Chapter 10 What’s Missing

Imagining Infrastructures for Whom

In imagining the infrastructural futures for digital humanities (DH) in India, we must first understand whom we are building for. Most importantly, conceptualizing infrastructural futures for DH in India requires us to move beyond traditional academic boundaries. As with many countries in the Global South, DH in India is not simply a product of academic innovations or robustly funded ecosystems but a response to public need and demand. The profusion of actors engaged in the interplay of digital and human activities goes beyond institutionally sanctioned spaces, producing a vibrant and conceptually demanding space (Dodd and Kalra 2021). This practical and epistemic diversity is, of course, characteristic of many countries in the Global South (Gobbo and Russo 2020). If we are to imagine a future for DH infrastructures in India, we must first assume a broader conception of DH, which implies a parallel assumption that infrastructures for DH in India need to be infrastructures for public humanities in India. Hence the infrastructures required presuppose a different trajectory than countries in the Global North, charting a course that embraces multimodal ways of designing, thinking, and building.

The first wave of DH in India (and in other South Asian states) over the past two decades has been characterized by the emergence of individual and crowdsourced initiatives without the benefit of official mandates in terms of funding or policy. This activity has, nevertheless, generated an important first generation of digital resources designed by several actors for the public good.1 Such practices, aligned to but distinguishable from the practice of minimal computing, have produced far more dispersed national DH infrastructure than would have resulted from initiatives by individual institutions (Roy and Dodd 2024, 248). Diasporic and national publics, representing broad aspects of society, have used digital methods to establish infrastructure in support of public memory and community history. This activity has, for instance, produced religious textual corpora for digital publics,2 with patronage frequently found outside of academic structures.

As Roy and Dodd remind us, practitioners and infrastructures of DH in India lie within and outside institutional systems (Roy and Dodd 2024, 243). The story of Indian DH is an extension of public humanities in India; infrastructural futures need to encompass this broad landscape. It is useful to view the emergence of DH in India as rhizomatic, where “the realisation of DH in research, pedagogy, and practice is [gained] primarily through acts of self-identification and not through previous affiliations to the existing big tents” (Roy and Dodd 2024, 240). That perspective should also animate imagined futures for DH labs and courses in India, that need to be informed by postcolonial histories and the diverse contexts of Indian classrooms.3

As with most countries around the world, in the Global North as well as South, design and delivery of DH curricula in India relies to a great degree on free and open-source software and extra-institutional resources and support (Shanmugapriya and Menon 2020, 7). Both inside and outside the university system, this has allowed for the scaffolding of DH work, despite continued challenges to connectivity and funding. The imaginative deployment of digital infrastructure, tools, and methods can help overcome these constraints. It is only by apprehending the uneven terrain of DH in India, and understanding its functional drivers, that we can delineate the challenges it faces. Infrastructures then need to address practitioners both inside and outside the university lab and are further informed by a broader ecosystem of internet freedom based on an emerging landscape of legal rights.

Digital India and Disconnected India

The futures of DH in India rely to a remarkable degree on an ability to contend with a rapidly developing and sociotechnically complex national digital infrastructure. In this regard, it is salient to note that for the past five years, India has topped the charts for internet shutdowns globally.4 This is not—it is important to stress—a result of technical failures but rather administrative direction. Denial of service has become a flash point for conversations about national internet governance, with providers reportedly shutting down services in the name of public safety. The effect is to deny users the right to connectivity by citing security concerns. According to Access Now and the #KeepItOn coalition’s new report, India imposed eighty-four internet shutdowns in 2022—the highest number globally for the fifth year in a row (Skok et al; Rajvanshi). The world’s largest democracy has been responsible for approximately 58 percent of the total number of shutdowns documented since 2016.

