Speaking back to the divide: Reassembling gender,mobile communication,and AI through feminist data justice in Zambia

Abstract

Drawing on the Ask Viamo Anything (AVA) Pilot in Zambia, a voice-first generative AI integrated into mobile phones, this article examines how mobile AI platforms reassemble the gendered digital divide. Through a feminist technology studies and feminist data justice lens, it argues that systems like AVA embody both empowerment and constraint: they enable women and youth to engage in private, trusted, voice-based communication while simultaneously reproducing structural exclusions through language, moderation, and infrastructural dependency. This commentary situates mobile AI within mobile communication scholarship, extending its concern with everyday mediation, intimacy, and mobility into the algorithmic age. In doing so, the article illustrates mobile AI as a new communicative form that redistributes agency, visibility, and voice along gendered lines.

Introduction

In recent years, public and scholarly discourse around artificial intelligence (AI) has focused on its disruptive potential, from large language models (LLMs) to automated decision-making generative tools. Yet beyond the high-profile debates about regulation, ethics, and innovation, mobile AI is increasingly embedded in everyday life. Thanks to smartphones, and other connected devices, mobile AI systems such as Google Gemini increasingly shape everyday communication and access to services.

Simultaneously, the digital divide continues to be a key area of research, policy, and intervention. Scholars and practitioners emphasize that digital inequality extends far beyond infrastructure or device access (Donner, 2015; GSMA, 2025a). It includes disparities in digital literacy, relevance of content, affordability, and language inclusion that shape meaningful engagement with digital technologies. As AI becomes central to everyday communication, mobile AI scholarship must be situated within broader debates about digital inequalities. These inequalities stem from technological, linguistic, and socio-economic constraints that shape who participates in digital futures and on what terms.

From a Science and Technology Studies (STS) perspective, socio-technical practices around mobile AI are co-constructed by both users and technologies. Earlier work on mobile communication (Oudshoorn & Pinch, 2005; Wamala, 2012) has shown that innovation and use are mutually shaped through everyday practices, negotiations, and cultural meanings. Building on this understanding, this article explores how mobile AI systems are appropriated, adapted, and given meaning within specific socio-technical contexts, and how these processes are shaped by gender.

This article draws on the Ask Viamo Anything (AVA) pilot launched in Zambia in 2023 by the social enterprise Viamo. As a social enterprise, Viamo partners with mobile network operators and development agencies, providing information services through Interactive Voice Response (IVR) technology. The analysis draws on secondary data from the Global System for Mobile Communications Association’s (GSMA's) study of AVA, ¹ which combined call-log analytics and qualitative interviews conducted in partnership with Viamo. As a commentary, the paper critically interprets these publicly available findings. Written from the standpoint of a researcher based in Sweden with long-standing collaborations in Sub-Saharan Africa, this commentary adopts a situated yet external perspective. Although I am not Zambian, my engagement with digitalization initiatives in the region, including Zambia, informs my critical attention to the linguistic and infrastructural asymmetries examined in this paper. AVA integrates voice-first generative AI into Viamo's existing IVR infrastructure, enabling users, many of whom have limited literacy or lack internet access, to ask questions and receive personalized spoken responses.

Viamo operates across 25 counties, primarily in Sub-Saharan Africa, delivering on-demand information in local languages through voice-based formats. The service is able to reach women, youth, and rural and peri-urban communities who are often on the margins of the digital information economy. Using IVR technology, Viamo delivers curated, localized messages co-developed with public and civil society partners across sectors such as health, education, agriculture, and finance. By 2024, the platform had recorded 24.7 million active users across its 25 core countries (GSMA, 2025a).

However, traditional IVR systems rely on pre-recorded messages, which, while effective for broad information delivery, limit interactivity and personalization. To address this, Viamo piloted its generative AI-based AVA product in Zambia, combining speech recognition, LLM processing, and text-to-speech technologies to enable conversational engagement accessible through basic mobile phones.

