Gagan Bansal

Gagan Bansal

Auto-generated profile, based on publicly available information

Gagan Bansal

Microsoft Research, United States

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 September 2023 in Proceedings of the ACM on Human-Computer Interaction
Verified
Journal Article
Open access
VCValerie Chen
Valerie Chen
QLQ. Vera Liao
Q. Vera Liao
JVJennifer Wortman Vaughan
Jennifer Wortman Vaughan
GBGagan Bansal
Gagan Bansal
AI explanations are often mentioned as a way to improve human-AI decision-making, but empirical studies have not found consistent evidence of explanations' effectiveness and, on the contrary, suggest that they can increase overreliance when the AI system is wrong. While many factors may affect reliance on AI support, one important factor is how decision-makers reconcile their own intuition---beliefs or heuristics, based on prior knowledge, experience, or pattern recognition, used to make judgments---with the information provided by the AI system to determine when to override AI predictions. We conduct a think-aloud, mixed-methods study with two explanation types (feature- and example-based) for two prediction tasks to explore how decision-makers' intuition affects their use of AI predictions and explanations, and ultimately their choice of when to rely on AI. Our results identify three types of intuition involved in reasoning about AI predictions and explanations: intuition about the task outcome, features, and AI limitations. Building on these, we summarize three observed pathways for decision-makers to apply their own intuition and override AI predictions. We use these pathways to explain why (1) the feature-based explanations we used did not improve participants' decision outcomes and increased their overreliance on AI, and (2) the example-based explanations we used improved decision-makers' performance over feature-based explanations and helped achieve complementary human-AI performance. Overall, our work identifies directions for further development of AI decision-support systems and explanation methods that help decision-makers effectively apply their intuition to achieve appropriate reliance on AI.

11 May 2024
Verified
Conference Paper
Open access
HMHussein Mozannar
Hussein Mozannar
GBGagan Bansal
Gagan Bansal
AFAdam Fourney
Adam Fourney
EHEric Horvitz
Eric Horvitz
Code-recommendation systems, such as Copilot and CodeWhisperer, have the potential to improve programmer productivity by suggesting and auto-completing code. However, to fully realize their potential, we must understand how programmers interact with these systems and identify ways to improve that interaction. To seek insights about human-AI collaboration with code recommendations systems, we studied GitHub Copilot, a code-recommendation system used by millions of programmers daily. We developed CUPS, a taxonomy of common programmer activities when interacting with Copilot. Our study of 21 programmers, who completed coding tasks and retrospectively labeled their sessions with CUPS, showed that CUPS can help us understand how programmers interact with code-recommendation systems, revealing inefficiencies and time costs. Our insights reveal how programmers interact with Copilot and motivate new interface designs and metrics.

25 April 2025
Verified
Conference Paper
Open access
WEWill Epperson
Will Epperson
GBGagan Bansal
Gagan Bansal
VDVictor C Dibia
Victor C Dibia
AFAdam Fourney
Adam Fourney
JGJack Gerrits
Jack Gerrits
SASaleema Amershi
Saleema Amershi

29 October 2024 in ACM Transactions on Computer-Human Interaction
Verified
Journal Article
Open access
HVHelena Vasconcelos
Helena Vasconcelos
GBGagan Bansal
Gagan Bansal
AFAdam Fourney
Adam Fourney
QLQ. Vera Liao
Q. Vera Liao
JVJennifer Wortman Vaughan
Jennifer Wortman Vaughan
Large-scale generative models have enabled the development of AI-powered code completion tools to assist programmers in writing code. Like all AI-powered tools, these code completion tools are not always accurate and can introduce bugs or even security vulnerabilities into code if not properly detected and corrected by a human programmer. One technique that has been proposed and implemented to help programmers locate potential errors is to highlight uncertain tokens. However, little is known about the effectiveness of this technique. Through a mixed-methods study with 30 programmers, we compare three conditions: providing the AI system's code completion alone, highlighting tokens with the lowest likelihood of being generated by the underlying generative model, and highlighting tokens with the highest predicted likelihood of being edited by a programmer. We find that highlighting tokens with the highest predicted likelihood of being edited leads to faster task completion and more targeted edits, and is subjectively preferred by study participants. In contrast, highlighting tokens according to their probability of being generated does not provide any benefit over the baseline with no highlighting. We further explore the design space of how to convey uncertainty in AI-powered code completion tools and find that programmers prefer highlights that are granular, informative, interpretable, and not overwhelming. This work contributes to building an understanding of what uncertainty means for generative models and how to convey it effectively.

