AI considered more as market opportunity

Artificial Intelligence (AI) is reverberating in India more than ever but the government, through the lens of “#AIForAll” policy, currently sees Artificial Intelligence (AI) only as a market opportunity. But there are societal implications as AI is not just a mathematical model but a socio-technical system with inherent bias. Considering the diverse nature of the country, the applications being developed are for a society with economic inequalities where discrimination persists.

A search on Google several times gave absurd responses, which may sound funny but are a painful reflection of society’s stereotypes. A Google search for “South Indian Masala” returns women instead of spices. Likewise, translations have gone haywire with “Dalit man” being returned as underdog. Treating the bias in the system as bugs only means the systems are fragile.

A lawyer and researcher, Vidushi Marda, in her paper ‘Artificial Intelligence Policy in India: A framework for engaging the limits of data-driven decision-making’ argues that data-driven decision-making is susceptible to inaccuracies, discriminatory outcomes, embedded and exacerbated bias, and even unintended consequences due to various limitations that occur through the process.

Ms Marda said, “Most AI systems train on historical data and are capable of uncovering patterns, learning from examples and predicting future outcomes for the purposes of decision-making. While these predictions and classifications are generalisations

■ A SEARCH on Google several times gave absurd responses, which may sound funny but are a painful reflection of society’s stereotypes.

■ MOST AI systems train on historical data and are capable of uncovering patterns, learning from examples and predicting the future. that humans wouldn’t be able to analyse at similar speeds and scale, they are not perfect. They learn from data that emerges from an unfair world, filled with historical discrimination and inequality”.

She adds that the data these systems train on is collected by individuals who have their own biases, as all humans do. Consequently, machines trained on data tend to reflect these biases. It becomes imperative to understand the sources of bias and pitfalls of these systems before we begin deploying them.

Experts said that the rapid rate at which development was going on and lack of transparency in such developments were irreversible. Due to this, the conventional method of building processes and deploying technology first and deliberating on their effects next will not work with AI.

An independent security researcher, said, “AI driven surveillance will give scope for self censorship as people are unsure of the status and implications. We are implementing cameras wherever we want, but do not really understand the privacy issues.”