Chat-HI: Exploring Indian National Identity Through Machine-Generated Text

Abstract

Situated at the crossroads of computational politics and intellectual history, the article abstracted here interrogates Chat-Hi:, the prototype of an interactive art installation using a natural language processing model trained on a large database of speeches by Indian prime ministers since 1946. Machine learning (ML)–powered, Chat-Hi: generates analytically intelligible answers to viewers’ questions, such as: What is the idea of India? The media installation acts as a conversational archive, gazing medium and interpretive layer [1]. It fosters playful, intimate, interactive and relational modes of historical attention [2], inviting participants to actualize the past through present political and social questioning. Building on existing accounts on the inclusive and modernist endeavor of the makers of postcolonial India, we interpret Chat-Hi:’s output to revisit historical argument by contrasting Jawaharlal Nehru’s emphasis on diversity with Narendra Modi’s stress on unity. Far from being anecdotal, this finding subtly signals the majoritarian turn of Indian democracy.With the help of Chat-Hi: and its human-machine question-answers digital interface, the article is concerned with the afterlives of historical figures, facetiously enjoining them to revisit a topical puzzle: What is the idea of India? Through interpellations of the sort addressed to Chat-Hi:, the article uses automated text generation trained on speeches by historical figures to convoke them into our turbulent times. We aim to interrupt the presentism [3] of politics to formulate answers, look for analogies [4] and—maybe—gather the courage to act from authoritative “specters” of the past [5]. Despite the fuzziness of modelgenerated texts, we insist on the usefulness of these political specters in facilitating the interpretation and identification of historical shifts in the representation of nationhood.The contrast between Prime Minister (PM) Jawaharlal Nehru’s (1946–1964) emphasis on diversity and PM Narendra Modi’s (2014–) focus on unity in Chat-Hi:’s outputs is a strong indication that the gradual political deepening of the Hindu nationalist project durably amends the statesponsored idea of India. The emphasis by Modi and his synthetic avatar on unity reflects the long-standing ideology of social harmony advocated by the Hindu nationalist movement, which gives primacy to a consolidated Hindu fold over the celebration of a multiplicity of religious and community identities. Machine-generated text by Modi affirms, for instance: “The reason for the culture of social cohesion and harmony in the society, just by doing research, I could clearly understand the processes.” Nehru’s contrary stress on diversity epitomizes a nationalism based on the normative belief [6] that what is to be represented as the people is not the Hindu majority but a more abstract collective bound by the principle of secular inclusiveness. This analysis finds echoes in Shiv Visvanathan’s study of India’s political modernity: “[Modi] erased a cosmopolitan India for a nationalist idea of the nation state, rejecting pluralism for predictability and uniformity, the univocality of patriotism. The Republic of Nehru sought plurality” [7]. The contrast between the two historical figures shows that there are at least two ideas of India, and that through conversing with automated specters of the past, the present can appear to us more clearly.The article further discusses the epistemic status of “human” insights derived from machine learning–generated content. We start by presenting the didactic potential of Chat-Hi:, in particular its ability to transmit historical emotions and ideas. Following this, we outline five potential methodological aids of the automated approach to historical inquiry. Chat-Hi: facilitates indeed a fresh encounter, thematic focus, comparability, confirmation and interactivity.The reader’s exposure to the computer-generated content triggers renewed attention, partly in the attempt to dispel ambiguity over the authenticity of the writing. Because puzzle, surprise and tension are feelings central to abductive reasoning [8] and interpretive research designs [9], they subject the reader to a fresh encounter [10] with characters such as Prime Ministers Atal Bihari Vajpayee or Chandra Shekhar. The speculative scrutiny of the synthetic text provides rich feedback, stimulating conversations between archival and machine-generated content. Such iterative analysis veers away from policy-oriented predicaments, focusing instead on historical insights. Second, the automated approach enables all the historical figures to answer the same question, thus offering thematically coherent text to the analyst. Responses to what the idea of India possibly is might venture into various political domains, referring to the country’s foreign policy or to its social issues, but they will ultimately give us clues about the ideological views of prime ministers on nationalism. Such structured output offers the possibility of comparing speakers, further facilitating the identification of contrastive and salient ideational as well as stylistic traits among them. Indeed, insights derived from ML speeches do not exist in a vacuum and can be further assessed in light of the original texts. This intertextual and contrastive approach can be further adjusted using more descriptive text analysis, which can ease the identification of lexical patterns among speakers in a given textual dataset.