‘Alexa, shut up!’
When Siri, Alexa, Cortana and…erm…Google were first introduced to the home there were some concerns over privacy. Something only natural after decades of film and TV warning us about crazed AI taking over the world.
Now, scientists at Imperial College London are investigating this as a possibility. However they’re not worried about a full world invasion, just on on a smaller scale.
A new study investigating whether smart devices can be accidentally activated by non human interaction. In this case non human interaction means voices from your TV.
Over a six month period the experiment observed smart home devices while TV shows played in the same room. The study used various ways to detect when smart speakers record audio. They filmed devices (to monitor activation light up) and monitored network traffic (to detect audio data sent to the cloud).
The experiment tested an Amazon Alexa system, a Google Home device, an Apple HomePod and a Microsoft home device. They paid close attention whenever TV characters used what might be considered a trigger word. This means words like “Alexa,” “Amazon,” “Echo,” and “computer,” or “hey,” “hi,” and “OK” (or “OK Google” for Google Home.)
The study used 125 hours of audio from various Netflix TV shows. The results found that the tested smart devices were activated between 1.5 and 19 times in a day! A bit of a terrifying figure when you consider 125 hours is only 5 days!
The most sensitive speakers?
The experiment found that the Apple HomePod and the Cortana home device were the smart speakers activated most frequently. The other devices accidentally activated most frequently were the Amazon Echo Dot series 2, the Google Home Mini, and then the Amazon Echo Dot series 3.
However, it’s worth noting that Amazon is on their third generation of the Echo Dot (series 3) – the frequency of activation declined as the series progressed. So Amazon are clearly improving their voice recognition in each iteration.
What’s interesting (or creepy?) is that certain speakers responded more often to certain shows. It’s unsurprising that the research found ‘dialogue heavy’ TV shows were the most triggering for the speakers. Notable shows that caused speaker reactions were The Office (US) and Gilmore Girls. Both are shows known for their rapid fire dialogue.
Certain words or phrases cause activations in a variety of TV shows. Phrases like “What kind of”, “Okay, but not” or “I can spare” activate Google devices the most.
For Apple devices words that rhyme with ‘Siri’ cause the most activations. Examples include “He clearly”, “They very”, “Hey sorry”. For Amazon Alexa it’s words that rhyme with Alexa such as “exclamation”, “Kevin’s car” and “congresswoman”.
That’s my name, don’t wear it out.
Good news, there isn’t any AI world domination in sight for now! Ultimately the experiment concludes that while smart devices can respond to unintended phrases they aren’t always recording us. Going forward the researchers will look into other factors that cause activation.
So if you’re home alone watching TV and Alexa starts up, don’t worry! She’s just a Gilmore Girls fan!