Shield telcos from liability even if AI blocks messages: Vi

India’s third-largest carrier has already harnessed AI-ML to craft a solution for blocking spam without directly deploying an AI-ML engine. That said it will take some more time to fully deploy AI-ML system as telcos are expecting some initial obstacles such as the accidental filtering of bona fide messages, incorrectly identified as spam.

“Deploying any specific AI/ML model in live requires validation against large data sets and is still work in progress. Models have high chances of blocking true positives and any decisions on AI/ ML has to be carefully assessed and can be proceeded only when there is immunity from both the data set owner and the regulator,” Babu said.

Even as telecom companies implement AI algorithms to thwart unsolicited communication, demand for such measures is growing following an uptick in fraudulent spam messages, especially on WhatsApp. Trai has instructed telcos to employ AI to assist millions of Indian users. Spammers are exploiting the Meta-owned messaging platform by making calls from international numbers to unsuspecting users to defraud them.

According to a Localcircles survey published on Wednesday, 46% of Indian WhatsApp users received unsolicited audio and video calls, which has surged in the past month. It said 59% of calls originated from unidentified international numbers.

In a previous attempt to block unsolicited communication, the regulator had, in March 2021, mandated telcos to use blockchain or distributed ledger technology , but it resulted in massive disruption in financial and UPI transactions as the scrubbing mechanism ended up blocking messages carrying one-time-passwords or OTPs Trai had to temporarily suspend the mandate but re-instated it later as the menace of spam messages and calls continued.

Babu said that Vodafone Idea has already methods and procedures in place that already have shown better results even without deploying the AI/ML model and has shown that it is able to better control and protect its own consumer base that receives spam messages using the deployed DLT technology. He added that the solution had been showcased to Trai and few banks.

He added that the AI model, which learns from inputs that are provided to it and evolve, could possibly identify marketing messages being sent by, for instance, a bank, as spam message to customers and could likely block them, thus may come in way of the marketing efforts.

He added that in such cases, feedback to close loop the analysis from the marketeer which is the actual messenger on an ongoing basis is needed to improve the model and deploy it effectively.

“AI/ML model takes longer time to mature to provide desired levels of accuracy” he said.