It is safe to predict that the overwhelming majority of news stories will say artificial intelligence is going to protect the public from harm by scanning their calls and warning them if it sounds like they are being scammed. But as you knew that already, you came here because Commsrisk does more than just uncritically repeating the marketing claims made by firms that sell technology. False positives, where genuine calls will also be blocked or soft-blocked, are inevitable. It makes a refreshing change to see news from KT, the South Korean telco, that admits they had false positives when using new AI software that runs on the handsets of consumers and listens to their conversations.
BreakNews reports that the AI identified most of the voice phishing calls received by handsets since the service was launched on January 22.
According to the analysis, the detection accuracy reached 90.3% among 1,528 verifiable voice phishing calls detected as ‘Caution’ and ‘Dangerous’ since the launch of the service. In particular, 392 cases (25%) were confirmed to be on the National Police Agency’s voice phishing blacklist or cases of prosecutors/police impersonation, proving that the notification function is effective in preventing actual damage.
They were less specific about the frequency with which legitimate calls were mislabeled.
However, there were some cases of false positives in normal calls related to billing or payment, and KT plans to further improve accuracy within the first half of the year by improving the AI engine and updating the whitelist to minimize these cases.
In case you have not noticed, there is a lot of enthusiasm for AI amongst telcos. One reason is because AI will slow the terminal decline of telephone numbers as an addressing system. Consumers will increasingly prefer closed systems like WhatsApp that require a user to accept an invitation to connect with another user before the channels of communication open further. This is a fundamentally better way of preventing bad people from contacting you. Businesses and governments will be keen to preserve the use of phone numbers as addressing systems because they are highly motivated to spam users, and most of the motivation for anti-spam programs comes from organizations who want to maintain the effectiveness of their spam by inhibiting other spam that competes for attention. Too much spam leads users to turn off completely. That is why we are going to see a lot of people jockeying for power over the way filters work, although they will all pretend they are only motivated by their desire to serve humanity.
Turning each user’s phone into their primary inbox for junk is cheaper than sending junk mail through the postal system and is better for eliciting a response than email. The fact that some people would prefer to deal with another human being face-to-face will not matter because employing people to work in a building accessible to the public is really expensive, employing them to sit in a call center is somewhat cheaper, and replacing people with bots is much cheaper. We are witnessing the emergence of a technological arms race between AI bots that make calls, AI to interfere with calls, and AI to combat the AI that interferes with calls. This begs the question of why users would not choose to use AI-powered filters to stop business and government spam too, unless governments and businesses will control the detail of how AI works, and thus give themselves more perfect control of human interactions than anything imagined by Orwell or the Stasi. Any innovation designed to work around such filters will likely be adopted by criminals as well.
You can read the BreakNews article here.
