Google Assistant stole the show at Google I/O 2018 with a demonstration of Duplex. And why not? Utilizing some of the latest advancements in deep neural networks, Duplex is the closest mankind has come to sci-fi levels of AI. Of course, “advanced” is a controversial word to use when it comes to any kind of technology. It’s time to take a look at the good, the bad, and everything in between about Google Assistant’s eerie future.
Duplex — Turning your Phone into a Secretary
Google Duplex is a marvel of modern computer sciences and a genius implementation of deep learning. Most importantly, it is an innovation that can see a lot of good done in its time. Duplex can serve as a platform for individuals suffering from speech impediments. It might be a useful tool for sufferers of social anxiety. It may also partially ease problems caused by language barriers between communicating parties.
Google Duplex does not tire of managing schedules. It does not lose patience with irate contacts. It doesn’t need a monthly salary. In essence, it turns your phone into a virtual personal secretary. And while it is still a far cry from Tony Stark’s JARVIS, it’s definitely another step in the right direction. If only the idea of conversing with a soulless machine faking humanity wasn’t so uncanny.
The Sounds of the Uncanny Valley
The term “uncanny valley” refers to the aversion people feel when presented with a stimulus that resembles humans, but is not actually human. Traditionally, this has referred to visual stimuli. However, the uncertainty over Google Duplex’s emulated speech seems to suggest it can apply to other stimuli as well; in this case, sound. Duplex initially sounds very natural, but the longer the conversation goes on, the more differences you can pick up versus the speech pattern of a human being. For instance, while the AI is capable of using mannerisms like “uhmm” and “hmm” in the right context, it seems to use them as preprogrammed responses to a set of conditions.
Consider the second phone call in the demo. “Uhh…” is commonly thought of as a verbal cue indicating uncertainty. When Duplex first asks about Wednesday the 7th, it uses an “uhh…” as if it is confirming with itself if the date is correct. This sounds very natural, as it is something most people do. It uses another instance of “uhh…” when questioned about the date. By the third time, it uses this when referring to the scheduled date, “uhh” no longer sounds like a verbal cue of uncertainty, and more like a guaranteed programmed response.
Duplex’s Shades of #7A7A7A
There are greyer areas than the subjective perception of the uncanny valley, however. For instance, what are the ethical implications of a machine that can mimic human conversation? What are the limits to the information AI can make use of in conversation? Should people be aware that the voice they are hearing is that of a machine? Remember: assistants need to know information about who they work for in order to get things done, and Google has incredible amounts of information on each user.
Is it possible to trick AI into divulging information? Even without those risks, one can argue that the abuse of Duplex’s advantages is also an issue. Duplex may be used as a tool for people with social anxiety, but what if these people become dependent on it? Would it still be a tool for their benefit? It’s a shady road we have to walk on from this point forward.
CEO Sundar Pichai mentioned that Google is still developing the technology, and so perhaps are dealing with these issues even as you read this. Until these questions are answered, I’ll probably make my own phone calls, thank you.