Global South Representation in Generative AI
What the Annotators Don't Know can Harm Us
With the rapid proliferation of artificial intelligence, there is growing concern over its potential to exacerbate existing biases and societal disparities and introduce novel ones. This issue has prompted widespread attention from academia, policymakers, industry, and civil society. While evidence suggests that integrating human perspectives can mitigate bias-related issues in AI systems, it also introduces challenges associated with cognitive biases inherent in human decision-making. Our research focuses on reviewing existing methodologies and ongoing investigations aimed at understanding annotation attributes that contribute to bias.
Harms in Text-to-Image Generation with focus on Global South
With the increasing integration of advanced generative models like Gemini and GPT in AI-as-a-Service (AIaaS) systems, concerns have emerged about these models' biases, particularly favoring majority demographics across sociodemographic lines. Despite calls for more diverse media representation, marginalized racial and ethnic groups continue to face misrepresentation, stereotyping, and exclusion within the AIaaS ecosystem. This work offers a critical review of research addressing these social harms and presents open-ended research questions to stimulate further inquiry and guide future efforts in mitigating biases in AI systems.
Nationality Bias Detection in Text Generation
This research investigates nationality biases in NLP models and their impact on fairness and justice in AI systems. Using a mixed-methods approach, the study quantitatively measures bias in AI-generated articles and qualitatively analyzes its implications through interviews. Findings show that biased NLP models amplify societal biases, potentially leading to harm in sociotechnical settings. The qualitative analysis reveals readers' altered perceptions of countries influenced by biased articles. The research emphasizes the importance of addressing biases in AI systems, correcting them to ensure ethical and equitable deployment, and recognizing the role of public perception in shaping AI's societal impact.
Social Media for Civic Awareness
Understanding the Workings of 911 Centres during Covid-19
Implementing these COOP plans required PSAPs to decentralize operations by moving staff out of “one big room” to multiple workspaces across primary and alternative facilities, including people’s homes. Unsurprisingly, our interviews highlight disruptions to shared physical spaces, or social infrastructures, caused by outbreaks of infectious disease like COVID-19, and the vulnerability of essential functions when performed by people in centralized, collocated workspaces.
Using Twitter for Crisis Response
During the pandemic social media gained special interest as it went on to become an important medium of communication. This made the information being relayed on these platforms especially critical. In our work, we aim to explore identification of fake news, misinformation and using Twitter for real-time information sharing. Our study is useful in establishing the role of Twitter, and social media, during a crisis, and more specifically during crisis management.
What’s Political on TikTok?
This study explores user exposure to political content on TikTok through a custom tool that tracks videos viewed by 358 participants and captures their perceptions. Findings show political content makes up about 13.7% of users' feeds, with demographics—particularly age, education, and political views—being key factors in exposure levels.
Learning Community Designs
Applying Social Constructivist Theories in Adaptive Learning
My thesis work revolves around applying social constructivist theories in the world of adaptive learning. With this work, I aim to develop adaptive learning systems that not only account for student performance but also further the goal of social learning as described by Bandura. This should be useful in the context of online learning both as motivation as well as for improved learning.
AI Curriculum Design
There are currently multiple undergraduate or certificate programs that are made up of specialized courses designed to meet the new found demand for AI professionals. However, there has been a mismatch between what is taught in professional degrees and what is needed in the job market. To this effect, our work explores the needs and demands of the industry through the eyes of AI experts.
We Are One: Study on Distributed Communities
Options for students to learn and connect with each other have diversified in recent years, with online resources and campuses playing an increasing role. Nonetheless, students want to feel sense of community with their peers and instructors; institutional bonds are in turn associated with enhanced learning. We explore the feelings of community among students studying at a geographically distributed university.
YouTube User Data Modelling
YouTube is one of the most popular websites. To better understand the characteristics and impact of YouTube on education, we analyzed a popular YouTube channel. Our analysis provides valuable information that can have major technical and commercial implications in the field of education. We perform in-depth time-series analysis of the channel data to reveal the trend, seasonality and temporal pattern for the educational videos on YouTube.
Plant Village (NLP supported learning tool for African Farmers)
PlantVillage has developed a triple A model (Algorithmic Agricultural Advice) that works to increase the yield and profitability for millions of farmers. It is our goal to reach hundreds of millions in partnership with an ecosystem of farmer facing organizations and the farmers themselves. Our algorithms come from our integration of AI, satellite technology and our unique field force (the Dream Team).