Reliable service and open access would seem to be obvious requirements for a developing national internet infrastructure, but this is not always viewed as the case by a state apparatus with complex motivations. As Nishant Shah observes, the patchiness of internet connectivity influenced the first wave of DH activity in India, focusing on questions of access and the provision of technological solutions for the preservation and circulation of data (169). This first wave sought to bridge the “connected” from the “to-be-connected,” which is most evident in the case of Facebook Basics (see Sosale). This developed into a politics of the disconnected subject. The notion of infrastructure-as-development has continually shifted, molded by parallel debates about citizenship rights. Uncertain legal privileges undermine connectivity rights; this basic reality needs to be factored into any discussion of DH infrastructural futures.

Impending legislation suggests that individual privacy, enshrined in frameworks like that of the European Union’s General Data Protection Regulation (GDPR), are not a priority for Digital India. The definition of high-quality infrastructural development is defined by the imagined user, and in India, this implies a late capitalist maximization of a vast emerging market. The rollout of 5G is a case in point, with national aspirations far exceeding reality, as the price point of the hardware was unattainable to the vast majority of consumers (ET Telecom). The desire for the creation of ever-larger markets is creating an untenable equation where affordances and affordability are not in sync. The problem crosses sectors and service categories: cheap data, smartphone growth, and the rollout of artificial intelligence (AI) technologies also assumes that digital inclusivity is around the corner. Conversations about the risks of emergent technologies in India are stuck in limbo, between concern about growing digital divides that impede human flourishing (Thirumal and Tartakov) and national anxiety about increasing access for digital belonging. And although there are legitimate anxieties around basic issues of access and privacy for the common citizen or heightened threats related to AI, more profound issues exist in draft legislation poised between surveillance (such as a blanket deployment of facial recognition software in the name of national security) and inadequate attempts to bridge often-substantial gaps between intention and implementation. On the eve of the proposed Digital India Act, there is no doubt that we are at an inflection point that divides an internet that we have known, as a space of potential empowerment albeit with uneven access, from a brave new digital world that we fear.

Digital India, inaugurated as state policy in 2015,5 represents a comprehensive vision for using technology to drive social and economic development, improve governance, and empower citizens across the country. India Stack, initiated as a set of digital infrastructure components designed to enable a secure and efficient ecosystem for online transactions and service delivery across the country, is central to this policy; and since November 2023, it has been termed “Digital Public Infrastructure.”6 Technology stakeholders often boast about the success of the initiative, noting that “what began with the Aadhaar digital identity way back in 2009—joined later by services like the Unified Payments Interface (UPI), JAM (Jan Dhan Yojana, Aadhaar and Mobile number) trinity, and Co-WIN (for managing the Covid-19 vaccination programme), among others—helped India achieve 80 per cent financial inclusion in just six years, which one paper from the Bank for International Settlements estimated would have otherwise taken 47 years to achieve” (Singal). While ordinary citizens seek personal data protection, a lot more is expected to be built on India Stack.

As India grapples with the challenges of delivering high-quality digital services to the last mile, it also increases engagement with what will simultaneously be the world’s most populous and youngest nation. This means that any analysis of digital infrastructure in India must contend with the reality of increasing state power, which seeks to deliver development through digital platforms and services, and the use of state power to rein in existing freedoms through the same technology. India Stack is touted as the state’s success story because of its contribution to digital inclusion, and it is marketed as a solution for developing countries needing to provide solutions at scale for identity, payments, health care, and education. Eight countries have signed memoranda of understanding (MoUs) with India: Armenia, Sierra Leone, Suriname, Antigua, Barbados, Trinidad and Tobago, Papua New Guinea, and Mauritius, which allow them to access Indian digital infrastructure with no-cost open-source access (Raghavan, Jain, and Varma 2019).

