of the body is shaped by actions, gestures, and expressed desires, forming the illusion of an internal, organised gender identity. Simply put, specific appearances and ways of using the body are standardized to align with the definitions of 'male' and 'female.' Bodies conforming to culturally recognized boundaries, hold greater political significance than those that do not.
In this context, there's a noteworthy similarity between gender and technology. Just as society attributes political significance to gendered bodies, humanoid robots also hold a similar recognition. They signify economic well-being and serve as an indicator of technological progress. The perpetuation of conventional gender stereotypes in robotics contributes to the heightened political importance of both gender binary and artificial intelligence.
Humanoid robots also conform to these gendered frameworks, actively contributing to the reinforcement of gendered bodies and behaviours. New technology and media consistently echo traditional perceptions of the female body as illustrated by robots that emulate societal gender stereotypes through their appearance, voice, mannerisms, movements, and demeanour.
Sex robots, for instance, perpetuate conventional physical gender roles and contribute to the 'objectification' of gender-specific bodies.
Similarly, Sophia, the humanoid robot crafted by Hanson Robotics, closely resembles a stereotypical woman in a remarkably realistic manner. The representations of gendered bodies in robotics, particularly those resembling women, sustain and replicate stereotypical appearances. Furthermore, these entities are assigned political significance, as in the case of Sophia's acquisition of Saudi Arabian citizenship.
Moreover, gender imposes fixations on bodies through surveillance, placing them in specific temporal and spatial contexts, and framing bodies as needing correction by eliminating those that challenge the established gender order. These humanoid depictions not only reinforce stereotypes but also contribute to the removal of bodies that pose a threat to the established gendered hierarchy.
When it comes to Virtual Personal Assistants or VPAs, experts examine how they fail to challenge traditional male and female dichotomies. These VPAs perpetuate gender stereotypes through language and the process of naming. By naming them with phrases like 'hey Siri' or 'hey Alexa,' VPAs are essentially summoned into existence and a power dynamic is established through this act of naming. Studies draw a parallel between the VPA's lack of self-determined subjectivity and its role as a supporter and helper, echoing the notion of the female figure as a devoted aide to humanity. The association of the feminine voice with the language of assistance also raises ethical concerns. These gender stereotypes not only adversely impact society but also foster surveillance and further reinforce the domestication of the feminine identity through the promotion of digital domesticity.
Gendered Knowledge And AI
Historically, the construction and access to knowledge have been significantly influenced by gender dynamics, with men often occupying privileged positions in this regard. In many societies, women were traditionally excluded from formal education and academic institutions, limiting their access to learning and knowledge creation. Educational opportunities were frequently reserved for men, and women's contributions to intellectual and scientific advancements were often overlooked or attributed to male figures.
Similarly, the concept of 'intelligence' in AI research is limited to a specific type of knowledge, often ignoring alternative ways of knowing highlighted by feminist perspectives. These perspectives reveal the exclusion of knowledge systems traditionally associated with women or the feminine. Consequently, AI contributes to a broader socio-technical marginalization of women's knowledge, fostering a gendered and racialized view of 'intelligence.'
AI could offer an opportunity to challenge assumptions about male and female ways of knowing. For instance, the dichotomy of 'hard' and 'soft' intelligence is often gendered respectively as masculine and feminine. 'hard' intelligence, associated with logic and rationality, is generally replicated in technological forms, reinforcing its perceived dominance. As machines excel in cognitive tasks, skills traditionally associated with women, such as creative problem-solving and empathy, may become more valuable. Whether AI reinforces masculinist ways of knowing or potentially elevates a feminine one in the job market, it perpetuates binary, and gendered stereotypes of knowledge systems.
Gender Bias
When it comes to Gender bias in AI systems, it can occur on various levels such as misrecognition of women's faces and voices, reproduction of gender stereotypes, and exclusion of female perspectives. These biases mainly result from non-inclusive and non-representative training data, along with social prejudices held by developers of technology.
In the field of artificial intelligence, achieving a balance between genders is a pressing concern. The development of AI systems introduces the potential for gender bias, as those creating algorithms may unintentionally embed their own biases. These biases, rooted in societal gender norms, can be unconsciously transferred to robots. This is evident in current machine learning trends that perpetuate historical stereotypes, depicting traits like humility and gentleness as feminine and associating them with service robots, whereas, security robots are often portrayed with male characteristics. Moreover, the assignment of female attributes to bots is seen as a way to enhance their perceived humanity and increase the societal acceptance of AI.
Gender biases in AI technologies generally manifest through facial and voice recognition technologies and conventional gender roles incorporated into virtual assistants.
One prominent example is in the case of workforce management, as AI technologies are increasingly used for hiring, task delegation, performance assessment, and promotions across industries. Despite the potential benefits, concerns arise due to the risk of gender bias and algorithmic discrimination associated with these technologies. Biased data sets used for algorithm training can perpetuate historically discriminatory hiring practices. Technical specifications embedded with assumptions about gender, race, or disability may also lead to biased candidate evaluations. Research has shown that facial and voice recognition technologies have lower accuracy for women and non-white individuals. Additionally, AI can perpetuate gender stereotypes for marketing purposes, such as designing virtual assistants with female features and in traditional secretarial roles.
Research on major gender classifiers in facial recognition technology by IBM and Microsoft revealed higher accuracy in recognizing male faces than female faces. These systems were also more accurate for lighter faces than darker faces and the least accurate performance was observed for darker female faces. This discrepancy was attributed to incomplete datasets lacking female voices or images, mislabelling of training data by humans, and optimization of machine learning models for white male faces and voices.
