Author: Sagi Eliyahu / Source: VentureBeat
The news about Amazon’s Alexa bursting out in spontaneous laughter would have been funnier if it hadn’t cast a spotlight on what’s wrong with the current approach to chatbots and virtual assistants. Amazon researchers programmed Alexa to deliver a rolling “ha ha ha” when users said, “Alexa, can you laugh?
” The device began misinterpreting audio, resulting in random, fairly terrifying cackling. Amazon quickly corrected this by changing both the command and response to behave more clearly and predictably: “Alexa, can you laugh?” now results in, “Sure, I can laugh. Tee hee!” This is a good, simple fix, but what can we learn from this event?Makers of chatbots and virtual assistants are trying too hard to make them seem human. It’s not just Alexa. Ask Siri “What is zero divided by zero?” and she’ll send you to a burn ward, or ask Google Home a series of standard questions and you’ll receive some weird answers. Sure, the cheekiness is fun, and a certain level of humanness makes them more accessible, but it is also what’s preventing this technology from reaching its fullest potential.
In reality, chatbots and virtual assistants are just another type of computer interface. Just like other interfaces, they are tools expected to process information and complete the tasks users assign to them. As with every other interface, the user experience is paramount. A number of studies have shown that assigning certain human traits (like voice tones or the ability to tell jokes) helps users relate to and trust computers.
It makes sense why companies would make chatbots and virtual assistants humanlike. But trying to make something accessible and trying to make something human are two very different things.Alexa’s laughing jag has been described as, among other things, “creepy,” “evil,” “bone-chilling” and “freaky.” That’s not particularly accessible. Humans expect computers to behave in predictable ways. When…
The post Alexa’s laughing bouts highlight a fatal flaw in bot development appeared first on FeedBox.