Despite AVA's explicit focus on inclusion and equity, its gendered implications have not yet been examined. This article takes a gendered perspective to analyze how AVA reconfigures access, participation, and voice among marginalized users, particularly women. Drawing on feminist technology studies, it adopts the notion of mutual construction of gender and technology, the idea that technologies are not neutral but are shaped by and, in turn, shape gendered relations of power (Wajcman, 2010; Larsson, 2019). Through this lens, AVA is examined as a gendered artifact: one that opens new communicative opportunities while simultaneously reproducing certain exclusions.

Positioning mobile AI within mobile communication scholarship

Early mobile communication research established how mobile devices reorganize sociality, presence, and everyday coordination, shifting from fixed-line to perpetual, portable connectivity (Katz & Aakhus, 2002; Ling, 2012). This scholarship mapped mobiles as infrastructures of intimacy, coordination, and place-making across diverse contexts, and later broadened mobile media, apps, platforms, locative services, and camera cultures, emphasizing hybridity of social, technical, and cultural practices (Goggin, 2025; Goggin & Hjorth, 2014). As the field consolidates, Mobile Media and Communication's recent editorials note a turn toward “different keys,” including infrastructures, automation, and planetary-scale concerns (Campbell & Komen, 2023), urging scholars to grapple with AI's growing role in mobile ecologies.

Within communication studies more broadly, emerging work theorizes AI-mediated communication to capture interactions where computational agents not only transmit but modify, augment, or generate messages, thereby challenging long-held assumptions about agency, authorship, and relational cues in interpersonal communication (Hancock et al., 2020). This calls for revisiting classic concerns, language style, identity signaling, trust, and disclosure, under conditions where AI systems increasingly participate as co-authors or filters in everyday exchanges (Hancock et al., 2020).

For mobile AI, the smartphone has become the dominant host and interface, embedding LLM chat apps, voice assistants, and on-device inferencing into routine activity. Recent syntheses in mobile media scholarship explicitly name mobile AI as a critical horizon after the smartphone, foregrounding how AI underpins search, translation, creativity, mobility services, and everyday decision support. These interventions reframe phones not just as terminals but as ambient AI interfaces that recalibrate everyday practices (Goggin, 2025).

Research on digital voice assistants documents how voice, accent, and gender presentation shape trust, perceived competence, and interactional dynamics, often reproducing social stereotypes such as critiques of default feminized assistants. Public discourse further shows men interrupting gendered voice assistants more often, raising design concerns about how feminine voicing norms may reinforce bias (Piercy et al., 2025). Drawing on work revealing the gendered and racialized design of voice assistants (Phan, 2023), these studies underscore that mobile AI is never neutral: model training data, voice/accent choices, and conversational scripts enact cultural politics that intersect with gender, race, and language (Piercy et al, 2025).

A parallel discourse on digital inequality cautions that access is multi-dimensional, encompassing infrastructure, affordability, literacy, and relevance, and that gaps persist even where coverage exists. GSMA's recent Mobile Gender Gap Report (GSMA, 2025b, p. 34) again shows disparities in mobile internet adoption and use by women across low- and middle-income countries, highlighting affordability, content/language relevance, and even safety concerns as enduring barriers. As AI becomes more central to mobile services, these barriers risk translating into AI access gaps, especially where local languages are underrepresented in training data and speech stacks.

Against this backdrop, voice-first mobile AI tools in low middle income contexts delivered via IVR and basic phones have emerged as notable interventions that bypass some barriers such as smartphone ownership, data costs, and literacy. At the same time, they have surfaced new ones such as language coverage, moderation governance, and speech recognition accuracy. Viamo's AVA embodies this shift, layering speech-to-text – LLM – text-to-speech atop a nationwide IVR. It reorients mobile communication from one-way, pre-recorded menus toward conversational, personalized exchanges that are reachable on feature phones. AVA's Zambia pilot indicates “strong uptake among women and youth” (GSMA, 2025a, pp. 18–19), with women comprising the majority of users and engaging in more follow-ups, especially on health and sensitive topics, suggesting that anonymity and voice access may open new communicative spaces for marginalized users.