As Adrian Athique and Akshaya Kumar have noted, “The digital development race in Asia has impelled infrastructure investments that entail close partnerships between national governments and local firms capable of operating universal digital infrastructure at scale” (1421). This reflects the “conceptually unruly” nature of infrastructures that Brian Larkin identifies: Whether viewed “as a system of substrates, networks or interconnected technological systems . . . [infrastructures] need to be seen as an amalgam of technical, administrative, and financial techniques” (328). In India, as elsewhere, digital platforms are market systems that connect domains of commerce, technology, sociality, and logistics (Athique and Parthasarathi). Adrian Athique and Akshaya Kumar demonstrate how this is playing out in India, noting that the “ambitions of Digital India are gargantuan, with multiple projects grouped under three core headings: ‘digital infrastructures’, ‘governance service’ and ‘digital empowerment’” (1421). While the story on state schemes continues to unfold, the legal landscape around markets, individuals, and rights remains nebulous.

Because DH (in India as much as elsewhere) do not represent a singular cultural phenomenon but rather a range of practices, its conditions of future possibility need to be connected to discourses related to the state and the market, and to private and personal limits to freedom. And this larger canvas needs to be positioned alongside the seemingly unstoppable processes of global digitalization that proliferate around us. While we were writing this chapter, many facts on the ground changed—people’s access to India’s digital infrastructure, reformulating state laws, and machinic possibilities are all fluidly recasting a future faster than can be apprehended. Laws are being redrafted (as with the much-revised Digital Personal Data Protection Bill 2022, the impending Digital India Act, and the National Data Governance Policy). These are among a plethora of initiatives, ranging from definitions of digital liability to new taxation, and symbolize a creeping centralization with its own dangers. Further, when all these initiatives occur in a fast-evolving landscape of AI technology, that itself seems to elude real-time regulatory thinking, and these rapid changes obscure who the real subject of these efforts indeed is.

Imagined and Excluded Subjects

The dizzying speed of regulatory changes can obfuscate the subject at the heart of all these infrastructural affordances, requiring us to recast the question again: Who are we building for? It is useful to return to the imagined subject that this mega-scaffolding is intended for. The challenges are more complex than those that can be met with promises of providing material infrastructures for bringing people online or providing an internet device. Digital divides are not simply remedied by providing connectivity and access. In a time in which we are all “data citizens” (Bowker, xiii) questions of trade off are inevitable. Just as Facebook Basics7 proved to be an unworthy deal (Mukerjee), we need to question whether Faustian bargains are being entered into in exchange for mass connectivity. If DH in India has a single message to offer the wider DH world, it is that digital public infrastructures are profoundly political, and arguments about what constitutes the public good can rarely be resolved in purely academic contexts (Larkin, 189; Appel, Anand, and Gupta, 2). The potential of critical infrastructure studies (CIS), in this sense, lies in its ability to problematize and potentially surmount academic, public, and private boundaries.

The relationship between CIS and DH in an Indian context is fraught with difficulty too, however. Although infrastructures are central to the practice of development, limits to digital technology render many DH practices fragile in India. In the Sisyphean task of building capacity, elements necessary for the development of DH as an inclusive field remain missing. Are connectivity and development alone adequate to bridge the digital divide, without an imaginative regulation that is capacious enough to handle a variety of needs on scale? Critical theory places such an emphasis on defining the problem (as if it were static) that we are often left holding fossils of meaning that were once accurate but inadequately reflect the reality of the present moment.

To explain this, we turn to the ideas of Prathama Banerjee. In her recent work, Elementary Aspects of the Political: Histories from the Global South, she challenges the givenness of certain assumptions as universal, such as the definition of the political. Just as the definition of what constitutes “political” can itself be unraveled, so can assumptions about the connections between infrastructure and development in the Indian context. Banerjee makes visible and unravels our “contemporary assumptions about the self-evidence, universality and primacy of the political” to show how there is “no essence to the political” and that “what we call political emerges only in terms of its differentiation from the non-political . . . and in terms of its delimitation by the extra-political that always returns to haunt it” (215). Her gesture informs our unpacking of the basic elements of infrastructure and serves as a heuristic device to guide our understanding.8

