Ironies of Generative AI: Understanding and mitigating productivity loss in human-AI interactions
Project Overview
The document explores the integration of generative AI (GenAI) in education, highlighting its potential to enhance productivity and learning experiences in knowledge-intensive tasks such as programming and writing. It discusses various applications of GenAI tools that utilize natural language processing to provide personalized learning support and improve student engagement. However, it also addresses significant challenges, including cognitive overload, ineffective workflows, and the transition of user roles from production to evaluation, which can lead to frustration and confusion with GenAI outputs. The findings suggest that while GenAI can facilitate enhanced educational outcomes, it may also result in decreased situational awareness and increased reliance on technology. By drawing on insights from Human Factors research, the document proposes design solutions aimed at improving user experience and effectiveness when interacting with GenAI systems, ultimately suggesting that with careful implementation and design considerations, the benefits of generative AI in education can be maximized while mitigating its challenges.
Key Applications
AI-Assisted Content Generation Tools
Context: Students and professionals in computer science using coding tools, and students engaging in creative writing exercises. This includes software engineering environments for programmers and creative workflows for writers.
Implementation: Integration of AI tools that assist in generating code completions, translating natural language into code or spreadsheet functions, as well as providing suggestions for creative writing, such as story ideas and narratives.
Outcomes: ['Improved coding efficiency and code quality.', 'Enhanced accessibility and usability of spreadsheets for non-experts.', "Increased creativity and reduced writer's block."]
Challenges: ['Users may struggle with effectively prompting the AI and debugging generated outputs.', 'Dependence on AI could stifle individual creativity.', 'The complexity of natural language processing can hinder usability.']
Implementation Barriers
Usability Barrier
High cognitive load and frustration experienced by users when interacting with GenAI systems, along with difficulty in effectively interacting with AI tools.
Proposed Solutions: Incorporate continuous feedback mechanisms, ecological interface design, user personalization, user training, and develop more intuitive interfaces to mitigate cognitive demands.
Workflow Disruption
GenAI tools can create unhelpful restructuring of workflows, causing interruptions and loss of task sequence.
Proposed Solutions: Implement task stabilization techniques and clear task allocation to enhance user control over their workflow.
Situational Awareness
Users experience reduced situational awareness due to the passive role of evaluating AI outputs.
Proposed Solutions: Enhance explainability and feedback to improve users' understanding of AI-generated outputs.
Technical Barrier
Challenges in the accuracy and reliability of AI-generated content.
Proposed Solutions: Continuous improvement through user feedback and iterative development.
Project Team
Auste Simkute
Researcher
Lev Tankelevitch
Researcher
Viktor Kewenig
Researcher
Ava Elizabeth Scott
Researcher
Abigail Sellen
Researcher
Sean Rintel
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Auste Simkute, Lev Tankelevitch, Viktor Kewenig, Ava Elizabeth Scott, Abigail Sellen, Sean Rintel
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai