3 reasons Google Assistant Duplex raises security, privacy concerns
Google Assistant Duplex introduces a new level in machine-driven conversation technology, but not without raising security and privacy concerns.
Back in May, Google unveiled Duplex, its artificial intelligence software that enables Google Assistant to conduct tasks over the phone via remarkably human conversations.
As the nascent virtual assistant space heats up, Duplex helps Google keep pace with Amazon, which established early dominance with Alexa. Google isn't touting Duplex as a business application as of yet. However, as conversational interfaces find their way into the workplace, Duplex enables Google to compete more aggressively with Alexa for Business and Microsoft Cortana.
How does Google Duplex work?
Relying on the engines of speech technology -- namely machine learning and natural language processing -- the Google Assistant Duplex brings believably human-like voice capabilities to Google Assistant, and it sets the stage for machine-to-person communication as a new channel for real-time engagement.
Not only has AI become adept at understanding speech, but it can now generate speech that humans cannot easily distinguish from a natural conversation with another person.
Virtual, Google assistant privacy concerns
While the prospect of every worker having their own virtual assistant is appealing, there are a host of privacy and etiquette issues to consider. Google certainly felt the "tech-lash" during the May demo, when it used Duplex to book an appointment at a hair salon without revealing the call was AI-generated.
This lack of transparency is challenging, because called parties think they're conversing with a human. Google has since moved to address that. Duplex-generated calls now state two things off the top: It's an automated service, and it's recording the call. It's easy to be wary of Google's intentions, and there is precedent for concern.
That said, Google Assistant Duplex seems to have passed the famous Turing test, where you can't distinguish between person and machine for the application. Aside from the duplicity that Duplex enables with unwitting participants, there are some other concerns of note:
- Consent for recording the call. As AI evolves, the implications around recording calls will only deepen. The legality of recording calls varies from state to state, so things can get messy quickly -- especially when using Google Assistant Duplex for things like national, or even global, outbound marketing campaigns.
- Data ownership. This begs the questions of where recordings are stored, for how long, by whom and for what purpose. And it raises an entirely new set of privacy and access concerns that could take AI in very dangerous directions.
- Duplex recordings used to simulate an individual's speech. Used with malicious intentions, Google Assistant Duplex could record your voice without consent. And even just a short conversation can produce enough audio samples to do a passable impersonation. With that in place, new possibilities open up for large-scale forms of identity theft, fraud and espionage.
Don't worry about Google's virtual assistant just yet
While Duplex could quickly become dystopian in the wrong hands, enterprises have little to fear for now. Google claimed it isn't targeting the enterprise or contact center spaces with Google Assistant Duplex, but that could certainly change -- especially if new business models emerge.
On a more granular level, Duplex thus far has only been developed for specific consumer use cases, like booking a salon appointment or a dining reservation. In these cases, the basis for conversation can be clearly defined. That's fine for a first-generation offering.
There's a big leap from that to general, open-ended conversation, where Duplex could have unlimited applications -- and opportunities for misuse. But that scenario doesn't appear to be an immediate concern.
It's worth noting that Duplex only works with Google Assistant, so it's a relatively closed environment. On the other hand, the machine learning engine for Duplex -- TensorFlow -- is open source and is used by companies such as LinkedIn, Coca-Cola, Tencent, Dropbox and Uber. There's no telling what types of conversational platforms these companies and others will build for themselves, so the story doesn't end with Google Assistant.
While the potential applications for enterprises are exciting, there's a lot to think about, and this is really just the beginning.