For mobile communication scholarship, these developments invite a reframing, from device-centered dyadic models to AI-inflected infrastructural models where voice user interfaces, guardrails, and language resources materially shape who can speak, what can be asked, and what kinds of answers circulate. Existing studies provide tools to analyze these shifts, combining mobile media's attention to everyday practice, AI-mediated communication's focus on agency and message authorship, and feminist technology studies’ insistence on co-construction. Together, they position mobile AI not as a frictionless layer added to mobility, but as a socio-technical configuration that can both expand participation for communities on the margins, and at the same time reinscribe exclusions.

Theoretical framework

Feminist technology studies (FTS) have long established that technologies are not neutral artifacts but are shaped by, and in turn shape, gendered relations of power (cf. Wajcman, 2010). Early feminist scholars exposed how design processes (Cockburn & Ormrod, 1993), workplace technologies (Cockburn, 1985), and media infrastructures encode assumptions about gender roles, labor, and competence (Wajcman, 2010). They argued that understanding technology requires examining the social practices and values built into design, as well as everyday negotiations through which users appropriate and reconfigure those technologies. This perspective remains vital for studying mobile AI, as systems like AVA are seen as social artifacts whose design and use are entwined with local gendered realities.

Building on FTS, feminist data science and design justice scholarship extend these insights to the algorithmic and data-driven age. Data feminism (D’Ignazio & Klein, 2020) reframes data as a form of power, arguing that equity in data systems requires questioning who collects data, whose voices count, and what forms of knowledge are privileged or erased. The design justice framework (Costanza-Chock, 2020) emphasizes community-led design processes that center those most affected by technologies, especially marginalized groups historically excluded from design decision-making. Both frameworks challenge the idea that digital inclusion is achieved merely through technical access or deployment; rather, justice-oriented design asks how technologies might redistribute agency and representation.

Applying these perspectives to mobile AI exposits how generative and voice-based systems operate within broader social hierarchies. Design choices such as the language models used, the moderation rules applied, or the tone and gendering of AI voices carry implicit norms that influence who feels recognized and who remains invisible. Feminist data and design frameworks also make visible how infrastructural decisions such as English-only models or high data-cost interfaces reproduce existing inequalities under the guise of technological neutrality. In AVA's case, the combination of voice interaction, mobile delivery, and generative AI reconfigures access to information for users, particularly women, who have been marginalized by traditional digital infrastructures. Yet the system's guardrails and language limitations also delimit what can be asked and said, shaping the contours of participation in gendered ways.

These theoretical knots offer a way to study mobile AI as a site of both empowerment and constraint. FTS provides the analytical vocabulary to examine the gendered assumptions embedded in AI infrastructures, while feminist data science and design justice extend that vocabulary to interrogate data flows, representation, and participation.

Gender, trust, and voices with AVA: Discussion and analysis

AVA demonstrates how mobile AI can extend access to information for populations historically excluded from digital infrastructures. AVA bridges the digital divide through voice, by combining IVR and generative AI. This modality bypasses literacy, cost, and smartphone barriers. Between May 2023 and January 2025, about 36,000 Zambian users engaged with AVA, submitting over 614,000 questions, 95% of which received spoken responses within an average of three seconds (GSMA, 2025a, p.18).

From a feminist data and design justice perspective (Costanza-Chock, 2020; D’Ignazio & Klein, 2020), this initiative reconfigures data power at the margins, where voice becomes a site of both representation and recognition. Instead of users adapting to text-heavy, English-dominant digital systems, AVA adapts to users’ embodied communication practices, echoing design justice's call to “design with rather than for” marginalized users. However, this co-production remains partial. English remains the sole language, restricting full participation and reinforcing what can be understood, drawing on Birhane's (2021) work, as a form of algorithmic coloniality, that is, the persistence of linguistic and epistemic hierarchies within AI systems.

