Research map / Large language models / AI and society
Position: Towards Bidirectional Human-AI Alignment
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Position paper arguing alignment should be defined bidirectionally, reviewing over 400 HCI/NLP/ML papers to build a framework covering aligning AI to humans and aligning humans to AI.
Problem
Alignment is used without a shared definition, and prior work treats it as a one-way, static process that ignores how humans and AI adapt to each other.
Approach
Proposes Bidirectional Human-AI Alignment, which adds the underexplored direction of aligning humans to AI (cognitive, behavioral, societal adaptation) to the traditional aligning-AI-to-humans direction, grounded in a systematic review of 411 papers and Schwartz's basic values.
Key results
- Systematic review of 34,213 retrieved records narrowed to 2,136 screened and 411 eligible papers.
- Inter-annotator joint agreement of 0.78 on whether papers investigate framework dimensions.
- Analysis finds gaps in implicit human feedback, inference-stage customization, human-in-the-loop evaluation, AI literacy, and ethical auditing.
Limitations
- Corpus is a non-exhaustive sample focused on computing fields (ML, NLP, HCI), excluding cognitive science, psychology, and STS.
- Keyword and criteria-based filtering may miss relevant work; the paper acknowledges this.
- Inter-annotator agreement of 0.78 is moderate, and coding involved many annotators with divided teams.
Experimental setup
Datasets: 411 papers from ACL Anthology, OpenReview, and ACM Digital Library (HCI, NLP, ML venues, 2019 to Jan 2024)
Metrics: paper counts per framework dimension inter-annotator agreement (joint probability 0.78)
Builds on: Schwartz Theory of Basic Values PRISMA
Closest papers by meaning
- Quantifying Misalignment Between Agents: Towards a Sociotechnical Understanding of Alignment (AAAI 2025 · similarity 0.92)
- An Evolutionary Perspective on AI Alignment (Student Abstract) (AAAI 2025 · similarity 0.91)
- Position: Challenges and Future Directions of Data-Centric AI Alignment (ICML 2025 · similarity 0.90)
- Position: A Roadmap to Pluralistic Alignment (ICML 2024 · similarity 0.90)
- MAP: Multi-Human-Value Alignment Palette (ICLR 2025 · similarity 0.90)
- Learning Human-like Representations to Enable Learning Human Values (NeurIPS 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.