Researchers Find A Possible Solution To The Problem Of Robocalling Using Machine Learning

The United States government has begun to take significant steps to eliminate robocalls. The FCC requires that phone companies use a cryptography-based technology called STIR/SHAKEN to authenticate all callers’ IDs beginning June 30, 2021.

Anyone hoping for robocalls to evaporate in a puff of regulation will be sorely disappointed. However, respite may be on the way, albeit slowly. The technology to block robocalls is developing, and STIR/SHAKEN is part of a trend in which phone consumers in the United States are no longer solely responsible for deciding whether or not to accept robocalls. Now it’s up to the phone companies to do their part.

Spam-call-blocking apps collect hard statistics on robocalls using the phones of their users as a sample. Brazil is the world’s spam call nexus, according to one such index published by the app Truecaller. In 2021, the average Truecaller user received 33 spam calls per month, roughly double the rate of Peru, which came in second.

Since 2015, another program, YouMail, has started tracking robocalls in the United States. Its indicator reveals that the number of robocalls in the United States increased significantly throughout the late 2010s until plummeting by nearly half in early 2020, owing to COVID-19. It was short-lived. Robocalls had already returned to 2018 levels by the end of the year.

If it’s up to apps to filter out unwanted calls, it’s a good thing they’re getting smarter. AI is currently being used by some. A group of researchers from Georgia Tech University developed one machine-learning-based solution.

The technology is described by the researchers as a virtual assistant that asks callers a series of inquiries, such as “whom do you want to speak to?” “How’s the weather where you’re phoning from?” or “How’s the weather where you’re calling from?” The natural language processing system can make an educated judgment as to whether a call is real by analyzing the answers—or details like if a caller interrupts the inquiry.

It’s feasible that robocallers will develop their own natural language processing algorithms. However, the researchers believe that making millions of calls for each and every target would necessitate a significant amount of resources.

The company hopes to release this technology as an app after it successfully identified 97.8% of robocalls. Researchers from Shanghai Jiao Tong University and the University of California, Berkeley developed a machine-learning-based system for Chinese mobile phones that had a similar 90 percent success rate.

Apps can operate effectively on the receiving end, one phone at a time—which raises a bigger question. Where do you put a stop to these phone calls?

What robocall-culling mechanisms there were, such as do-not-call lists, were never especially effective and only affected legal callers. As robocallers embraced new methods like VoIP and forging caller IDs in the 2010s, the lists became even more worthless.

The FCC revised the rules a few years ago, saying that a call can be banned if it’s not a real phone number coming in, for example.

The STIR/SHAKEN regulation now places even more pressure on providers to reduce robocalls by combating caller ID spoofing.

It’s too early to tell how effectively the adjustments are working, but there are reasons to be cautiously optimistic. Experts believe that the number of robocalls has decreased by about 10% on a monthly basis.

Even while proponents of STIR/SHAKEN try to expand the protocol beyond North America, others claim that it does not go far enough. The rule, as well as most of the press around it, contains terms like ‘scammers’ and ‘bad guys.’ However, not all robocalls are fraudulent.

The top 15 robocallers in the United States are all genuine callers, according to YouMail. Some have clear value—for example, school messages—but others are payment reminders, pointless customer contacts, or basic telemarketing, all of which are nuisance calls that don’t require caller ID. They will not be stopped by STIR/SHAKEN.

Thankfully, there is a precedent for going even further. Many jurisdictions have already made it illegal to make cold calls on weekends and holidays. Cold calling a private number without the recipient’s specific authorization is typically prohibited in the European Union.

Starting in 2022, all telemarketers in Brazil’s nuisance-call-infested country will be obliged to use a number with the 0303 prefixes. In principle, this should make it easier to detect and filter out those calls.

Similarly, the law in the United States encourages silent phones. With a few exceptions, the Telephone Consumer Protection Act stipulates that robocalls be performed only with the consumer’s prior consent.


It’s unlikely that robocalls will ever be eradicated entirely from the airwaves. However, when the technology to combat them improves and regulations change, it’s feasible that they’ll become like spam emails: present, nasty, but not clogging the tubes of a whole medium, according to those fighting them.



🐝 Join the Fastest Growing AI Research Newsletter Read by Researchers from Google + NVIDIA + Meta + Stanford + MIT + Microsoft and many others...