Periodic Reporting for period 5 - TheSocialBusiness (The advantages and pitfalls of elicitated online user engagement)
Berichtszeitraum: 2024-06-01 bis 2025-05-31
TheSocialBusiness - The advantages and pitfalls of elicited Online User Engagement
The research led by Prof. Gal Oestreicher-Singer is consisted of several aims that are manifested in a variety of projects. Below you can see the abstract of the project published since our last scientific report.
"Implicit bias in LLMs: Bias in financial advice based on implied gender"
• join work with Prof. Inbal Yahav and Dr. Shir Etgar
• Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4880335(öffnet in neuem Fenster)
Abstract: For the first time in human history, the era of Large Language Models (LLMs) has enabled humans to communicate directly with AIs in conversation-like interactions. For efficient communication, people are encouraged to prompt LLMs with contextual information. However, previous research in machine learning indicates that such information can reveal implicit group affiliations. This study explores whether implied gender affiliation, conveyed through stereotypically gendered professions, affects AI responses to financial advice-seeking prompts. Using GPT-4, we initiated 2,400 financial advice-seeking interactions with an LLM. Each prompt included either feminine or masculine gender cues. We found that advice given to implied women was less risky, more prevention-oriented, and more simplified and patronizing in tone and wording than advice given to implied men. These findings call attention to implicit biases in LLMs, which are more challenging to identify and debias than biases based on explicit group affiliation, and which could have tremendous societal implications.
The research led by Prof. Gal Oestreicher-Singer is consisted of several aims that are manifested in a variety of projects. Below you can see the abstract of the project published since our last scientific report.
"Implicit bias in LLMs: Bias in financial advice based on implied gender"
• join work with Prof. Inbal Yahav and Dr. Shir Etgar
• Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4880335(öffnet in neuem Fenster)
Abstract: For the first time in human history, the era of Large Language Models (LLMs) has enabled humans to communicate directly with AIs in conversation-like interactions. For efficient communication, people are encouraged to prompt LLMs with contextual information. However, previous research in machine learning indicates that such information can reveal implicit group affiliations. This study explores whether implied gender affiliation, conveyed through stereotypically gendered professions, affects AI responses to financial advice-seeking prompts. Using GPT-4, we initiated 2,400 financial advice-seeking interactions with an LLM. Each prompt included either feminine or masculine gender cues. We found that advice given to implied women was less risky, more prevention-oriented, and more simplified and patronizing in tone and wording than advice given to implied men. These findings call attention to implicit biases in LLMs, which are more challenging to identify and debias than biases based on explicit group affiliation, and which could have tremendous societal implications.
Over the course of the project, I have made substantial progress across all three aims, supported by a strong team of PhD students, postdoctoral researchers, and research assistants. Together we advanced ongoing projects, initiated new research streams, and brought several studies to completion and publication. The project has now reached a mature stage, with most initiatives either published, accepted, or under advanced review at leading journals.
Aim 1 (Online Engagement in Different Contexts):
We conducted multiple studies exploring engagement in diverse digital settings.
• Mobile context: Research on user engagement with videos and conversion formed the basis of a PhD dissertation and is now in the final stages of journal submission, with a related industry collaboration underway.
• Crowdfunding: Our work on entrepreneur–funder engagement was published in MIS Quarterly, with a follow-up article for managerial audiences. This line of research supported a postdoctoral fellow who has since secured a tenure-track position.
• Generative AI: We completed and submitted a paper on AI-mediated advice seeking, presented it at top conferences and workshops, and received wide media coverage. A related project on human–AI interaction with ChatGPT is ongoing. ,
• In addition, the project previously reported as preliminary results was accepted at Marketing Letters and featured in INSEAD Knowledge.
Aim 2 (Mechanisms of Engagement):
We advanced research on how users perceive AI-based tools on digital platforms. This project was presented at major conferences and is under review at a leading marketing journal. It also serves as the foundation for an upcoming PhD dissertation.
