By Vicente Radito David
In an age of whole chat conversations made only through emojis, a phone call feels outdated and (to my introvert self), invasive. Just send me a DM through my messenger, or if you’re feeling old-school, even an SMS. But for goodness sake, don’t call me. Even my mother doesn’t call me – and I’m her favorite.
To be fair, a phone call still has its uses. If your loved ones are away, it’s always a relief to hear their voices at the other end of the line. When you need quick solutions to things that would require explanations, it’s usually faster to just pick up a phone than type a long message. And when you have a complaint or need clarifications from your bank or internet provider, your first instinct would be to call their Customer Support to voice your concerns.
A few days ago, Google emphasized the utility of phone calls by wow-ing the world with a new AI-system designed to accomplish real-world tasks over the phone: the Google Duplex.
NOT YOUR USUAL ROBOCALLS
Automated voice messages are hardly new. In fact, we’ve learned to hate them already. Utilized by incessant companies and shrewd politicians, robocalls use a computerized autodialer (a robot) to deliver a pre-recorded message.
But to compare a robocall with Google Duplex is like equating the common housecat to a Bengal Tiger – some similarities are in place, but it’s a different animal altogether.
In CEO Sundar Pichai’s keynote at the Google I/O 2018, he shared an astounding demo of Google’s Duplex AI system where the program made a call to a salon to make a haircut appointment. To say that the demo was a revelation would be an understatement. The AI system successfully talked to a human and passed the Turing Test with flying colors. Unlike robocalls that you can identify in seconds, at no point did the person at the end of the line realize that she wasn’t talking to an actual human. The conversation sounded natural – with natural pauses, the umms and uhuhs of human dialog. And even more important, the AI was able to accomplish its task.
[For online, here’s a link to the official Duplex haircut audio: http://www.gstatic.com/b-g/DMS03IIQXU3TY2FD6DLPLOMBBBJ2CH188143148.mp3 ]
A second demo (a call to reserve a table at a restaurant), was even more impressive as it showcased how good the AI system was in understanding context. Despite the mistakes made by the person on the line, Google Duplex was able to sift through the errors and provide an appropriate response. This, more than anything, signaled that Duplex has the potential to be much, much more than a virtual assistant making appointments through the phone.
[Link to the Official Duplex restaurant audio: http://www.gstatic.com/b-g/KOK4HAMTAPH5Z96154F6GKUM74A3Z1576269077.mp3]
WHAT MAKES IT SPECIAL?
Not only is the Google Duplex good enough to sound like an actual human on the phone (a remarkable feat in itself), it also needs to be able to handle data and make decisions on the fly. Suppose that you asked Google Duplex to set an appointment with your doctor on Monday, at 3 pm. If your doctor’s secretary said that the only time the doctor is available would be at 1pm, Duplex would have to understand what the answer means, know if you’re busy at that time (do you have an office meeting set?), and even if you can make it to the doctor’s office at that hour (if you’re engaged until noon, will there be enough time to go to the doctor’s office given the usual traffic at that hour and day?). Duplex needs to parse all that data and come up with a decision fast enough for the human waiting on the phone.
Furthermore, Duplex is able to understand the nuances of language – at least, to a narrow field (for now). Just imagine the different ways we can say ‘sure’ – why not, for real, you betcha, 5 by 5, clear as crystal, certainly, no probs, hell yeah… it’s a lot. Duplex would need to analyze the phrases in the conversation and put them in the right context to follow along.
As writer Arthur Clarke would describe it, the technology behind Duplex would certainly seem like magic.
Right now, Google Duplex is limited to a few basic tasks such as making appointments and reservations. But the technology is promising.
It would be easy to imagine how we can use Google Duplex in the future. Say you’re an enterprising bride-to-be who’s handling all the details of her wedding. Instead of spending hours trying to call various caterers and asking for a quotation for a menu you want to try, you can ask Google Duplex to make the same inquiry to dozens of caterers and save time, energy, and patience – invaluable resources for a woman about to have a wedding.
What if the government can use Google Duplex for an automated national voice service? It can be used to call people to remind them to register as voters before the deadline comes, or to give emergency information to people – specific to their needs – at times of calamities.
Imagine how useful Google Duplex would be if it were integrated to other technologies (like self-driving cars, or even computers) for people who may have particular physiological challenges or just plain tech-shy. The ability to just talk to something that sounds like a person may be of tremendous help for some people.
THE DARK SIDE
If you were a bit creeped out after watching the Google Duplex demo, you’re not alone. The idea that a program could so effectively mimic a human in a voice conversation can be truly disconcerting. But the list of concerns for Google Duplex doesn’t end there.
Imagine how much more effective scammers would become if they can have access to technology similar to Duplex. Imagine a budol-budol call – scaled for millions, believable as heck, individually contextualized for maximum effect. Imagine a corrupt politician engaging in smear campaigns using a trustworthy-sounding Duplex. Imagine manipulating elections by calling voters (projected to choose a certain candidate) in a specific barangay, telling them that their precinct has been moved.
Once Google Duplex grows in complexity and expertise, will we still need human customer support staff? I don’t see how, especially since Google could program Duplex to be much more patient and efficient than a human agent. What would happen to economies that are significantly supported by such industries?
What if the person you thought you knew – the person you’ve been sharing intimate details with – turned out to be a mere computer program? Are we ready for a world evocative of the Joaquin Phoenix/Scarlett Johannson film ‘Her’?
Technology is always a double-edged sword. As useful as it is, someone can always twist its usage and use it for harm.
Google’s new AI system is truly exciting and can potentially initiate a paradigm shift in many industries – a term not to be awarded lightly. It would be good to monitor the Google Duplex for its future uses, and even more importantly, to be vigilant against the possible abuses villains can do with it.