The reference to assumptions in how practitioners use or interact with machine learning (ML) systems is ubiquitous in HCI and responsible ML discourse.However, what remains unclear from prior works is the conceptualization of assumptions and how practitioners identify and handle assumptions throughout their workflows.This leads to confusion about what assumptions are and what needs to be done with them.We use the concept of an argument from Informal Logic, a branch of Philosophy, to offer a new perspective to understand and explicate the confusions surrounding assumptions.Through semi-structured interviews with 22 ML practitioners, we find what contributes most to these confusions is how independently assumptions are constructed, how reactively and reflectively they are handled, and how nebulously they are recorded.Our study brings the peripheral discussion of assumptions in ML to the center and presents recommendations for practitioners to better think about and work with assumptions.

Though ML practitioners increasingly employ various Responsible ML (RML) strategies, their methodological approach in practice is still unclear. In particular, the constraints, assumptions, and choices of practitioners with technical duties–such as developers, engineers, and data scientists—are often implicit, subtle, and under-scrutinized in HCI and related fields. We interviewed 22 technically oriented ML practitioners across seven domains to understand the characteristics of their methodological approaches to RML through the lens of ideal and non-ideal theorizing of fairness. We find that practitioners’ methodological approaches fall along a spectrum of idealization. While they structured their approaches through ideal theorizing, such as by abstracting RML workflow from the inquiry of applicability of ML, they did not systematically document nor pay deliberate attention to their non-ideal approaches, such as diagnosing imperfect conditions. We end our paper with a discussion of a new methodological approach, inspired by elements of non-ideal theory, to structure technical practitioners’ RML process and facilitate collaboration with other stakeholders.

The death of Indian film star Sushant Singh Rajput at the peak of the COVID lockdown triggered chaos on the news cycle in India with a range of conspiracy theories that led to a witch hunt of sorts, and the hounding of several entertainers and public figures in the months that followed. Using data from Twitter, YouTube, and an archive of debunked misinformation stories, we examine the drivers and consequences of social media outrage in this case. We analyse these patterns from the framework of conspiracy and astroturfing and contextualize our findings to the socio-political background currently prevalent in India. Primarily, retweet rates on Twitter suggest that commentators benefited from talking about the case, which got higher engagement than other topics. Moreover, we report evidence of political hands in the way the discourse has shaped online, but more importantly that the story bears warnings for the shape and impact of witch-hunts in the backdrop of a fractured media environment. In conclusion, we consider the effects of Rajput’s outsider status as a small-town implant in the film industry within the broader narrative of systemic injustice, as well as the gendered aspects of mob justice that have taken aim at his former partner in the months since.

Celebrity influencers are increasingly central to political discourse as they engage in, and get engaged with, on matters of electoral importance. In this paper, using Twitter data from 1432 sportspersons and entertainers and their engagement with the 1000 of the most followed ruling party and opposition politicians from India, we propose a new method to measure partisanship of celebrities along different modes of engagement. Our examination of polarization, through topical and retweet analyses, shows patterns related to both party incumbency and the level of internal organization. We find that the ruling BJP has been more effective than the opposition, the INC, in organized outreach to celebrities, by eschewing explicit party-based partisanship, and instead employing non-partisan narrative techniques, such as maintaining nationalism as the central theme in tweets. We find that while entertainers are equally engaged by both the ruling and opposition parties, sportspersons, who often enjoy a nationalist appeal by virtue of representing the country, tend to have a much more partisan relationship with the incumbent party.

Twitter is increasingly important for political outreach and networking around the world. While electoral politics and social relations in India are heavily organized by caste, a broader rhetoric of castelessness among upper-caste politicians has led to the eschewing of caste publicly to appear strategically secular. This has rendered caste dynamics more implicit than explicit. Social media, often cited as a tool for inclusion, offers a unique look into the networks of covert exclusion. Our study analyzes three structural properties of the Twitter network of Members of Parliament in India - influence, bridging capital, and mutual connectivity, to understand how caste manifests as social capital in the information economy. Our results show that those higher in the caste hierarchy are structurally poised for higher social capital through higher influence, incoming bridging capital, and higher propensity for mutual connections with other MPs in the network. Our study offers a methodological window into these invisible relations to show how structural advantages of Brahmanical supremacy are being co-produced and stabilized on social media at the highest level of politics.