19 April 2023
Verified
Conference Paper
Open access
GBGagan Bansal
Gagan Bansal
ZBZana Buçinca
Zana Buçinca
KHKenneth Holstein
Kenneth Holstein
JHJessica Hullman
Jessica Hullman
ASAlison Marie Smith-Renner
Alison Marie Smith-Renner
SWSherry Wu
Sherry Wu
As humans increasingly interact (and even collaborate) with AI systems during decision-making, creative exercises, and other tasks, appropriate trust and reliance are necessary to ensure proper usage and adoption of these systems. Specifically, people should understand when to trust or rely on an algorithm’s outputs and when to override them. Significant research focus has aimed to define and measure trust in human-AI interaction, and design and implement interactions that promote and calibrate trust. However, conceptualizing trust and reliance, and identifying the best ways to measure these constructs and effectively shape them in human-AI interactions remains a challenge, especially across contexts and domains. This workshop aims to establish building appropriate trust and reliance on (imperfect) AI systems as a vital, yet under-explored research problem. The workshop will provide a venue for exploring three broad aspects related to human-AI trust: (1) How do we clarify definitions and frameworks relevant to human-AI trust and reliance (e.g., what does trust mean in different contexts)? (2) How do we measure trust and reliance? And, (3) How do we shape trust and reliance? The workshop will build on the success from running it at CHI 2022,1 with a focus on “Learning from Practice" — how can we better tie theory-building to real-life use cases? As these problems and solutions involving humans and AI are interdisciplinary in nature, we invite participants with expertise in HCI, AI, ML, psychology, and social science, or other relevant fields to foster closer communications and collaboration between multiple communities.

11 May 2024
Verified
Conference Paper
Open access
ZAZahra Ashktorab
Zahra Ashktorab
GBGagan Bansal
Gagan Bansal
ZBZana Buçinca
Zana Buçinca
KHKenneth Holstein
Kenneth Holstein
JHJessica Hullman
Jessica Hullman
WZWenjuan Zhang
Wenjuan Zhang
State-of-the-art AIs, including Large Language Models (LLMs) like GPT-4, now possess capabilities once unique to humans, such as coding, idea generation, and planning. Advanced AIs are now integrated into a plethora of platforms and tools, including GitHub Copilot, Bing Chat, Bard, ChatGPT, and Advanced Data Analytics. In contrast to conventional, specialized AIs that typically offer singular solutions, these LLMs redefine human-AI dynamics, with a growing trend toward humans viewing them as collaborative counterparts. This shift leads to enhanced dialogues, negotiations, and task delegation between humans and AI. With these rapid advancements, the nature of human roles in the AI collaboration spectrum is evolving. While our previous workshops CHI TRAIT 2022 and 2023 delved into the trust and reliance concerning traditional AIs, the pressing question now is: how should we measure trust and reliance with these emerging AI technologies? As these systems witness widespread adoption, there’s also a need to assess their impact on human skill development. Does AI assistance amplify human skill progression, or does it inadvertently inhibit it? Considering the multifaceted challenges and solutions that revolve around human-AI interactions, we invite experts from diverse fields, including HCI, AI, ML, psychology, and social science. Our aim is to bridge communication gaps and facilitate rich collaborations across these domains.