It uses artificial intelligence (AI) to infer latent goals: The thing you actually want, not what you exactly asked.
Alexa Will Now Make Better Suggestions
The new feature uses a mixture of past behavior and current dialogue to determine what suggestions work best for the user. The announcement goes on to highlight the multi-step process required to get behavior like this right.
A Complex New AI Model
The machine learning-based trigger model determines whether there should be a suggestion based on a perceived latent goal. Each suggestion is decided by a deep learning-based sub module responsible for analyzing the likelihood of a follow-up interaction or whether the user’s initial question was meant to be a request.
This is paired with a model responsible for semantic role labeling—the process of assigning meaningful labels to words in a sentence. All of this information feeds into an optimized system for tracking which recommendations are helpful and which should be suppressed in the future.
Another Step Towards Natural Alexa Interactions
This addition is the latest in a long line of improvements to Alexa’s conversational skillset. Natural turn-taking was introduced in September 2020. This improves Alexa’s ability to determine when a user has finished speaking and whether the user is talking to the device or not.
If you want the interaction to feel more natural, you can also change Alexa’s name.
