Left side: Example of a message exchange with AI support (eg, Smart Reply enabled Messenger). Typical examples of Smart Replies and how to present them to the user are shown in orange at the bottom. Right side: Abstract representation of the impact of AI on interpersonal communication. Either or both participants can access AI support (eg in the form of smart responses). When granting access to an AI support, participants may choose to use it or not (actual use). However, apart from actual use, participants make assumptions about AI support (perceived use). Both actual use and perceived use influence the general exchange of messages and the perceptions that people form of one another. credit: Scientific reports (2023). DOI: 10.1038/s41598-023-30938-9
A group of researchers at Cornell University has found that people have more effective conversations, use more positive language, and perceive each other more positively when using an AI-powered chat tool.
Postdoctoral researcher Jess Hohenstein is the lead author of “Artificial Intelligence in Communication Affects Language and Social Relations” published in Scientific reports.
Co-authors include Malte Jung, Associate Professor of Information Science in the Cornell Ann S. Bowers College of Computing and Information Sciences (Cornell Bowers CIS), and Renee Kizelcek, Assistant Professor of Information Sciences (Cornell Bowers CIS).
Generative AI is poised to influence all aspects of society, communication, and work. Every day brings new evidence of the technical capabilities of Language Large Models (LLMs) such as ChatGPT and GPT-4, but the social consequences of integrating these technologies into our daily lives are still not well understood.
AI tools have the potential to improve efficiency, but may have negative social side effects. Hohenstein and his colleagues studied how the use of AI in conversations affects the way people express themselves and view each other.
“Technology companies tend to emphasize the usefulness of AI tools to get things done faster and better, but they ignore social distancing,” Young said. “We do not live and work in isolation, and the systems we use influence our interactions with others.”
In addition to increased competence and positivity, the group found that when participants believed their partner was using more AI-suggested responses, they perceived that partner as less cooperative, and felt less affiliation with them.
“I was surprised to find that people tend to rate you more negatively simply because they suspect you are using AI to help you text, regardless of whether you actually are,” Hohenstein said. “This illustrates the persistent public skepticism that people seem to have about artificial intelligence.”
In their first experiment, co-author Dominic DeFranzo, a former postdoctoral researcher in the Cornell Robotics and Clusters Laboratory and assistant professor at Lehigh University, developed a smart reply platform called the group “moshi” (Japanese for “hello”), similar to Google’s “Allo”. (French for “hello”), the first Smart Reply platform, was unveiled in 2016. Smart Replies are generated from LLMs to predict the next reasonable responses in chat-based interactions.
A total of 219 pairs of participants were asked to talk about an issue related to politics and assigned one of three conditions: both participants could use Smart Responses; Only one participant can use Smart Replies; Or no participant can use Smart Replies.
The researchers found that the use of smart responses increased communication efficiency, positive emotional language, and positive evaluations by communication partners. On average, Smart Replies made up 14.3% of messages sent (1 in 7).
But participants whose response partners suspected of having intelligent responses were rated more negatively than those who were thought to have written their own responses, consistent with common assumptions about the negative effects of AI.
In a second experiment, 299 pairs of randomly assigned participants were asked to discuss a policy issue under one of four conditions: No intelligent responses. the default Google Smart Replies; intelligent responses with a positive emotional tone; and the other with a negative emotional tone. Having positive, intelligent responses from Google made conversations have a more positive emotional tone than conversations with negative or no smart responses, highlighting the impact AI can have on language production in everyday conversations.
Hohenstein said, “While AI may be able to help you write, it changes your language in ways you might not expect, especially by making you sound more positive. This suggests that by using AI to generate text, you are sacrificing some of your personal voice.” .
“What we observe in this study is the impact of AI on social dynamics and some of the unintended consequences that may result from integrating AI into social contexts,” Jung said. This suggests that who controls the algorithm may have an impact on people’s interactions, language, and perceptions of each other. “.
more information:
Jess Hohenstein et al., Artificial Intelligence in Communication Affects Language and Social Relations, Scientific reports (2023). DOI: 10.1038/s41598-023-30938-9
the quote: Study Uncovers Social Cost of Using AI in Conversations (2023, April 4) Retrieved April 4, 2023 from https://phys.org/news/2023-04-uncovers-social-ai-conversations.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.
