Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Amplify Initiative: Building A Localized Data Platform for Globalized AI

Qazi Mamunur Rashid, Erin van Liemt, Tiffany Shih, Amber Ebinama, Karla Barrios Ramos, Madhurima Maji, Aishwarya Verma, Charu Kalia, Jamila Smith-Loud, Joyce Nakatumba-Nabende, Rehema Baguma, Andrew Katumba, Chodrine Mutebi, Jagen Marvin, Eric Peter Wairagala, Mugizi Bruce, Peter Oketta, Lawrence Nderu, Obichi Obiajunwa, Abigail Oppong, Michael Zimba, Data Authors

TL;DR

Amplify Initiative addresses the global AI data gap by building a participatory, locally anchored data platform that engages domain experts to generate multilingual, culturally nuanced adversarial queries. The Sub-Saharan Africa pilot co-created 8,091 queries across seven languages with 155 experts, using an Android app and a seven-step methodology that emphasizes trust, training, validation, and fair recognition. Key findings reveal strong health-focused misinformation patterns, gendered mental health expressions, disability concerns in education, and rich ethnolinguistic diversity, all informing safer and more contextually relevant AI evaluation. The work demonstrates a scalable path to broaden AI safety and relevance for the Global South, while highlighting challenges in recruitment, data quality, cultural nuance, and ethical data governance.

Abstract

Current AI models often fail to account for local context and language, given the predominance of English and Western internet content in their training data. This hinders the global relevance, usefulness, and safety of these models as they gain more users around the globe. Amplify Initiative, a data platform and methodology, leverages expert communities to collect diverse, high-quality data to address the limitations of these models. The platform is designed to enable co-creation of datasets, provide access to high-quality multilingual datasets, and offer recognition to data authors. This paper presents the approach to co-creating datasets with domain experts (e.g., health workers, teachers) through a pilot conducted in Sub-Saharan Africa (Ghana, Kenya, Malawi, Nigeria, and Uganda). In partnership with local researchers situated in these countries, the pilot demonstrated an end-to-end approach to co-creating data with 155 experts in sensitive domains (e.g., physicians, bankers, anthropologists, human and civil rights advocates). This approach, implemented with an Android app, resulted in an annotated dataset of 8,091 adversarial queries in seven languages (e.g., Luganda, Swahili, Chichewa), capturing nuanced and contextual information related to key themes such as misinformation and public interest topics. This dataset in turn can be used to evaluate models for their safety and cultural relevance within the context of these languages.

Amplify Initiative: Building A Localized Data Platform for Globalized AI

TL;DR

Amplify Initiative addresses the global AI data gap by building a participatory, locally anchored data platform that engages domain experts to generate multilingual, culturally nuanced adversarial queries. The Sub-Saharan Africa pilot co-created 8,091 queries across seven languages with 155 experts, using an Android app and a seven-step methodology that emphasizes trust, training, validation, and fair recognition. Key findings reveal strong health-focused misinformation patterns, gendered mental health expressions, disability concerns in education, and rich ethnolinguistic diversity, all informing safer and more contextually relevant AI evaluation. The work demonstrates a scalable path to broaden AI safety and relevance for the Global South, while highlighting challenges in recruitment, data quality, cultural nuance, and ethical data governance.

Abstract

Current AI models often fail to account for local context and language, given the predominance of English and Western internet content in their training data. This hinders the global relevance, usefulness, and safety of these models as they gain more users around the globe. Amplify Initiative, a data platform and methodology, leverages expert communities to collect diverse, high-quality data to address the limitations of these models. The platform is designed to enable co-creation of datasets, provide access to high-quality multilingual datasets, and offer recognition to data authors. This paper presents the approach to co-creating datasets with domain experts (e.g., health workers, teachers) through a pilot conducted in Sub-Saharan Africa (Ghana, Kenya, Malawi, Nigeria, and Uganda). In partnership with local researchers situated in these countries, the pilot demonstrated an end-to-end approach to co-creating data with 155 experts in sensitive domains (e.g., physicians, bankers, anthropologists, human and civil rights advocates). This approach, implemented with an Android app, resulted in an annotated dataset of 8,091 adversarial queries in seven languages (e.g., Luganda, Swahili, Chichewa), capturing nuanced and contextual information related to key themes such as misinformation and public interest topics. This dataset in turn can be used to evaluate models for their safety and cultural relevance within the context of these languages.

Paper Structure

This paper contains 52 sections, 5 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Availability of Data Resources
  4. Impact of Limited Data
  5. When Languages Are Not Represented
  6. When Lived Experiences Are Not Represented
  7. When Critical Needs Are Not Met
  8. Community-focused Approaches
  9. Amplify Initiative's Approach
  10. Introduction to Amplify Initiative
  11. Pilot Across Sub-Saharan Africa
  12. Methodology For Creating Localized Data
  13. Data Analysis
  14. Data Author Statistics
  15. Linguistic Diversity
  16. ...and 37 more sections

Figures (5)

  • Figure 3: Map of African continent showing the countries where data was collected and number of queries per language.
  • Figure 4: Distribution of number of queries per language and domain across all 5 countries.
  • Figure 5: Distribution of number of queries per theme and domain across all 5 countries.
  • Figure 6: Manually constructed ontological representation of entities in health and misinformation queries from Malawi, Uganda, and Ghana.
  • Figure 7: Distribution of Number of Queries per Language and Domain across all 5 Countries for non-English Languages