The challenge
As the world begins to rely more on AI to deliver robust technological solutions, data bias has become a real concern.
Ensuring that the training data that is fed into AI models is diverse and inclusive of the collective, rather than one dominant group’s opinions and judgments, is of utmost importance.
According to the United Nations, in 2021 there were more than 84 million refugees worldwide. This figure has nearly doubled in the past decade.
These populations have typically been marginalized when it comes to recruitment opportunities. For this reason, TELUS Digital sought out a partner who could help us add this integral population to our AI Community workforce.
Recruiting data annotators from across Southeast Asia allows us to draw on the vast life experiences of a wide range of individuals for more inclusive AI models that effectively combat bias.
The TELUS Digital approach
To ensure accurate data representation, TELUS Digital selected YICF as a partner because of their commitment to creating opportunities for Indonesians and refugees in transit to the country. YICF is a locally rooted and globally connected organization focused on transforming lives in greater Jakarta through education, vocation and community.
Our team worked with YICF to expand our global workforce by engaging with refugees from throughout Southeast Asia. As part of the partnership, we participated in the organization’s recently launched Bersama program, with its mission to provide a life-changing, vocational learning experience for unemployed Indonesian youth and refugees.
Our collaboration focused on three main objectives for the refugee population in Indonesia, who are without legal rights to employment while they await resettlement to a third country:
- Working - Gaining work experience through data annotation
- Learning - Improving their language and professional skills through continuous education activities
- Belonging - Enjoying membership within a diverse, supportive coworking community
The results
Our participation in YICF’s Bersama program has led to the development of a roster of 60-90 workers, delivering more than 2,700 hours of work per week, across more than 50 projects. Bersama provided a diverse workforce that spanned across seven unique nationalities with a variety of ages, backgrounds and languages represented.
The team has also expanded their support of simple collection and categorization tasks to include more complex image annotation, audio and image transcription, categorization and quality assurance capabilities.
Project highlights include:
- High retention rate of 80%
- 75-80% of workforce tackling multiple tasks concurrently
- Greater than 80% pass rate across all optical character recognition projects
- 351,949 images labeled
- 5,200+ image and video files collected
- Speech-to-text audio transcription in English, Persian, Indonesian and Arabic
- 459,968 files reviewed (quality assurance)