AI: Alexa guesses what you want even before you ask for it
Amazon engineers are modifying Alexa’s algorithm to help the virtual assistant guess user requests and resolve them even before they are spoken.
After asking, for example, how long it takes for water to boil tea, Alexa will suggest setting a timer for the recommended number of minutes.
Studies on Alexa
Anjishnu Kumar and Anand Rathi, engineers from Alexa, explained that continuous improvement comes from the desire to make interactions with the virtual assistant as natural as possible.
Chatting with Alexa should be as natural as talking to another human being and one of the main keys to this, according to them, is to allow technology to anticipate what comes next in the conversation.
“Now, we are taking another step towards natural interaction with an ability that allows Alexa to deduce clients’ latent goals – goals that are implicit in clients’ requests but not directly expressed,” said Kumar and Rathi.
Achieving this degree of intelligence for a virtual assistant is difficult and requires a number of sophisticated algorithms.
Difficulty of implementation
In order to understand what the users’ latent goal is, Alexa has to analyse the multiple characteristics of their requests and compare them with previous interaction models.
The model has to learn from customers’ behaviours, remembering for example that users who ask for how long they need to boil water to make a tea, often later require the setting of a timer for that period of time.
Equally challenging is the process of creating a follow-up suggestion based on the information Alexa identified in the first request.
The algorithm must understand the context in which the user’s words are spoken in order to structure the information for the next task to be performed.
Amazon engineers have developed a so-called “context carryover model” to enable the transition.
One of the most difficult tasks was to find out if the virtual assistant could understand the users’ unspoken requests.
“Our first experiments have shown that not all dialogue contexts are suitable for discovering latent goals,” said Kumar and Rathi. “When a customer asked for ‘chicken recipes’, for example, one of our initial prototypes asked erroneously: Do you want me to reproduce the sound of chicken?”
The engineers used an in-depth learning model that took various elements into account in the dialogue with the customer before deciding whether or not a suggestion should be activated.
The algorithm makes an assessment based on factors ranging from the text of the dialogue to users’ previous behaviour towards the virtual assistant.
“We are excited about this invention because it helps to discover Alexa’s skills and provides greater utility to our customers,” Amazon engineers said.
Although Alexa’s experts insisted that the algorithm suggests a follow-up only when it finds the right context, it is easy to imagine how invasive Alexa can become if the technology is malfunctioning.
If it starts to misunderstand the context of customers’ needs and starts asking irrelevant questions, the technology could become a nuisance to users.
For now, the new feature is available in the US, and requires no further action by developers to activate it.