The Indian Air Force must shift from fixed systems to AI-based probabilistic models for its 'zero-error' air operations, said Air Vice Marshal Ajay Kunnath. He highlighted the goal is to achieve trust, failsafe operations, and full autonomy.
Air Vice Marshal Ajay Kunnath on Tuesday said that the Indian Air Force should shift how it uses technology for air operations, highlighting the importance of how such operations function in a "zero-error" environment. Speaking with ANI on the sidelines of the AI Impact Summit, he explained that the force is moving away from fixed systems toward AI-based models.
"We are in a deterministic area at this point. We have to get into a probabilistic domain," he said. Because air operations function in a near "zero-error" environment, he stressed that trust is essential. "The type of domain that we talk about, like air operations specifically, actually mandates trust. And that is implicit. We are in a domain which is sort of zero error. So we need to now have probabilistic solutions which take us to that trust and failsafe operations."
The Goal: From Automation to Autonomy
Summing up the goal, he said, "If you really want me to sum it up, it is trust, failsafe, moving from automation to autonomy."
Evolving Human Involvement
He also described how human involvement changes as AI systems advance. "We gravitate from human in the loop, where he becomes a decision maker, to human on the loop, where he uses it to his advantage, and finally autonomy, where it goes human out of the loop. You have to decide which area you are into and which facet of the human you want to use -- in, on or out."
Key Challenges in AI Implementation
He warned that AI solutions cannot simply be copied from one field to another. "You can't take one particular scenario and think that it can replicate or manifest itself in the other domain," he said.
Closing the Accuracy Gap
Finally, he pointed out a key challenge with AI accuracy. "Your probabilistic takes you to 95 per cent. But the fact remains that it is the last five percentile that actually determines a leaker who can come in," he said, adding that better models and higher-quality data will be needed to close that gap. (ANI)
(Except for the headline, this story has not been edited by Asianet Newsable English staff and is published from a syndicated feed.)