Moreover, AI-driven technologies tend to perpetuate conventional gender roles to enhance marketing effectiveness. Virtual assistants like Alexa, Cortana, and Siri are intentionally designed with female attributes, including names and voices, portraying them as helpful, empathetic, and pleasant. These digital assistants carry out administrative tasks traditionally associated with women, such as scheduling and setting reminders.
The feminization of these voice assistants is noteworthy given their widespread adoption and the growing reliance on voice-based human-computer interaction. The design choices for these voice assistants reinforce stereotypical notions about women being subservient, constantly available, and susceptible to flattery or even sexual harassment and verbal abuse. This reinforcement may contribute to normalizing gender stereotypes and, in some cases, gender-based violence, particularly in the form of verbal sexual harassment. In response to these concerns, AI research organizations in Denmark have developed genderless voice assistants like Q, aiming to challenge stereotypes, break away from the gender binary, and promote inclusion and diversity through technology.
Another notable example of algorithmic perpetuation of gender stereotypes is observed in AI-powered translation tools, such as Google Translate. These tools employ neural machine translation, utilizing machine learning algorithms on extensive datasets. While well-trained algorithms can generate fluent sentences, if trained on biased data, they reflect and amplify those biases in their translations.
This issue became apparent when Google started translating between languages that do and do not assign gender to nouns. For instance, in professions, the translation would often provide a single stereotypical option, neglecting the possibility of both feminine and masculine forms. The translation algorithm would default to the pronoun most commonly associated with a particular profession, without recognizing the underlying sexist perspective it had acquired.
The persistent gender bias in algorithmic decision-making can be attributed to the lack of diversity among professionals involved in designing, coding, engineering, and programming AI technologies. Additionally, the male-dominated leadership in major tech companies, coupled with a lack of gender sensitivity, contributes to this bias. The absence of diversity among AI experts and designers results in a limited focus on researching algorithmic bias and its associated risks. Diverse teams are more likely to be attuned to potential sources of bias, identifying issues with negative societal implications before a product is launched. This may involve highlighting underrepresented demographic groups in training data sets, improving data labelling accuracy, testing the limitations of image searches and facial/voice recognition systems, and flagging stereotypical or sexist design elements.
The evolving capabilities of robots suggest a future where they not only become more intelligent but also form meaningful connections with humans, sparking new questions about how society perceives robots in various relationships, be it as companions or potential rivals. Experts predict that artificial intelligence will surpass human intelligence soon, yet warn that technological progress may lead to AI perpetuating societal biases, particularly concerning race, gender, and hostility towards women.
The rapid advancements in AI and robotics underscore the pressing issues of gender bias in algorithms, with concerns that future algorithms could specifically target women, posing risks to their roles in different aspects of life, as discrimination is no longer a purely human problem and has extended to the realm of artificial intelligence. Decision-makers are now aware of the illegality and immorality of discriminating based on attributes like gender, prompting them to conceal discriminatory practices behind seemingly constructed excuses.
A few approaches that can possibly tackle gender bias in AI could be implementing fair and ethical AI development, reducing algorithmic bias, and ensuring diversity in AI development teams. The dominance of male designers in AI algorithm creation has been identified as a potential cause of gender bias in resulting applications, making diversity in the development sector an effective strategy. Various techniques, such as gender tagging, bias control training, AI-enabled analysis, trade-offs, and data augmentation, have been proposed by experts to address gender bias. The integration of fairness principles in AI the development of algorithms is considered crucial for mitigating gender bias, with a focus on justice and fairness in procedures and outcomes.
Conclusion
AI technologies influence societal perceptions by embodying and perpetuating gender and race stereotypes. Feminist scholars have long explored the connection between gender and technology, drawing parallels between AI operation and learned gender behaviours.
The relationship between AI, the gender binary, and colonialism reveals a shared goal of establishing identity hierarchies. Notably, AI technology, particularly in facial recognition, can lead to discriminatory outcomes based on fixed notions of gender and race.
The embodiment of gender in artificial intelligence is evident in humanoid robots and virtual personal assistants (VPAs). Humanoid robots, such as Sophia, reinforce traditional gender roles through appearance, voice, and behaviour. Similarly, VPAs like Siri and Alexa perpetuate gender stereotypes through language, reinforcing power dynamics.
When it comes to gendered knowledge, historical biases have significantly influenced the construction and access to knowledge, often favouring men and marginalizing women. AI research, too, tends to align with traditional notions of intelligence, overlooking alternative ways of knowing associated with women.
Gender bias in AI systems can be apparent in various applications, including facial and voice recognition technologies, virtual assistants, and translation tools. These biases stem from non-inclusive training data, perpetuating societal prejudices held by tech developers, often leading to discriminatory outcomes. Moreover, Gender bias in AI also extends to workforce management, credit assessment, and social media algorithms, impacting women's opportunities and reinforcing historical stereotypes. Algorithmic translation tools and the promotion of AI products also contribute to gender biases, reflecting societal prejudices.
Addressing gender bias in AI requires fair and ethical development, reducing algorithmic bias, and ensuring diversity in development teams. The dominance of male designers in AI development is identified as a cause of gender bias, making diversity a crucial strategy. Techniques like gender tagging, bias control training, and fairness principles in algorithmic development aim to mitigate gender bias and promote equality in AI systems. There is a crucial need for vigilance in AI development to prevent discriminatory practices and promote fairness and diversity.