Technology giants like Google and Microsoft are betting on a 27-year-old Stanford-educated computer engineer to make artificial intelligence (AI) work for over a billion Indians.
Read More: FPIs Withdraw Rs 3,400 Crore During First Three Trading Sessions In November
Why tech giants want AI to do well in India
Manu Chopra, the founder of a startup called Karya, is attempting a mammoth task – to collect text, voice, and image data in India’s local languages. Speakers of local languages collect and label the data that AI is used by AI chatbots and virtual assistants to generate relevant results in any language.
In conversation with Officenewz, Manu Chopra, explained how poor pay for such work is the failure of the tech industry, even as AI tools are normalised across the board. AI companies around the world benefit from the work of underpaid workers who tag and label the data in local languages.
What do Microsoft and Google want from Indians?
Microsoft has used Karya’s resources to source local speech data for its AI products. Through the Bill & Melinda Gates Foundation, Microsoft is working to reduce gender biases in data that permeate large language models that power chatbots.
Read More: Q&A: Is it the right time to invest in Gold, how much and in what form – Physical, ETF, or Gold MFs?
Google, too, is seeking assistance from Karya and other local players to gather speech data in 85 Indian districts. Google, then, is also planning to expand to every district in order to include the majority language or dialect spoken there. Based on this, the company hopes to build a generative AI model for 125 Indian languages.
Why Indian languages are important to tech giants
Current generative AI models that deal with language are focused on the English language, learning from data in the same language. India and its diverse pool of languages hold a lot of potential for Silicon Valley companies – as over a billion potential users are in India.
Unfortunately, non-English datasets are usually low quality and extremely limited. In addition, large language models (LLMs) often make up words and struggle with basic grammar.
Karya is headquartered in Bengaluru, Karnataka and is supported by grants. It targets workers in rural areas to tag and label local language content. The app works without internet and can provide voice support for those with limited literacy. Over 32,000 crowdsourced workers have logged into the app, according to Bloomberg, and 40 million paid digital tasks have been completed.