Viewed in its full sociotechnical complexity, the discourse of infrastructure-as-development comes perilously close to info-solutionism (Dodd, “Querying Info-Solutionism”), suggesting lessons for the global DH community. It is easy to position DH infrastructures as merely in need of more development—as suffering from a resource deficit problem—rather than being imbricated in broader patterns of sociopolitical and technical control. This has flow-on effects that obscures clear thinking on other topics. We assume that we need more formal funding structures, for example, but ignore other less resource-rich models that are capable of meeting DH needs (Roy and Dodd, 248). These sorts of problems probably stem from the fact that development narratives are embroiled in the modernist teleology that always demands a singular temporality, instead of being open to multiple ones (Gupta 2018, 70). Modernity impels us to look at infrastructural solutions in terms of possessing a phone, having Wi-Fi or 5G. And yet a phone and an internet connection do not guarantee connectivity or imply access, representation, or equality, any more than a DH lab does. Viewed in the broader context of digital colonization (Dhapola), we need to continually ask how many people are automatically excluded not only from connectivity but also representation (which, certainly in the context of a DH lab, could occur without individual connectivity). Beyond the legal landscape and the connectivity question lies the human reality of India’s current internet architecture. Are we building DH for them? Whom are we building these imagined futures for?

Exclusions by/of Language

For a country with the largest population in the world, it is striking that not a single Indian language appears on the list of the top ten most used languages on the internet (Brandom). The structural exclusion of non-English-speaking peoples is a defining impediment to curricular DH in higher education in India, as it is in many countries around the world, framing institutional possibilities (Roy and Dodd). Consider how many common Indian languages are present online. The first-ever platform survey to collect and analyze global interface languages, the pioneering “State of the Internet’s Languages Report,”9 throws some light on this. The report found that “in South Asia, almost half of the platforms surveyed do not offer interface support for any regional language, and major South Asian languages such as Hindi and Bengali, spoken by hundreds of millions of people, are not as widely supported as we might expect. Support for South-East Asian languages is similarly mixed: while Indonesian, Vietnamese, and Thai tend to be very well supported by the platforms we surveyed, most other South-East Asian languages are not” (Whose Knowledge?, Oxford Internet Institute, and the Centre for Internet and Society [India]).

In their survey of eleven websites, twelve Android apps, and sixteen iOS apps (selected across widely used platforms), one finding stands out. Even though Wikipedia (“the largest collaborative effort in human history”) began with a single English-language edition, it is now available in over 300 language editions. Wikipedia’s user interface has been translated into more languages than any of the commercial platforms, including Google (available in 150 languages) and Facebook (106 languages).

It is important to recognize that this situation also extends to the code that underwrites global digital infrastructure as well. In an article for Wired magazine in 2019, Gretchen McCulloch bluntly noted that “even huge languages that have extensive literary traditions and are used as regional trade languages, like Mandarin, Spanish, Hindi, and Arabic, still aren’t widespread as languages of code. I’ve found four programming languages that are widely available in multilingual versions.”

Comparing the journey from Latin to English also serves as a reminder that the first website was not written in just Hypertext Markup Language (HTML)—it was written in English HTML. The lack of bridges between languages and code has resulted in a warped representation of lived realities online and offline. As Mark Graham of the Oxford Internet Institute noted in The Guardian’s feature on the digital language divide, “rich countries largely get to define themselves and poor countries largely get defined by others” (Young). Even if it is heartening that Wikipedia’s 300 languages and several translation tools enabled the transfer of content across these languages, the fact is that what has been left behind is staggering. The point to remember here is that we are not only referring to the absence of mother tongues and semantic worlds, but entire worldviews that cannot be accurately portrayed by mere translation. So, where is the imagined nation of Indian-language speakers online?

