The critical role of impact sourcing on AI model expansion
Successful machine learning models rely on the quantity, quality and diversity of the data used to train them. It's crucial to feed artificial intelligence (AI) models training data that's inclusive of the collective – rather than one dominant voice – in order to achieve high-quality datasets. This helps to reduce bias and ensure algorithms function as intended for all users, a tenet of responsible AI.
Following a set of responsible AI guidelines helps to deliver better AI outcomes. These principles can outline fair payment for task workers and wellness strategies; prioritize data security, privacy and confidentiality throughout all project stages; and foster a safe workplace free of discrimination, harassment or abuse of any kind.
In addition to creating a safe and productive work environment, following a set of responsible AI principles can combat bias by ensuring your team of data annotators is made up of diverse individuals from a variety of ages, backgrounds, ethnicities, nationalities, abilities, genders and geographies. And when it comes to diversifying your talent pool, impact sourcing can be instrumental in achieving your objectives.
The essential guide to AI training data
Discover best practices for the sourcing, labeling and analyzing of training data from TELUS International, a leading provider of AI data solutions.
What is impact sourcing?
Also known as socially responsible outsourcing, this practice allows companies to access diverse, previously untapped pools of talent from wide-ranging backgrounds. "In contrast to traditional outsourcing providers who often draw from a recycled pool of the same resources, impact sourcing providers bring fresh talent into the workforce and leverage their wide-ranging skills and willingness to learn," according to Indi Village Foundation, a nonprofit organization focused on community development in rural India.
As a workforce model, impact sourcing gives opportunities to high-potential individuals who are marginalized. It provides workers with formal opportunities at a fair wage, skills training and the opportunity for professional development and advancement.
While impact sourcing has a direct impact at the individual level, it also has the potential to significantly improve the lives of families and the greater community. The Rockefeller Foundation estimates that three to four family members benefit from the income brought forth by impact sourcing initiatives. Further, the same philanthropic institute found that impact sourcing has a wider impact on communities, estimating a 3.5 to 4.0 multiplier effect on the local economy.
Enterprises who demonstrate corporate responsibility through impact sourcing have also benefited from improved business outcomes. For example, analyst firm Everest Group has found that attrition rates can be as much as 40% lower for impact workers when compared to regular hires. Their research also indicated that impact workers have high motivation levels that contribute to improved performance over time, as well as lower sourcing and training costs.
Impact sourcing in action
Leveraging impact sourcing can help in the expansion of AI models in a number of ways, including bias prevention, greater language capabilities and longer tenured employees.
Preventing data bias
Bias occurs when the data you have doesn't accurately reflect the conditions your model will operate in. Whether gender, racial, observer, selection or any of the other types of bias, avoiding all forms increases the chances that your machine learning model will work correctly.
Impact sourcing can help reduce the presence of data bias by providing access to a large and diverse talent pool from around the world, and consequently, has the ability to improve overall AI data quality.
For example, TELUS International partnered with YICF, an Indonesian not-for-profit organization, to recruit displaced persons and refugees from Southeast Asia to work as data annotators. A more inclusive AI model was made possible by drawing on the life experiences of this varied workforce that spanned seven nationalities with differing ages, backgrounds and languages. In turn, the project provided these marginalized individuals with the opportunity to gain work experience, improve their language and professional skills through education activities and be a part of a diverse, supportive coworking community.
Multilingual model expansion
Impact sourcing provides access to individuals with divergent vernacular language capabilities, which helps with the expansion of multilingual models. "Developing models that work for more languages is important in order to offset the existing language divide and to ensure that speakers of non-English languages are not left behind," says computer scientist Sebastian Ruder in a blog post on ruder.io.
Machine learning models that support multiple languages help to better serve more people equally, ensuring the AI has a wider reach globally and a greater use case.
Attrition rates tend to be relatively low among impact workers, which ensures reliable and timely service delivery. One study conducted in South Africa by The Rockefeller Foundation showed that the average monthly attrition rate for entry level talent was 5% to 6% for traditional workers compared to 2% to 3% for impact workers.
Results such as these are typically attributed to the fact that for some of these workers, it's their first job that offers them a living wage. As a result, they tend to be highly motivated and committed to their employer.
Impact sourcing: ethical considerations
There are countless ways in which impact sourcing has benefited individuals and communities around the world. Impact sourcing typically leads to improvements in workers' lifestyles through increases in individual income, reductions in stress levels, higher confidence and greater tendencies to remain in their communities rather than migrate, according to The Rockefeller Foundation.
However, impact sourcing is not without its criticisms, with some saying the system leaves opportunity for exploitation. To counter this, corporations using impact sourcing practices must commit to socially responsible principles like paying impact workers at least the minimum total compensation required by local law, providing training and professional development opportunities and practicing inclusive recruiting and hiring processes. These tenets of ethical impact sourcing practices go hand-in-hand with principles of responsible AI.
Expanding AI models through impact sourcing
The task of sourcing a diverse team of individuals can be a drain on both your time and resources. Seeking the help of a service partner – particularly one that leverages impact sourcing – can help.
For example, TELUS International has committed to broaden the focus and participation of our AI impact sourcing programs by increasing participants 15% by the end of 2023.
Through our commitment to impact sourcing programs, our team of experts are well equipped to deliver leading AI Data Solutions that have diversity baked in from the start. To learn more about how TELUS International can help you build inclusive, high-quality datasets for training your AI models, contact one of our experts.