Feature attributions and counterfactual explanations are popular approaches to explain a ML model. The former assigns an importance score to each input feature, while the latter provides input examples with minimal changes to alter the model's predictions. To unify these approaches, we provide an interpretation based on the actual causality framework and present two key results in terms of their use. First, we present a method to generate feature attribution explanations from a set of counterfactual examples. These feature attributions convey how important a feature is to changing the classification outcome of a model, especially on whether a subset of features is necessary and/or sufficient for that change, which attribution-based methods are unable to provide. Second, we show how counterfactual examples can be used to evaluate the goodness of an attribution-based explanation in terms of its necessity and sufficiency. As a result, we highlight the complimentary of these two approaches. Our evaluation on three benchmark datasets --- Adult-Income, LendingClub, and German-Credit --- confirms the complimentary. Feature attribution methods like LIME and SHAP and counterfactual explanation methods like Wachter et al. and DiCE often do not agree on feature importance rankings. In addition, by restricting the features that can be modified for generating counterfactual examples, we find that the top-k features from LIME or SHAP are often neither necessary nor sufficient explanations of a model's prediction. Finally, we present a case study of different explanation methods on a real-world hospital triage problem.

In the Global South, mobile airtime payment has emerged as a popular way to incentivize different research studies, including ones on survey completion or disseminating information to people. Building on this literature, we report deployment experiences from three different studies in India that used airtime incentives. The first was used to promote awareness about HIV/AIDS, the second for promoting awareness and surveying preparedness for an upcoming election, and the third to measure learning and encourage people to vote in a conflict-hit region for a different election. Unlike past work, we found that a delivery mechanism that focuses on asking questions first, rather than presenting a tutorial and then asking questions, worked well in practice. In addition, we found multiple challenges in adoption of the technology and tried different ways to incentivize peer sharing of our system. Between the three deployments, we also addressed other technical and human-centered challenges such as delayed airtime payments and people using the system on behalf of someone else. We hope that our experiences and insights can be helpful to others seeking to deploy applications that utilize mobile airtime payments for learning, sharing, and survey response.

The primary obstacle to developing technologies for low-resource languages is the lack of usable data. In this paper, we report the adaption and deployment of 4 technology-driven methods of data collection for Gondi, a low-resource vulnerable language spoken by around 2.3 million tribal people in south and central India. In the process of data collection, we also help in its revival by expanding access to information in Gondi through the creation of linguistic resources that can be used by the community, such as a dictionary, children’s stories, an app with Gondi content from multiple sources and an Interactive Voice Response (IVR) based mass awareness platform. At the end of these interventions, we collected a little less than 12,000 translated words and/or sentences and identified more than 650 community members whose help can be solicited for future translation efforts. The larger goal of the project is collecting enough data in Gondi to build and deploy viable language technologies like machine translation and speech to text systems that can help take the language onto the internet.

We studied the topical preferences of social media campaigns of India's two main political parties by examining the tweets of 7382 politicians during the key phase of campaigning between Jan - May of 2019 in the run up to the 2019 general election. First, we compare the use of self-promotion and opponent attack, and their respective success online by categorizing 1208 most commonly used hashtags accordingly into the two categories. Second, we classify the tweets applying a qualitative typology to hashtags on the subjects of nationalism, corruption, religion and development. We find that the ruling BJP tended to promote itself over attacking the opposition whereas the main challenger INC was more likely to attack than promote itself. Moreover, while the INC gets more retweets on average, the BJP dominates Twitter's trends by flooding the online space with large numbers of tweets. We consider the implications of our findings hold for political communication strategies in democracies across the world.

The Hindu caste system plays an important role in the socio-political landscape of India. In recent years, Indian politicians have moved a lot of direct communication online. Social media is now an important space for the articulation and performance of their political positions. In this paper, we study ways in which the political performance of caste relations can be captured from the online connections and messaging of parliamentarians in India. We run tests of odds ratios among the members of LokSabha (lower house) of India to find the extent to which their engagement is insular to their own caste group versus other groups. We observe that in the LokSabha network, Members of Parliament (MPs) have higher odds of getting retweeted by others whose caste is the same or closer to their own in the caste hierarchy. The findings of this research shed light on an understudied, yet critical, social relation of caste in the study of political behavior on social media.

Programs that reward people for referring their friends are increasingly being used to raise awareness about important topics. With a fixed budget for referral incentives, a natural goal for such referral programs is to maximize the number of people reached. Unlike a typical influence maximization problem, however, the social network of potential adopters is unknown apriori. Further, people's response to a referral incentive can depend on various factors such as their preference for the content, size of their social network, and their estimated value for sharing. Therefore, we introduce an incentive-aware variant of the influence maximization problem and formalize it under an online learning setting. Given the lack of initial information about the social network or how people respond to referral incentives, we use an explore-exploit strategy and present a contextual bandit agent CoBBI that optimizes the incentives for each user by learning from the results of its past actions. We demonstrate the effectiveness of CoBBI on data from a real-world referral program for raising land rights' awareness among farmers. Compared to a wide range of baselines, we find that CoBBI is consistently more cost-effective, across a wide range of influence probabilities and people's response to incentives.