The gendered dynamics of AVA are striking. Of all users, 59% were female, and women not only used AVA more frequently than men but also asked more follow-up questions, which translates to a 35% follow-up rate overall, with women and young users (18–24) most likely to continue the dialogue (GSMA, 2025a, p.18). Data shows that the average user was female, aged 18–24, and from Lusaka or the North-Western Province (GSMA, 2025a, p.18).

This data challenges long-standing patterns in mobile communication, where men typically dominate digital access and interaction (cf. Wamala-Larsson, 2019). The AVA findings suggest that voice-first generative AI, when coupled with anonymity and trust, can create alternative spaces of gendered agency. Women reported feeling comfortable asking questions they could not raise publicly, particularly about health, sexuality, and HIV. AVA in FTS terms can be seen as an intimate infrastructure (Wilson, 2016) – a socio-technical space where privacy and embodied voice intersect to reframe access as relational and affective.

From a design justice view, this user trust demonstrates epistemic inclusion and users defining what matters to them. Yet it also raises ethical questions about how trust is engineered. Viamo's moderation protocols, designed to filter sensitive content, potentially reproduce paternalistic logics of care: protecting users by precluding certain conversations, for example around reproductive rights. FTS reminds us that safety and control systems are not neutral but gendered design decisions (Bardzell & Bardzell, 2016; Wajcman, 2010). Thus, while AVA empowers women to speak, it also curates what they are allowed to know.

User interviews conducted by GSMA (2025a, p. 21) identified four perceived benefits of AVA: trust, detail, ease, and anonymity. These align directly with feminist data science's redefinition of access, not only as infrastructural connectivity but also as meaningful participation. Users likened AVA to a “search engine without internet, (ibid. p.21)” emphasizing the comfort of asking taboo or awkward questions privately.

This anonymity redefines mobile communications affordances. Instead of connectivity being social and visible as with Facebook and other social media, AVA's value lies in invisible, individualized connection. For women and youth, especially those navigating social stigma, this anonymity transforms the phone from a surveillance device into a safe space, a rare inversion of the gendered gaze of technology. Yet, as feminist scholars note, invisibility is ambivalent; it shields but also isolates (Harcourt, 2017). The lack of visible community exchange means that women's voices are recognized by the system but remain inaudible to collective publics, or what Couldry and Mejias (2019) call data colonialism, where intimate queries become datafied but rarely feed back into users’ social worlds.

Roughly 98% of the Zambian population, for whom English is not a first language, were excluded from AVA's English-only design GSMA (2025a, p.14). This exclusion illustrates how infrastructural inclusion can co-exist with epistemic exclusion. From an FTS perspective, this reveals the tension between the universality of AI design and situated knowledges, or put differently, technologies claim neutrality but are built upon linguistic hierarchies inherited from colonial infrastructures. Additionally, Viamo's content moderation and guardrails, which filter out sensitive or harmful topics, raise concerns about who defines harm and whose values are encoded as safe knowledge (cf. Couldry & Mejias, 2019). The guardrails, co-defined with development agents, represent institutional negotiation of gendered speech, where some forms of female curiosity, such as reproductive autonomy or violence, may be silently disqualified. Feminist design justice would argue for community-defined safety, where content moderation is shaped by the lived experiences of users themselves, not institutional proxies.

Conclusion

AVA embodies a transition from listening systems (IVR) to responsive systems (AI), shifting agency from the sender to the user. Through an FTS lens, AVA can be read as a moment of gendered reconfiguration in mobile communication. Yet this agency remains bounded by infrastructural and epistemic asymmetries such as language constraints and top-down moderation. AVA's pilot privileges oral literacies over written ones. It demonstrates how trust and anonymity can lower gendered barriers to participation.