Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Why AI Is WEIRD and Should Not Be This Way: Towards AI For Everyone, With Everyone, By Everyone

Rada Mihalcea, Oana Ignat, Longju Bai, Angana Borah, Luis Chiruzzo, Zhijing Jin, Claude Kwizera, Joan Nwatu, Soujanya Poria, Thamar Solorio

TL;DR

This paper presents a vision for creating AI systems that are inclusive at every stage of development, from data collection to model design and evaluation, and focuses on the need for diverse representation among the developers of these systems.

Abstract

This paper presents a vision for creating AI systems that are inclusive at every stage of development, from data collection to model design and evaluation. We address key limitations in the current AI pipeline and its WEIRD representation, such as lack of data diversity, biases in model performance, and narrow evaluation metrics. We also focus on the need for diverse representation among the developers of these systems, as well as incentives that are not skewed toward certain groups. We highlight opportunities to develop AI systems that are for everyone (with diverse stakeholders in mind), with everyone (inclusive of diverse data and annotators), and by everyone (designed and developed by a globally diverse workforce).

Why AI Is WEIRD and Should Not Be This Way: Towards AI For Everyone, With Everyone, By Everyone

TL;DR

This paper presents a vision for creating AI systems that are inclusive at every stage of development, from data collection to model design and evaluation, and focuses on the need for diverse representation among the developers of these systems.

Abstract

This paper presents a vision for creating AI systems that are inclusive at every stage of development, from data collection to model design and evaluation. We address key limitations in the current AI pipeline and its WEIRD representation, such as lack of data diversity, biases in model performance, and narrow evaluation metrics. We also focus on the need for diverse representation among the developers of these systems, as well as incentives that are not skewed toward certain groups. We highlight opportunities to develop AI systems that are for everyone (with diverse stakeholders in mind), with everyone (inclusive of diverse data and annotators), and by everyone (designed and developed by a globally diverse workforce).

Paper Structure

This paper contains 28 sections, 7 figures.

Table of Contents

  1. Representation in AI
  2. Lack of Cultural Knowledge.
  3. Data
  4. Data Sources
  5. Lack of Data Transparency.
  6. Lack of Data Diversity.
  7. Data Annotations
  8. Lack of Inclusive Annotation Standards.
  9. Lack of Diverse Annotators and Curators.
  10. Models
  11. Model Design
  12. Models Exhibit Biases in Pre-training and Alignment.
  13. Model Performance
  14. Poor Model Generalization.
  15. Security Vulnerabilities and Propagation of Harmful Stereotypes.
  16. ...and 13 more sections

Figures (7)

  • Figure 1: Desiderata and areas of research to expand the reach and impact of AI to everyone.
  • Figure 2: Outsider perspectives on the history and culture of groups not represented in AI models often conflict with the insider perspectives and can be misleading.
  • Figure 3: Models trained on non-inclusive datasets hinder the representation of stakeholders in mainstream media.
  • Figure 4: Uneven model performance for different languages leads to incorrect output and can favor those speaking the languages for which the model performs better.
  • Figure 5: A misalignment between evaluation metrics and cultural values can lead to misleading estimation of a tool's effectiveness.
  • ...and 2 more figures