The Absent Training Set

As Anasuya Sengupta has noted, only 7 percent of the world’s languages are found in published materials, and even fewer are represented on the internet (Vrana, Sengupta, Pozo, and Bourterse; Spence). English and Chinese dominate the languages available online, which include just 500 of the 7,000-plus languages of the world. This is significant to note, especially since 75 percent of internet users are from the Global South, indicative of an expansive gulf between the users and producers of digital content and infrastructures (Vrana et al.). This gains further urgency when one considers that the training sets of large language models (LLMs) used by AI models are primarily trained on the English-dominated internet for their capacities. Kalika Bali is a principal researcher at Microsoft India, whose work is featured in a report noting that India is “a profoundly multilingual country with 16 languages enjoying primary official language status at state level and 29 languages with more than a million native speakers” (Spence, 10). Given this, and noting the reliance of many AI, natural language processing (NLP), and speech technology systems on a very small number of languages, Bali highlights the challenges of building technology for low-resourced languages. “In global terms, no indigenous Indian language qualifies in her category of the fifteen ‘highly-resourced’ global languages in terms of digital support, and most sit in the ‘under resourced’ or ‘no source’ categories” (Spence, 10).

To create in one’s own image is to tell one’s own story (and also, in a time of generative AI, to multiply it), rendering cultural and linguistic absence deeply impactful. Acknowledging the profound biases of AI training sets draws attention to the specter of the stillborn user. With the rapid development of AI, quickly increasing its capacity from one generation to the next, these absences are dire. It is telling that India provides ChatGPT (the AI chatbot service that famously earned 100 million downloads within two months of its launch), with its second-largest user base (at 7.1 percent, compared with 15.1 percent from the United States) (Upadhyay), but without data from Indian sources, LLMs will merely extend the biases of the internet into the future.

Missing Infrastructures and AI

Indigenous innovation will be needed to truly harness the power of AI for India, and the Indian state is aware of this. To accelerate AI, in May 2022 the Ministry of Electronics and Information Technology (MEITY) launched Bhashini, an Indian languages database that can be used by research institutes (Das). The government is also setting up a draft National Deep Tech Startup Policy to explore the development of LLMs (Koshy). As two office-bearing government representatives opined in a recent newspaper piece,

The current landscape of LLM development is dominated by US-centric models. . . . akin to the global impact of social media platforms like Facebook, Google . . . the risk of perpetuating this US-centric paradigm is immense . . . For reference, GPT-3 was trained on 45 terabytes of text data, roughly equivalent to 500 billion tokens . . . and OpenAI has invested in the neighbourhood of $10 billion to ‘train’ GPT-4 . . . The Nilekani Centre at IIT Madras’ AI4Bharat has collected over 21 billion open text tokens [and] also collected 100,000 hours of YouTube videos in Indian languages that have been published under Creative Commons (CC) 4.0 licence. However, even this massive data collection is about two orders of magnitude too small to build GPT-4-style models (D’Monte and Kolla).

The assumption is that by unlocking existing Indian-language data, developing an Indigenous LLM, and forging public-private partnerships, India can create an indigenous LLM that counterbalances the influence of dominant players and ensures a more equitable distribution of AI benefits. The spirit of nation-building is invoked for this, as they point out that “last year, the Bhasha Daan initiative of the Bhashini project was launched in which citizens could ‘donate’ their existing voices to help build datasets for Indian-language AI. This programme may be reimagined and revamped to crowdsource data for Indian languages that may be relevant to collect data to ‘train’ an LLM” (Vempati and Raghavan).

In line with these policies, Indian academia is initiating an increasing number of generative AI research projects, many of which are seeking to understand how technology can help create tools similar to ChatGPT using Indian languages. Researchers indicate a host of challenges for such projects, however, the biggest of which lies in sourcing ample amounts of Indian languages data. The cost of such projects and the scale of computing power needed are equally challenging problems to resolve (Das). The scale of the training sets needed for developing AI across India will require crowdsourcing to resolve. While the state has already turned to AI to take on questions of scale, such as filtering tax returns and generating court transcripts (Economic Times), the need for greater diversity in input also requires citizen activity and other forms of social datasets (Haini).

