Periodic Reporting for period 1 - AI4Dignity (Collaborative AI Counters Hate)
Periodo di rendicontazione: 2021-01-01 al 2022-06-30
Online extreme speech has emerged as a significant challenge for democratic societies worldwide. Governments, companies, and academic researchers have responded to this phenomenon by increasingly turning to Artificial Intelligence (AI) as a potential tool that can detect, decelerate, and remove online extreme speech.
AI deployment is expected to bring scalability, reduce costs, and decrease human discretion and emotional labor in the removal of objectionable content. However, even though digitalization is now a global phenomenon, AI tools for extreme speech detection that are globally applicable, inclusive, and yet resource-efficient and feasible are lacking.
The key objective of AI4Dignity has been to operationalize the “human-in-the-loop” principle by developing a community-based human–machine process model with curated space of coding to detect and categorize extreme speech.
Towards this goal, the project has successfully implemented the project goals by
• organizing collaborative coding with factcheckers, ethnographers and NLP researchers to gather and label extreme speech in five languages—Brazilian Portuguese, German, English, Hindi and Swahili—covering India, Brazil, Kenya and Germany.
• creating a frontend for the classification tool as a pilot
• releasing a AI4Dignity toolkit with step-by-step instructions about organizing collaborative coding events (Counterathon); technical details about gathering and annotating extreme speech; and building the classification tool frontend
• creating extreme speech database for the selected country case studies for more research
• launching policy reports on community participation in AI-assisted content moderation
• publishing research articles for the academic audience
• actively engaging with media (print news and podcasts) and the corporate sector for dissemination of results
• consulting for the United Nations Peace Operations for developing holistic approaches to counter online extreme speech, including replication of AI4Dignity’s process model for content detection in under-served languages.
Across these activities and outputs, the project has stressed the need for involving communities in identifying and labeling online extreme speech and incorporating their contextual knowledge in building responsible training datasets in different languages. The project has also proposed the framework of “ethical scaling” to highlight moderation process as political praxis. As a normative framework for platform governance, ethical scaling calls for a transparent, reflexive and replicable process of iteration for content moderation with community participation and global parity, which should evolve in conjunction with addressing algorithmic amplification of divisive content and resource allocation for content moderation.
The various outcomes of the project are curated on a dedicated website: https://www.ai4dignity.gwi.uni-muenchen.de(si apre in una nuova finestra)
Researchers interested in accessing the curated extreme speech datasets can approach the team. Civil society organizations and companies interested in implementing AI4Dignity can download the policy handout and the toolkit, and get in touch with the team. As a social intervention project, AI4Dignity has stressed the need for investing resources in developing iterative and reflexive processes for content moderation with community involvement.
AI deployment is expected to bring scalability, reduce costs, and decrease human discretion and emotional labor in the removal of objectionable content. However, even though digitalization is now a global phenomenon, AI tools for extreme speech detection that are globally applicable, inclusive, and yet resource-efficient and feasible are lacking.
The key objective of AI4Dignity has been to operationalize the “human-in-the-loop” principle by developing a community-based human–machine process model with curated space of coding to detect and categorize extreme speech.
Towards this goal, the project has successfully implemented the project goals by
• organizing collaborative coding with factcheckers, ethnographers and NLP researchers to gather and label extreme speech in five languages—Brazilian Portuguese, German, English, Hindi and Swahili—covering India, Brazil, Kenya and Germany.
• creating a frontend for the classification tool as a pilot
• releasing a AI4Dignity toolkit with step-by-step instructions about organizing collaborative coding events (Counterathon); technical details about gathering and annotating extreme speech; and building the classification tool frontend
• creating extreme speech database for the selected country case studies for more research
• launching policy reports on community participation in AI-assisted content moderation
• publishing research articles for the academic audience
• actively engaging with media (print news and podcasts) and the corporate sector for dissemination of results
• consulting for the United Nations Peace Operations for developing holistic approaches to counter online extreme speech, including replication of AI4Dignity’s process model for content detection in under-served languages.
Across these activities and outputs, the project has stressed the need for involving communities in identifying and labeling online extreme speech and incorporating their contextual knowledge in building responsible training datasets in different languages. The project has also proposed the framework of “ethical scaling” to highlight moderation process as political praxis. As a normative framework for platform governance, ethical scaling calls for a transparent, reflexive and replicable process of iteration for content moderation with community participation and global parity, which should evolve in conjunction with addressing algorithmic amplification of divisive content and resource allocation for content moderation.
The various outcomes of the project are curated on a dedicated website: https://www.ai4dignity.gwi.uni-muenchen.de(si apre in una nuova finestra)
Researchers interested in accessing the curated extreme speech datasets can approach the team. Civil society organizations and companies interested in implementing AI4Dignity can download the policy handout and the toolkit, and get in touch with the team. As a social intervention project, AI4Dignity has stressed the need for investing resources in developing iterative and reflexive processes for content moderation with community involvement.