CoChat

Exploring the Impact of AI Value Alignment in Collaborative Ideation: Effects on Perception, Ownership, and Output

Open access
Guo, Alicia, Pataranutaporn, Pat, Maes, Pattie · 2024 · Extended Abstracts of the CHI Conference on Human Factors in Computing Systems

The DOI matches its Crossref record; 24 of the 24 references it cites were found in a registry or an index; none of the 4 sources its claims cite is checked for support yet.

Abstract

AI-based virtual assistants are increasingly used to support daily ideation tasks. The values or bias present in these assistants can influence output in hidden ways and affect how people perceive the ideas produced with these AI assistants, leading to implications for the design of AI-based tools. We explored the effects of AI assistants with different values on the ideation process and user perception of idea quality, ownership, assistant competence, and values present in the output. Our study tasked 180 participants with brainstorming practical solutions to a set of problems with AI assistants of different values. Results show no significant difference in self-evaluation of idea quality and perception of the assistant based on value alignment; however, ideas generated reflected the AI’s values and feeling of ownership is affected. This highlights an intricate interplay between AI values and human ideation, suggesting careful design considerations for future AI-supported brainstorming tools.

Overview

CoChat Summary

TL;DR

This study examined how brainstorming with GPT-4 agents prompted to hold Pro, Neutral, or Con values on Economic, Automation, and Hierarchy domains affected users’ perceptions and the values in their final ideas. It found AI values appeared in outputs and AI collaboration reduced perceived ownership, while value alignment did not significantly affect self-rated idea quality or agent perception.

Key findings

  • Final ideas reflected the AI agent’s values: a one-way ANOVA found a significant difference between conditions (F(3, 219)=14.03, p=2.2E-8). Pro AI responses scored higher on the value dimension (M=7.49, SD=2.76) than Neutral AI (M=5.51, SD=2.97; p=.005), Con AI (M=3.65, SD=3.45; p<.001), and No AI (M=5.91, SD=3.16; p=.039); Con AI was also lower than Neutral AI (p=.01) and No AI (p=.001), while Neutral AI and No AI did not differ significantly (p=.907).
  • Human-AI value alignment showed no statistically significant differences for self-reported idea quality or for any perception-of-AI-agent questions.
  • AI collaboration significantly reduced self-reported idea ownership compared with No AI: one-way ANOVA F(3, 60)=22.53, p=7.3E-13; No AI M=3.87, SD=0.69 vs Same M=2.59, SD=0.99; Different M=2.98, SD=1.10; Neutral M=2.68, SD=0.99, all p<.0001. Different AI values did not significantly affect ownership ratings.
  • Process ratings differed: Neutral AI made it easier to answer the prompt than Different AI and No AI (F(3, 60)=5.62, p=0.001; Neutral M=4.31, SD=0.85 vs Different M=3.73, SD=1.17, p=0.037; No AI M=3.47, SD=1.31, p<0.0001), and Neutral AI and Same AI were rated as more useful brainstorming time than No AI (F(3, 60)=4.37, p=0.005; Neutral M=4.33, SD=0.89, p=.043; Same M=4.41, SD=0.75, p=0.010; No AI M=3.87, SD=1.13).

Methods

  • Online experiment with 180 participants (60 per domain), US residents aged 18–60 from Prolific, run on Qualtrics. Participants self-rated their values on a 0–10 scale, then completed two timed brainstorming sessions—one independently and one with an AI agent—in random order, followed by post-surveys and demographics.
  • AI agents were built with GPT-4 prompting for Pro, Neutral, or Con values across Economic, Automation, and Hierarchy domains; temperature was 0.7, and prompts included “Reply in 30 words or less.” Measures included self-evaluation of idea quality, ownership, perception of the AI agent, and process/empowerment on 5-point Likert scales; final ideas were also scored 0–10 for value presence by a GPT-4 prompt.
  • Analysis used descriptive statistics and ANOVA models comparing experimental and control conditions. The No AI responses were sampled down to 20 to match Pro, Neutral, and Con condition sizes. Depending on sample size and Shapiro-Wilk normality results, the authors used Kruskal-Wallis with post-hoc Dunn and Bonferroni correction, or Levene tests followed by Welch ANOVA with Tukey post-hoc or basic ANOVA.

Limitations

  • The paper states the experiment was a Qualtrics survey with crafted brainstorming prompts users had no personal stake in, not a real-world setting; five minutes was felt to be too short and may have led participants to follow AI suggestions. It also states participants completed only a single AI brainstorm session, so behavior across all AI conditions could not be assessed, and the original all-conditions design was too long and tiring, resulting in 20 participants per AI condition per domain value.
  • The paper states user values were determined by one self-scored question, chosen as the simplest method after testing multiple measures, and that real-world values are more nuanced and do not exist on a single dimension. It also notes the method of evaluating values in brainstorming output is subject to limitations, and the text describes the GPT-4 evaluation as “not at the level of human rater sophistication.”
  • Evident from the text: the No AI condition was downsampled to 20 to match AI conditions, which may limit comparisons, and the value domains were deliberately simplified into clear dichotomies that “does not take into account all of the nuances of the values.”

