Reports from the Heart of the Machine
Editors: Zenan, Boat
Regulators in some countries have begun to seek to address the root causes.
Since last year, the U.S. government has taken the problem of robot calls seriously. Effective June 30, 2021, the FCC requires phone providers to verify the IDs of all callers using an encryption-based protocol called STIR/SHAKEN.
As always, this approach doesn't cure the root cause, and anyone expecting robocalls to disappear from regulation will be disappointed, but new solutions are coming. Technology to block incoming bot calls is improving, and STIR/SHAKEN is part of the trend: phone users are no longer the only contributors to rejecting harassing calls. Now, phone providers must also share the responsibility.
The spam phone blocking app collects raw data about automated phone calls based on a sample of the user's phone. According to an index by the app Truecaller, the world's spam phone hit is Brazil. There, Truecaller users received an average of 33 spam calls per month in 2021, nearly double that of Peru in second place. (Even for Brazilians, that number has fallen sharply from a few years ago.) ）
Another app, YouMail, has been tracking automated calls in the U.S. since 2015. Its index shows that the number of automated calls in the country rose steadily for much of the late 2010s before falling by nearly half in early 2020 – possibly "thanks" to the COVID-19 pandemic. But the trend didn't last long, and by the end of the year, robophones had returned to 2018 levels.
"The growth of telemarketing has been ongoing, as it has been for a long time." YouMail CEO Alex Quilici said.
Fortunately, after the rise of artificial intelligence technology, apps will become smarter and smarter if apps eliminate spam calls. Computer scientist Mustaque Ahamad at The Georgia Institute of Technology and his graduate student Sharbani Pandit and others have created a machine learning system to try to solve this problem. His research, Applying Deep Learning to Combat Mass Robocalls, was recently included in the IEEE Symposium on Security and Privacy.
Ahamad describes the system as a "virtual assistant" that screens callers by asking a few questions, such as "who do you want to talk to?" or "What's the weather going on there," and by judging the other person's answer — or details like whether the caller interrupts the question — the natural language processing system can make an educated prediction about whether the call is a real person.
Could this be an AI-to-AI contest? It's entirely possible for bot callers to deploy their own natural language processing systems. But Ahamad says, "If you're going to make millions of calls and do it for every goal, you need a lot of computing power."
The system has a 97.8 percent success rate in pinpointing automated phones, and its creators want to deploy it as an application. Previously, researchers at Shanghai Jiao Tong University and UC Berkeley achieved a similar 90% success rate in another machine learning-based system for Chinese phones (the paper "A Machine Learning Approach to Prevent Malicious Calls over Telephony Networks").
Apps can work well on the receiver side, inferred from the phone – hinting at a larger problem. "Is this the most efficient solution?" Ahamad asked.
Historically, this is not the job that U.S. phone providers need to do. "Ten years ago, operators had a legal obligation to get all calls through without compromise," said Jim McEachern, an advisor at the Telecom Industry Solutions Alliance.
Theoretically, methods of blocking bot calls, such as blacklists, are never particularly defensible and only affect "legitimate" callers. In the 2010s, these lists became useless as bot callers adopted new methods such as VoIP and forged caller IDs.
"A few years ago, the FCC changed the rules to allow 'one call' to be blocked when it's not a real phone number,' McEachern said.
The STIR/SHAKEN directive now gives suppliers more responsibility to eliminate automatic spam calls with the help of caller ID.
While it's too early to say these effects, Quilici believes the number of robosets is already decreasing: "Probably an average of 10% less per month."
George Slover, policy advisor at Consumer Reports, a consumer product assessment magazine, said: "I think this is reducing the number of spambots that go through. Of course, not all of them will be blocked. As technology evolves, the system will be revisited and updated."
YouMail's results of tracking U.S. robot callers show that some robot phones have obvious application value, such as phone messages from schools; others are simple telemarketing, which is harassing calls that don't require a caller ID.
However, even as proponents of STIR/SHAKEN try to export the approach outside of North America, others argue that it is not enough. The regulation, and much of the report around it, is rife with comments such as "scammers" or "bad guys." But not all robocalls are scams.
Fortunately, there are some further precedents. Many jurisdictions have banned phone calls on weekends or holidays. In the EU, it is often illegal to dial a private number coldly without the explicit permission of the recipient.
In Brazil, where harassing calls are rife, starting in 2022, all telemarketers must use a number prefixed with 0303. In theory, this should make it easier to identify these calls and filter them out.
Similarly, U.S. law prefers to eliminate problems at the root. "The Telephone Consumer Protection Act requires, with some limited exceptions, that automated calls may only be made to consumers with their prior consent," Slover said.
Although it is impossible to completely eliminate spam calls, there have been people working on it.
For reference:
https://spectrum.ieee.org/ai-robocalls