Aim 3 (Trust and Disclosure):
Our research progressed from preliminary findings to a full publication in MIS Quarterly, demonstrating how social cues and trust cues jointly shape willingness to disclose private information online. This project supported two postdoctoral researchers, one of whom was a co-author.
Dissemination and Exploitation:
The results have been actively disseminated through top-tier journal publications, conference presentations, invited talks, and media coverage. Findings have been shared with both academic and practitioner audiences, ensuring impact beyond academia. Collaborations with industry partners (e.g. mobile app developers) enhance the practical exploitation of results, while international collaborations with leading scholars expand the project’s global reach.
Overall progress:
The project delivered on its proposed aims, significantly advancing knowledge on online engagement, mechanisms of interaction, and trust in digital platforms. Despite challenges posed by COVID-19 and later the war in Israel, the team maintained strong dissemination and collaboration activities. Many of the researchers trained through this project have transitioned into successful academic careers, further demonstrating its long-term impact.
Aim 1 (Online Engagement in Different Contexts):
We conducted multiple studies exploring engagement in diverse digital settings.
• Mobile context: Research on user engagement with videos and conversion formed the basis of a PhD dissertation and is now in the final stages of journal submission, with a related industry collaboration underway.
• Crowdfunding: Our work on entrepreneur–funder engagement was published in MIS Quarterly, with a follow-up article for managerial audiences. This line of research supported a postdoctoral fellow who has since secured a tenure-track position.
• Generative AI: We completed and submitted a paper on AI-mediated advice seeking, presented it at top conferences and workshops, and received wide media coverage. A related project on human–AI interaction with ChatGPT is ongoing. ,
• In addition, the project previously reported as preliminary results was accepted at Marketing Letters and featured in INSEAD Knowledge.
Aim 2 (Mechanisms of Engagement):
We advanced research on how users perceive AI-based tools on digital platforms. This project was presented at major conferences and is under review at a leading marketing journal. It also serves as the foundation for an upcoming PhD dissertation.
Aim 3 (Trust and Disclosure):
Our research progressed from preliminary findings to a full publication in MIS Quarterly, demonstrating how social cues and trust cues jointly shape willingness to disclose private information online. This project supported two postdoctoral researchers, one of whom was a co-author.
Dissemination and Exploitation:
The results have been actively disseminated through top-tier journal publications, conference presentations, invited talks, and media coverage. Findings have been shared with both academic and practitioner audiences, ensuring impact beyond academia. Collaborations with industry partners (e.g. mobile app developers) enhance the practical exploitation of results, while international collaborations with leading scholars expand the project’s global reach.
Overall progress:
The project delivered on its proposed aims, significantly advancing knowledge on online engagement, mechanisms of interaction, and trust in digital platforms. Despite challenges posed by COVID-19 and later the war in Israel, the team maintained strong dissemination and collaboration activities. Many of the researchers trained through this project have transitioned into successful academic careers, further demonstrating its long-term impact.
Several outputs of the project go beyond the state of the art. Our MIS Quarterly study on social cues and trust in digital platforms offered new insights into how design features shape private information disclosure, advancing theories of trust and goal systems. More recently, our research on AI-mediated advice seeking and human–ChatGPT interaction is among the earliest systematic examinations of large language models in online engagement. This work advanced academic understanding and drew wide media attention, underlining its societal relevance.
Two outcomes were unexpected. First, the generative AI line of research, prompted by recent technological advances, quickly became one of the project’s most impactful contributions. Second, several postdoctoral researchers supported by the project secured tenure-track positions, exceeding initial expectations and demonstrating the project’s role in capacity building.
Two outcomes were unexpected. First, the generative AI line of research, prompted by recent technological advances, quickly became one of the project’s most impactful contributions. Second, several postdoctoral researchers supported by the project secured tenure-track positions, exceeding initial expectations and demonstrating the project’s role in capacity building.