Current Projects Academic Year 2026-27

HuskyAI: An AI Coaching Platform for Strengthening Human Reasoning and Agency

HuskyAI is an AI coaching platform currently in active development, built on a premise that grows more pressing as AI systems become everyday intellectual companions: the skill that matters most is not fluency in using them, but the strength and independence of the reasoning a person brings to the exchange.

Rather than teaching prompt engineering, HuskyAI challenges students with authentic exercises and evaluates the quality of their reasoning as they work with AI—how effectively they question, test, and judge model outputs rather than how cleverly they formulate prompts.

This assessment is being developed around an evolving framework called the Platform Engagement Index (PEI), which seeks to move beyond surface-level tool usage and instead measure underlying habits of mind. A cohort-level view allows students to track how their reasoning develops relative to their peers over time.

The broader goal is formative: to make the quality of thinking visible, measurable, and improvable while cultivating the judgment and agency required to remain in control of AI interactions rather than deferring to the machine.

As HuskyAI matures, it seeks to reframe AI literacy as a question of epistemic integrity—developing individuals who can think with AI systems without surrendering their thinking to them.

AI News Coverage Tracker: A Storybench + Media Cloud Project

AI News Coverage Tracker is a computational instrument currently in development for the longitudinal study of how news media represent artificial intelligence. The project is a collaboration with Media Cloud and Northeastern's Storybench.

It draws its corpus from Media Cloud's open-source archive, which continuously collects content from tens of thousands of online news sources. Analysis is orchestrated using Media Cloud's Python-based sous-chef analytics pipeline through scheduled workflows for discovering AI-related coverage, executing analytical tasks, and aggregating daily results.

Planned analytical layers include:

• Semantic embedding and dynamic topic modeling
• Framing, stance, and hype classification
• Source and quotation extraction
• Near-duplicate and semantic similarity detection

Rather than producing static snapshots, the tracker is designed around change. Interactive dashboards will surface emerging narratives, burst detection, and change-point alerts to reveal shifts in how AI is framed, who is granted epistemic authority, and how uniformly stories propagate across outlets.

The project approaches AI coverage not as a topic to summarize, but as an evolving information ecosystem whose diversity and susceptibility to hype directly influence public understanding.

Glass Ledger: An AI-Assisted Investigation of Massachusetts Campaign Finance

Every campaign dollar raised in Massachusetts leaves a public record: who contributed, how much was given, when it was donated, and where the money was ultimately spent.

Glass Ledger combines AI-assisted text analysis and data visualization to make these financial flows transparent and investigate anomalies and accountability questions hidden within them.

The project draws data from the Massachusetts Office of Campaign and Political Finance (OCPF), focusing on:

• Political contributions
• Campaign expenditures

At its core is a collection of specialized analytical agents, each examining the data from different perspectives to identify patterns that might otherwise remain hidden, including unusual spending categories, suspicious timing, and financial outliers.

Two primary outputs are being developed:

• An interactive dashboard for exploring spending patterns by politician and category
• An analytical report explaining discovered anomalies and contextualizing findings

Glass Ledger ultimately aims to transform a sprawling public record into an accessible instrument of civic accountability.

Hyperlocal AI Video News Experiments

Hyperlocal AI Video News Experiments is an early-stage research initiative exploring the production of short-form, social-media-native video news from real-time civic data at a level of geographic specificity that traditional local newsrooms can no longer sustain.

The project is a collaboration with Northeastern's Reinventing Local TV News Project and focuses on metro Boston.

It integrates three streams of municipal data:

• 311 service requests
• Weather data
• Traffic information

An agentic pipeline transforms a single data point into a finished video story—from a pothole reported on a specific street to neighborhood-level weather forecasts and localized traffic disruptions.

The system is being designed around station-specific editorial voices learned from historical news archives, along with reusable AI-generated anchor avatars that preserve a recognizable brand identity.

Because the project produces synthetic news content, it intentionally engages with questions surrounding disclosure, consent, and authenticity rather than avoiding them.

The ultimate goal is to evaluate whether agentic AI can responsibly extend trustworthy local journalism into hyperlocal coverage gaps.

Public Perceptions of AI in Media Across the U.S., U.K., and Canada

Public Perceptions of AI in Media is an ongoing cross-national survey study tracking attitudes toward AI-driven content moderation in the United States, United Kingdom, and Canada.

The 2026 survey wave (n = 1,000 participants per country) extends a baseline study conducted in August 2023 and presented at the Trust and Safety Research Conference.

The study examines public acceptance of interventions such as:

• Algorithmic flagging
• Warning labels
• Conversational AI assistants
• Automated suspensions without human review

Findings from the 2023 baseline revealed that traditional demographic variables were largely non-predictive, while self-reported AI usage emerged as the strongest predictor of acceptance across most categories.

The 2026 study investigates changes against a transformed technological and regulatory backdrop, including:

• Mainstream adoption of large language models
• The proliferation of AI-generated misinformation
• Deployment of automated moderation systems
• Emerging global AI regulations

By tracing how public attitudes evolve as AI becomes embedded in everyday life, the project aims to provide empirical foundations for future governance and platform policy discussions.

Computational Methods for the AI Skeptical

Vaishnavi Sarmalkar (graduate researcher) and Prof. Katherine Haenschen

This project investigates how U.S. newspapers have represented "Relational Organizing" over the past 25 years. Relational Organizing broadly refers to organizing practices built on personal relationships, spanning grassroots community organizing through one-on-one meetings and faith-based groups, to tech-driven political mobilization using apps and friend-to-friend voter outreach.

To analyze a corpus of newspaper articles at scale, we are building a computational pipeline that uses large language models to automatically classify articles against a human-validated codebook. Claude's labels are validated against human coding using Cohen's kappa before running on the full corpus, ensuring the approach meets established standards for reliability in computational content analysis.

The project sits at the intersection of AI and media research, using modern NLP tools to trace how a political concept has been framed and understood in American journalism over time.

“It’s not all going to be good, but it’s going to be different:” Community News Editors Respond to the Implementation of an AI-Supported Content Management System

Prof. Meg Heckman, Caleb Okereke (doctoral researcher), Prof. John Wihbey

This project explores findings from a year-long qualitative study documenting how editors at a U.S. family-owned newspaper chain are responding to the implementation of an AI-supported CMS and other management efforts to promote GenAI adoption.

Drawing on two rounds of interviews conducted one and six months after the system's launch, the study examines shifts in production routines and newsroom attitudes toward GenAI. Four themes emerged: the new system as a catalyst for broader experimentation with AI tools, a desire to amplify the human dimensions of journalism, ethical tensions centered on transparency, and a pattern of task type dictating tool adoption.

Findings reveal a paradox in which community journalists are experimenting with GenAI while simultaneously questioning what the technology means for journalistic independence, audience trust, and long-term sustainability. The study contributes to scholarship on technological disruption in community journalism and offers practical insights for news organizations navigating AI adoption.