Embedded Files
Scideator
Scideator
Objective: to support scientists in exploring potential research ideas and evaluating them for novelty
Methods: lo-fi prototyping (mockups), system development (RAG-augmented human-LLM workflow), lab user study (survey, interview, interaction logging), data/statistical analysis, thematic analysis
Tools: Google Slides, Figma, React, TypeScript, Python, LLM APIs, Semantic Scholar API, R, Google Sheets
Key Insight: Scideator (LLM tool for facet-based ideation) significantly increases creativity support compared to a strong baseline
Marissa Radensky, Simra Shahid, Raymond Fok, Pao Siangliulue, Daniel S. Weld, Tom Hope (in submission)
Where: University of Washington, collaborating with AI2
Papers-to-Posts
Papers-to-Posts
Objective: to assist scientists in translating research papers into blog posts
Methods: lo-fi/hi-fi prototyping (mockups, coded), pilot study (think-aloud, iterative design), system development (human-LLM workflow), lab and deployment user studies (survey, interview, interaction logging), data/statistical analysis, thematic analysis
Tools: Google Slides, React, TypeScript, Python, LLM API, Semantic Scholar API, R, Google Sheets, Jupyter Notebooks
Key Insight: Papers-to-Posts (LLM tool with interactive reverse source outline) significantly increases writers' blog post satisfaction under time constraints, compared to a strong baseline
Marissa Radensky, Daniel S. Weld, Joseph Chee Chang, Pao Siangliulue, Jonathan Bragg (in submission)
Where: AI2 internship
AI Recommendation Confidence Patterns
AI Recommendation Confidence Patterns
Objective: to help music listeners understand their song recommendations
Methods: lo-fi prototyping (mockups), Wizard-of-Oz user study (survey, interview), data/statistical analysis, thematic analysis
Tools: Google Slides, R, Google Sheets
Key Insight: trust-related design guidelines for confidence signals in conversational recommender systems (e.g., consider both quality and novelty expectations when designing confidence levels)
Paper: “I Think You Might Like This”: Exploring Effects of Confidence
Signal Patterns on Trust in and Reliance on Conversational
Marissa Radensky, Julie Anne Séguin, Jang Soo Lim, Kristen Olson, Robert Geiger (FAccT 2023)
Where: Google internship
Anomalous AI Input/Output Alerts
Anomalous AI Input/Output Alerts
Objective: to improve AI-augmented clinical decision-making (use case: physicians using AI to detect pneumonia in chest x-rays)
Methods: lo-fi prototyping (mockups), formative study (think-aloud, survey, interview), dataset curation, image classification, survey study, data/statistical analysis, thematic analysis
Tools: Microsoft PowerPoint, PyTorch, R, Microsoft Excel
Key Insight: AI anomaly alerts are desired by physicians, but no evidence indicating that they improve physician-AI team accuracy
Paper: Exploring How Anomalous Model Input and Output Alerts Affect Decision-Making in Healthcare
Marissa Radensky, Dustin Burson, Rajya Bhaiya, and Daniel S. Weld (CHI 2022 Workshop on Trust and Reliance in AI-Human Teams)
Where: Microsoft internship
Explainable AI Scholar Recommendations
Explainable AI Scholar Recommendations
Objective: to support scientists in discovering scholars of interest
Methods: formative study (think-aloud, interview, survey), lab user study (survey, interview), part of data analysis, thematic analysis
Tools: Google Slides, Google Sheets, Python
Key Insight: Bridger (facet-based AI scholar recommender system) helps surface recommendations useful for generating novel research directions
Paper: Bursting Scientific Filter Bubbles: Boosting Innovation via Novel Author Discovery
Jason Portenoy, Marissa Radensky, Jevin West, Eric Horvitz, Daniel S. Weld, and Tom Hope (CHI 2022)
Where: University of Washington, collaborating with AI2
Local and Global AI Explanations
Local and Global AI Explanations
Objective: to assist scientists in understanding and personalizing their research-paper feeds
Methods: lo-fi/hi-fi prototyping (paper, mockups, coded), formative study (interview), exploratory and controlled user studies (think-aloud, survey, interaction logging), dataset curation, data/statistical analysis, thematic analysis
Tools: Google Slides, React, TypeScript, Python, R, Google Sheets
Key Insight: combining local and global explanations may help users explain how to improve recommendations better than either alone
Paper: Exploring the Role of Local and Global Explanations in Recommender Systems
Marissa Radensky, Doug Downey, Kyle Lo, Zoran Popović, and Daniel S. Weld (CHI 2022 Late-Breaking Work)
Where: AI2 internship
Page updated
Google Sites
Report abuse