The Future of Infrastructure

DH curricula could support the development of crowdsourced and inclusive language datasets if designed with this goal in mind; the process of decolonizing a curriculum and developing content for the curriculum are one and the same. Reflecting on what is missing in our thinking about infrastructures for DH in India requires us to think along three interconnected planes. First, we look at the subject and content of DH: Who are the audiences for DH infrastructures? Whom are we building for? Is there enough recognition of DH in India? Second, what are some of the challenges in building for this audience, and what factors might influence or shape the ecosystem? Because building infrastructures for DH in India intersects with public humanities, infrastructures for DH are imbricated in the existing and shifting legal and rights infrastructures (not to mention the existing language inequalities) of the internet. Third, imagining futures for DH infrastructures in India requires a critical understanding of infrastructures outside the teleological logics of modernity. Is it possible to draw out the contours affording DH possibilities in India as a function of both imagination and infrastructure?

The simple fact is that the use of digital tools rests on a global access system predicated on the English (and to a significant extent, Chinese) language, something that both DH and CIS need to contend with. The need to develop infrastructures goes beyond commercial and instrumentalist ends. By returning this challenge to the lived practices of DH in India, we might yet accord digital citizenship to those who do not live in the English-speaking world. Until then, the view that we have of the future is but an elementary aspect of temporary infrastructures that shall always fall short.

Notes

1. 1. See Project Madurai, founded in 1988 (https://www.projectmadurai.org/); the Punjab Digital Library, established in 2003 (http://www.panjabdigilib.org/webuser/searches/mainpage.jsp); the Indian Memory Project, established in 2010 (https://www.indianmemoryproject.com/); Sahapedia, established in 2016 (https://www.sahapedia.org/); pad.ma (https://pad.ma/); the Nepal Public Library (https://www.nepalpicturelibrary.org/); and the South Asian American Digital Archive (https://www.saada.org/).

2. 2. See the Tibetan and Himalayan Digital Library (https://www.thlib.org/) and Pali Tipitaka (https://tipitaka.org/).

3. 3. See Roy and Dodd; Shanmugapriya and Menon.

4. 4. India has ranked first in the world for shutting off the internet over the past five years, according to SFLC.in and other digital rights watchdogs. For the case of Manipur, see Gupta; and for the larger context, see Gupta and Shih and Ali.

5. 5. This program was launched on July 1, 2015. See https://csc.gov.in/digitalIndia.

6. 6. India Stack (https://indiastack.org/) was conceptualized to support the government’s Digital India initiative and promote financial inclusion. The key components of India Stack are a unique identity number known as Aadhar, a number that is issued to Indian residents; an electronic identification process known as e-KYC, the paperless governance of a Digital Locker that serves as an online vault for verification documents, backed by electronic signature or e-sign; and a United Payments Interface (UPI), a digital payment methods that use Aadhaar-enabled Payments System (AePS). For more on India Stack as a digital public infrastructure, see Singal.

7. 7. The saga that began ten years ago in India in 2013 is well delineated (Mukherjee). Almost a decade later, instead of the state, Reliance Jio and Meta entered a deal for chat-based shopping (enabling grocery delivery on WhatsApp) and are now said to be in talks to set up physical infrastructures for large AI applications (Kar).

8. 8. There is a longer history of the use of the term elementary in Indian scholarship. We build upon the form of elementary in the same vein as Prathama Banerjee (examining elementary aspects of the political) and Ranajit Guha (examining elementary aspects of peasant insurgency in colonial India).

9. 9. Begun in 2019 to advocate against language inequality, the State of the Internet’s Languages Report is a collaboration among three organizations: Whose Knowledge?, Oxford Internet Institute, and the Centre for Internet and Society (India).

Bibliography