When we started Fluenty in 2014, Artificial Intelligence on personal devices was in its infancy. Amazon Echo and Google Allo didn’t exist, and the #chatbots hashtag wasn’t a thing. But we knew that the use of AI to make our lives simpler was going to be big. As machine learning engineers, we had seen the major advances of deep neural networks and glimpsed the opportunity. It was an exciting time.
To be a good companion to humans, AI needs to learn from them. By knowing what the human will want to do next, it can provide an answer before the human even asks for it. And the best way to anticipate a future query or action is to pay special attention to repeatedly used phrases and user behaviors. Luckily, humans are creatures of habit. We tend to do similar things over and over. And it’s particularly true when it comes to language. That’s why we first created Fluenty as an AI companion that can communicate for you by answering your messages.
We used Deep Learning paired with Long Short Term Memory technology to create the first AI that can answer text messages for you. Our deep learning model does not treat a word as keyword. Instead, it assigns meaning to a word in the context of conversations. For instance, our model understands that “thank you” and “I appreciate it” are very similar to each other in human’s conversational context.
We started by mapping over 700 million public chat conversations, searching for patterns in replies most frequently used by people. We found that around 30% of answers we send as replies to messages we receive can be predicted and thus suggested by AI. This research yielded incredible results that allowed us to launch “Talkey,” the original name for our product. But we didn’t want our app to be another messaging service. We made it so that it integrates with the ones already in use. From Facebook, through WhatsApp to classic SMS, the idea was to let people continue conversing with their network of friends faster and in a smarter way.
Over the last two years, we’ve iterated on Fluenty to make it even smarter. We reduced loading times from 7 seconds to 50 milliseconds, and we made it so that it can learn from each user. Now Fluenty is able to suggest smart replies that are specific to the person typing them. Let’s say for example that someone likes to say “Cheers!” instead of “Bye!” when closing a conversation. Fluenty is able to pick that up and will suggest it as a way to say goodbye in future messages.
But sending smart replies was only the first step towards what we had envisioned for Fluenty. Not all text messages prompt a written answer. That’s why we extended smart replies to smart-linking. The newest version of Fluenty connects with non-messaging apps and can anticipate not just what a person may want to say next, but also what they may want to do.
Layers below, a deep linking engine forges cross communication synapses with the most popular apps in use. Just at the mention of a movie in a message, Fluenty loads the trailer on YouTube and readies driving directions to the theater on Google Maps. No dinner, no problem. Fluenty’s assistive communication is on the task. Upon deciphering the issue from a message, it brings up a favorite restaurant on Yelp and makes booking a table one click away.
We designed Fluenty with the uniqueness of each user in mind. It saves time and frees people to focus on more demanding tasks. Fluenty is smart. The app thinks, assisting with decision-making. Fluenty is safe. All communication is anonymized as it crosses wires, and our servers encrypt it, keeping messages private. And finally, using Fluenty is fun. With its smart engine and a plethora of customizable tools, it makes not only its own intelligence, but also each unique personality shine.
Looking ahead, the ability to predict a user’s messaging response using Smart Reply, represents a significant opportunity in predictive notifications domain. This is in line with the direction much of interactive software is about to take. Instead of serving the purpose of leading a user down its usual destination, apps are becoming a network of a much larger, deeply-linked ecosystem that can be launched from a central location.
We are confident that Fluenty can serve as a central hub for launching and directing such a multi-app system. Our vision is to significantly broaden the capability and reach of assistant apps. This will allow the system’s learning ability to go beyond Smart Replies and move towards organizing and streamlining the user’s entire day.