PapersWithELO
← ICLR 2024 leaderboard

Culture in Artificial Intelligence: A Literature Review & Proposal

Cesa Salaam, Danda Rawat

fairness, safety & privacyFATEFairnessBias MitigationCulture in AIHuman-Value Alignment
0.80100
Fused
band ≈ ±15 pct pts (from σ = 0.30)
1.00100
Mimo
band ≈ ±19 pct pts (from σ = 0.38)
0.00100
DeepSeek
band ≈ ±23 pct pts (from σ = 0.47)

OpenReview ground truth

Rejected

TL;DR — This paper presents a comprehensive examination of the intersection of AI and culture, highlighting the importance of considering cultural factors in AI development.

Abstract

Within the last few years, there has been an explosion of various Artificial Intelligence technologies poised to change the world. However, a lot of these technologies are made by the few to represent the many. This absence of diversity amongst researchers and overall innovators creates technology that is microscopic in worldview. It brings questions of Fairness, Accountability, Transparency, and Ethics (FATE) to the forefront. Most research undertaken in the context of FATE is done within a Western cultural context, which, in turn, imparts Western values. However, most research does not holistically address the question of the relationship between culture and AI. In this paper, we conduct a literature review of relevant research on Artificial Intelligence and culture and its importance in analyzing concepts of FATE. Additionally, we argue for and propose a definition of Culture in AI. We assume that through a combination of activation points (data collection and annotation, algorithm choice/development, problem framing, etc.), AI systems/agents perpetuate and produce culture in their dealings. This paper posits the need to situate Artificial intelligence systems within specific cultural paradigms consistent with their operative environments. We end by discussing future areas of study to be considered.

Author context

Most prolific author: 1 submissions (credibility 1.00).

No mass-submission penalty for this paper (authors within normal submission volume).

Aggregate statistics only — no individual author rankings.

Ranking trajectory

Percentile by tournament round — convergence indicates rating stability.

Battle history — 38 comparisons

Ranked above opponent in 20% of matchups.

Judge assessments

Mean overall score 0.0 ± 0.0 (n = 38)