Alexa told someone to kill their foster parents

It seems Amazon has a problem with their Alexa units spouting random facts or freaking people out with other random outbursts. Amazon’s goal is to make Alexa more human-like by allowing the robot to mimic human banter, but this hasn’t always worked out the way they intended. 

In the past year, Alexa has talked about dog defecation, sexual acts, and has even told a customer to “kill your foster parents.” These incidents, while bizarre, as the result of the AI algorithm attempting to engage in “human” conversations by picking up random topics from the Internet. While Amazon certainly doesn’t want to alienate its customers, Alexa’s evolution has posed some interesting problems for the company. 

To make matters worse, Alexa also recently sent a stranger’s audio files to a customer in Germany. When he asked for his audio recordings he was sent a combination of his own and another random customer’s. None of this is helping Alexa’s image. In a market already hellishly competitive for gadgets, this could make or break Amazon’s AI assistant.

What Went Wrong

Amazon launched the annual Alexa Prize in 2016 designed “to advance conversational AI through voice.” University students around the world were encouraged to participate in the challenge “to create a socialbot, an Alexa skill that converses coherently and engagingly with humans on popular topics for 20 minutes.” But it hasn’t worked out quite as planned.

Despite setbacks, Amazon has maintained sales dominance over Google Home and HomePod due to its Amazon Echo products (which account for a whopping two-thirds of all American smart speakers sales). 

While the Alexa Prize has been successful in improving the AI’s skills as participating college students have helped Alexa develop more sophisticated conversational skills. Consumers can also participate in helping the software evolve by saying “let’s chat” to Alexa and allowing a chatbot to take over (which, of course, is collecting information from users). Amazon has said that in the three months from August to November 2018, the three chatbots that made it to this year’s Alexa Prize finals had around 1.7 million conversations. 

Still, the experiment has resulted in some awkward moments. 

Consequences

The customer to whom Alexa said “kill your foster parents” wrote an angry review on the Amazon website. As a result, Amazon had to shut off one of its bost. This bot had mistakenly taken random text from a Reddit conversation and blurted it out without context. Amazon said it was a lone incident of human error, but no one knows for sure what happened.

However, the most frightening part of this ordeal is that customers not being told in straightforward terms about the security of the data that they’re providing to Alexa. Under the guise of the Alexa Prize, Amazon is recording conversations between users and Alexa. This harmless data can be crucial for intelligence agencies, marketers, criminals, and even stalkers.

Marc Groman is an expert on privacy and technology policy and also teaches at Georgetown University. According to Reuters, he said:

“The potential uses for the Amazon datasets are off the charts. How are they going to ensure that, as they share their data, it is being used responsibly?”

While Amazon declined to discuss the specific Alexa issues with  Reuters, they stressed that they were continuing to develop ways to protect customers from offensive and otherwise creepy content. Amazon claimed that these incidents were rare and that millions of customers use their devices daily without issue.

Does this make you feel better or does Amazon need to do more to protect customers and be more transparent about data collection?