Content

Speaker

Adam Coscia (Georgia Tech)

Image
A photo of Adam Coscia.
A photo of Adam Coscia.

Abstract

People rely on analysis tools like Excel every day to help them research, organize, and make important decisions with data. Today, those tools are increasingly AI-driven. If AI can retrieve, summarize, and explain our data for us, how should we rethink the way we build analysis tools?

In this talk, I will explore the trade-offs in building analysis tools that promote agency and communication between AI and human—from fully automating analysis to keeping humans in the driver's seat. By examining the different ways AI can be used, I argue that the best tools find a smart balance of using AI to handle the heavy lifting of finding and summarizing relevant data, while relying on human intuition and judgement at critical decision points. Towards this vision, I will demo two new prototype research systems for AI-driven visual data analysis. The first, VisPile, explores agency by integrating an LLM and a knowledge graph (KG) into unstructured text analysis, enabling rapid exploration of competing hypotheses and validation of AI-generated evidence while reducing time spent reading. The second, Debrief, explores communication by embedding an LLM into a visualization dashboard to control, modify, and explain the views in real-time. Finally, I'll step back and reflect on my broader vision for new research directions in AI-driven data science.

In person event posted in Research