Just after Thanksgiving the Cox Media Group (CMG) began marketing an advertising product that they say targets consumers based on private conversations heard by smart devices. This bold claim generated a good amount of media scrutiny, with most outlets saying Cox’s claimed capabilities were exaggerated. CMG has since taken their “active listening” marketing page down.
In my latest video I demonstrate how it’s possible to listen in on private conversations without ever having to upload audio data – just transcriptions generated by on-device AI. Smartphone processors have had enough horsepower to do this since at least 2017 if not earlier.
I conducted an experiment to test these capabilities. Using a piece of software called MacWhisper, which utilizes OpenAI’s models for on-device transcription, I transcribed a conversation from my home. The software efficiently converted the audio into text, which was then uploaded and summarized using ChatGPT. The results were surprisingly accurate and detailed, capturing various topics from health concerns to shopping plans.
The resulting transcript uploaded to ChatGPT was only 3k in size – a file small enough to be transmitted in just a few seconds using a 1980’s 1200 baud modem and mere milliseconds on a modern broadband connection. If anyone was monitoring the network traffic coming out of a smart television a transmission that small would likely be dismissed as just some random telemetry.
And you don’t even need a powerful computer to transcribe text on device. Google Pixel phones since the Pixel 4 could do it and Apple has had this capability since the iPhone X’s release. Conceivably every TV, phone, tablet, smart speaker and just about any other device made in the last five years is fully capable of on-device transcription.
In a statement, CMG denied they were listening to conversations but did not deny somebody else might be:
“CMG businesses do not listen to any conversations or have access to anything beyond a third-party aggregated, anonymized and fully encrypted data set that can be used for ad placement.“
So it’s entirely possible they’re working with a third party vendor that is conducting this activity through apps running on smart devices. CMG could just be buying the “output” of this transcription and AI processing. As of this posting CMG did not respond to my follow-up question asking if they were doing just that.
Is it legal? Cox Media Group thinks so. From their now deleted marketing page:
While the Cox Media Group’s claim about their advertising product could have been exaggerated, I demonstrated that it is now entirely plausible to listen in on private conversations, transcribe the audio to text in real time on-device, and transmit back very small blobs of text that can be interpreted by AI for advertising targeting.
I’d like to believe that CMG’s claims were exaggerated but it’s entirely possible advertisers have found a new way to invade our privacy for profit.