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3613905.3650892

Cites

24 of 24 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Verification

Claims the paper makes

0 Valid · 0 Not valid · 3 Not checked · 44 Own results and interpretations

17 more claims

Create a free account to see all 47 claims and to verify the 4 references still open.

See all claims — free account
  • Probably not a substantial claim. 20
  • It may state two separate claims as one. 10

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 24 references

Research works printed but not linked: 9

  1. Mostafa M. Amin Erik Cambria and Björn W. Schuller. 2023. Will Affective Computing Emerge from Foundation Models and General AI? A First Evaluation on ChatGPT. arxiv:2303.03186 [cs.CL] · arXiv
  2. Emilio Ferrara. 2023. Should ChatGPT be Biased? Challenges and Risks of Bias in Large Language Models. ArXiv abs/2304.03738 (2023). https://api.semanticscholar.org/CorpusID:258041203 · arXiv
  3. Yuen han. Chiu and Chen Chun-Ching. 2023. Investigating the Impact of Generative Artificial Intelligence on Brainstorming: A Preliminary Study. 2023 International Conference on Consumer Electronics - Taiwan (ICCE-Taiwan) (2023), 193–194. https://api.semanticscholar.org/CorpusID:261433351
  4. Daphne Ippolito, Ann Yuan, Andy Coenen, and Sehmon Burnam. 2022. Creative Writing with an AI-Powered Writing Assistant: Perspectives from Professional Writers. ArXiv abs/2211.05030 (2022). https://api.semanticscholar.org/CorpusID:253420678 · arXiv
  5. Janin Koch Prashanth Thattai Ravikumar and Filipe Calegario. 2021. Agency in Co-Creativity: Towards a Structured Analysis of a Concept. https://www.semanticscholar.org/paper/Agency-in-Co-Creativity%3A-Towards-a-Structured-of-a-Koch-Ravikumar/46bd05136805aae23bd2cd0015bb1f6398e9cbe8
  6. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In Advances in Neural Information Processing Systems, I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (Eds.). Vol. 30. Curran Associates, Inc.https://proceedings.neurips.cc/paper_files/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf
  7. Fernanda Viégas and Martin Wattenberg. 2023. The System Model and the User Model: Exploring AI Dashboard Design. arXiv preprint arXiv:2305.02469 (2023). · arXiv
  8. Qian Wan Siying Hu Yu Zhang Piaohong Wang Bo Wen and Zhicong Lu. 2023. "It Felt Like Having a Second Mind": Investigating Human-AI Co-creativity in Prewriting with Large Language Models. arxiv:2307.10811 [cs.HC] · arXiv
  9. Justin D. Weisz, Michael J. Muller, Stephanie Houde, John T. Richards, Steven I. Ross, Fernando Martinez, Mayank Agarwal, and Kartik Talamadupula. 2021. Perfection Not Required? Human-AI Partnerships in Code Translation. CoRR abs/2104.03820 (2021). arXiv:2104.03820https://arxiv.org/abs/2104.03820 · arXiv

Cited by

See all 34

Metadata

Authors
Guo, A., Pataranutaporn, P., & Maes, P.
Affiliations
  • Massachusetts Institute of Technology (US)
  • MIT Media Lab (US)
Venue
Extended Abstracts of the CHI Conference on Human Factors in Computing Systems
Publisher
ACM
Published
2024-05-11
Pages
1-11
OpenAlex
W4396832253
Cited by
32
References
22
Citations by year
2026: 14 · 2025: 17 · 2024: 1
FWCI
10.18
Type
conference paper
Language
en
Open access
Locations
  • CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) (publishedVersion)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc)
  • arXiv (Cornell University) (cc-by)
Topics
  • Physical Sciences › Computer Science › Human-Computer Interaction › Innovative Human-Technology Interaction
  • Physical Sciences › Engineering › Mechanical Engineering › Design Education and Practice
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
Keywords
AI alignment, algorithmic bias, collaborative ideation, brainstorming, AI virtual assistants, user perceptions of AI

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-04, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-04 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Exploring the Impact of AI Value Alignment in Collaborative Ideation: Effects on Perception, Ownership, and Output. https://app.cochat.ai/papers/10.1145/3613905.3650892
The paper
Guo, A., Pataranutaporn, P., & Maes, P. (2024). Exploring the Impact of AI Value Alignment in Collaborative Ideation: Effects on Perception, Ownership, and Output. Extended Abstracts of the CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3613905.3650892

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.