Info-tech

‘Data biases can skew outcomes of AI-based systems’

KV Kurmanath Hyderabad | Updated on July 30, 2019 Published on July 30, 2019

How good or bad Artificial Intelligence is? It is as good or as bad as the data that is fed into the system. If data is biased, the outcome of an Artificial Intelligence system could well be biased as well.

While AI could significantly improve the qualify of our lives, it can pose challenges depending on who uses it and how. It can also devour jobs as it engulfs sector after sector.

Subbarao Kambhampati, a professor with Arizona State University (the US) and an AI thought leader, shared his thoughts on all these and much more at a Manthan meeting held here on Monday.

“AI is a two-edged sword. It can be used by the good and the bad. While all technologies have dual use, Intelligence is the ultimate dual use technology,” he pointed out. “Biases can be amplified. Data biases that exist in the data that is piled up will lead to biases in understanding and outcomes of the AI systems,” he says.

He gave an example of how biases can influence the AI outcomes. “You key in the filter ‘professor’ in a search engine and you get hundreds of search results with the images of men. You have to scroll down deep to find women professors. If the AI systems are fed with such data, the biases would creep in the outcomes,” he cautioned. Subbarao said AI systems are far from becoming more intelligent than human beings. “They don't have common sense yet. Computers are super intelligent, but in narrow areas,” he pointed out.

“There is a ‘land rush’ mentality in industry to exploit AI technology before their competitors. Anecdotal evidence suggests that there is a push to convert all problems to pure data,” he said.

Faking machines

“AI systems themselves can pose headaches as they could talk to your Alexas in your own voice or your mother’s voice. They can create fake news, and can churn out fake images and fake narratives,” he warned.

He, however, felt that the problems that AI creates can be solved by AI itself. “It can’t kill non-routine jobs, while automating several routine jobs,” he observed.

Published on July 30, 2019
This article is closed for comments.
Please Email the Editor