[Paper Review] Cultural Incongruencies in Artificial Intelligence

Vinodkumar Prabhakaran, Rida Qadri+1 more|arXiv (Cornell University)|Nov 19, 2022

Language and cultural evolution37 citations

TL;DR

The paper argues that AI systems reflect culturally shaped human behavior and highlights five harms from cultural misalignments, proposing a research agenda for culturally cognizant AI.

ABSTRACT

Artificial intelligence (AI) systems attempt to imitate human behavior. How well they do this imitation is often used to assess their utility and to attribute human-like (or artificial) intelligence to them. However, most work on AI refers to and relies on human intelligence without accounting for the fact that human behavior is inherently shaped by the cultural contexts they are embedded in, the values and beliefs they hold, and the social practices they follow. Additionally, since AI technologies are mostly conceived and developed in just a handful of countries, they embed the cultural values and practices of these countries. Similarly, the data that is used to train the models also fails to equitably represent global cultural diversity. Problems therefore arise when these technologies interact with globally diverse societies and cultures, with different values and interpretive practices. In this position paper, we describe a set of cultural dependencies and incongruencies in the context of AI-based language and vision technologies, and reflect on the possibilities of and potential strategies towards addressing these incongruencies.

Motivation & Objective

• Identify how culture influences AI development and use.
• Characterize harms arising from cultural incongruencies in AI systems.
• Propose a research agenda to advance culturally cognizant AI.
• Encourage interdisciplinary approaches and diverse perspectives in AI research and deployment.

Proposed method

• Synthesize existing literature on culture, communication, and AI to identify cultural dependencies in development and use of AI systems.
• Describe five concrete harms arising from cultural incongruencies: cultural barriers, hegemonic classifications, safety gaps, violation of cultural values, and erasure.
• Propose high-level research directions and questions to mitigate cultural incongruencies in AI.
• Advocate for interdisciplinary collaboration and participatory processes in AI design and evaluation.

Experimental results

Research questions

• RQ1 Which aspects of AI systems are dependent on culture and how does this affect suitability in different ecosystems?
• RQ2 What incongruencies and harms emerge when AI systems’ implicit cultural predispositions clash with target cultural ecosystems?
• RQ3 How can we identify, measure, and mitigate cultural harms in AI technologies?
• RQ4 How can culturally situated evaluations and diverse perspectives be integrated into AI development and deployment?
• RQ5 What are practical pathways to incorporate broader cultural viewpoints and co-creation in AI pipelines?

Key findings

• Cultural dependencies in AI arise both during development and use, through data, resources, and the cultural norms of developers and researchers.
• Five harms from cultural incongruencies are identified: cultural barriers, imposing hegemonic classifications, safety gaps, violations of cultural values, and cultural erasure.
• Efforts to ensure safety, fairness, and performance may fail if they do not account for target cultural contexts and local norms.
• Cultural erasure can propagate through pre-trained models, leading to homogenization and misrepresentation of diverse cultures.
• A research agenda is proposed focusing on measuring cultural harms, culturally situated evaluations, and integrating diverse perspectives into AI pipelines.