Alexa Conversations – FI

Alexa Conversations Now Available to Developers in Beta with a $100,000 Contest – Here’s What it Really Means for Amazon and Devs

Alexa Conversations this week emerged from its year-long restricted-access preview as a beta feature in Alexa Skills Kit (ASK). Amazon is hoping to increase developer adoption by running a contest in collaboration with Devpost with a $100,000 prize purse. The company has run many such contests in the past in order to incentivize developers to adopt new features and have more working examples in the Alexa skill store. However, it has not run any contests over the past year so you can take that as a signal that Amazon is very keen to have developers come on board with Alexa Conversations.

What is Alexa Conversations?

Alexa Conversations was first demonstrated at Amazon’s Re: MARS conference in 2019. At the time, Amazon VP Rohit Prasad company highlighted Alexa’s ability to fulfill a complex customer intent that required multiple third-party Alexa skills to complete. The original demonstration was for a customer planning a night out that included a dinner reservations, purchasing movie tickets, and arranging a car service. However, Prasad characterized that feature as cross-skill goal completion and a couple of different terms have been used since then. Many media mischaracterized this as Alexa Conversations after the demonstration.

Alexa Conversations is actually a narrower feature for developers. It enables developers to define specific goals required to complete an interaction and define a sample dialog. Amazon’s Paul Cutsinger told Voicebot in an interview, “You just lay out the main script to identify what information you want and what API call you want. I don’t need a dialog for every possible combination…The developer annotates what slots they want and what goal is for the utterance. Alexa Conversations builds out [the solution].”

Most Alexa skills are developed today by anticipating the wide variety of ways a conversation could play out and then defining the dialogs. This requires a lot more code and time for the developer. Steven Arkonvich of Philosophical Creations, the developer of the popular Big Sky Alexa skill, worked with Alexa Conversations during the preview. He told Voicebot that with Alexa Conversations “is a very different experience compared to custom skill development. Although some basic concepts carry over, for example, slots and sample utterances, their application – what the developer does with them – is much different. You don’t really send intents and slots to your backend and keep track of the conversational state and responses there. Instead, you are modeling much more of the conversation in the developer console.  Where the user is in the conversation, and what sorts of responses are given, are kept track of by the Alexa conversations AI, guided by how you’ve modeled the conversation and its possible permutations.”

“It does promise some significant improvements, I think, for the end-user and the developer. There is less backend code to write – though there is more work required on the front-end. And there are gains in context carryover. So, for example, if the user changes their mind about what they said two conversation turns ago, Alexa Conversation should be able to keep track of that. That was doable before, but considerably more difficult,” Arkonvich added.

Amazon provided an example of this scenario in a demonstration video embedded below. It offers an example of how a pizza order could be placed by a user. The skill needs certain information, such as the size of the pizza, crust type, and toppings in order to complete the order. The information can be gathered in any sequence because Alexa Conversations knows what is required and maintains session context of what was said earlier. In addition, Alexa Converations enables users to go off script such as asking “how many people can that feed” for a particular pizza size and answer that question before coming back to ask for the information still required to complete the order.

Developers and designers in the past would have to model these many variations. Now, they can focus more on defining a “happy path” dialog, identifying which slots (i.e. information needed) should be filled by Alexa Conversations and annotating that sample dialog. The expectation is that this will be easier for the developer and provide more robust user experience for the user. For those of you with experience in this type of development, it is a departure from standard practice of defining numerous dialogs and in many ways more similar to what is asked of developers by Samsung’s Bixby.

It is worth noting that Amazon expects most people to add Alexa Conversations initially to existing skills. Developers can simply leverage Alexa Conversations for particular interactions while keeping the rest of the skill as it is today.

Alexa Conversations is Seen as Strategic to Amazon

There are a number of reasons why this new approach is important to Amazon. First, Alexa Conversations has the potential to make complex interactions, such as purchase transactions, easier and more natural for users. They won’t have to follow some rigid process to provide the necessary information the retailer needs to complete the purchase accurately. It does mimic humanlike conversation in that users can go “off script” and the system will automatically adjust while still moving toward the end goal of gathering required information.

Second, it could make complex Alexa skills easier to develop. Some developers have shied away from complex interactions because of the effort required to build out large numbers of dialog variations. Alexa Conversations should enable developers to focus more on the core experience they want to deliver or the transaction they would like to enable.

There is a third reason that is generally left unsaid, but pushing more user experience over to Alexa will enable Amazon to provide more consistency in overall skill experience. Amazon has little control over third-party skills. If developers attempt to support more complex interactions but fail to adequately define needed dialog variations, users are likely to rate the experience lower. That reflects badly on the developer who may be less motivated to support the platform and can undermine user confidence in Alexa as a platform because few differentiate between first-party and third-party capabilities. However, Amazon doesn’t have much control over third-party efforts and is largely stuck with what they can deliver. Alexa Conversations, if widely adopted, could enable Amazon to raise the general level user experience across both first-party and third-party skills.

Finally, the more Alexa handles in the conversations in third-party skills, the more it will be able to assess usage on the system at scale. In essence, Alexa Conversations asks skill developers to provide more information to the system and handles more of the interactions. And it is automated in terms of data collection in a way that the certification process, for example, is not. This should enable Amazon to accelerate the accuracy of Alexa given the additional data.

Details about the Alexa Conversations Contest

Developers can sign up for the contest at Devpost. The skill must be submitted for review by Devpost by September 14, 2020 and winners will be announced in October. Ten finalists will each receive $5,000 and a grand prize winner will receive an additional $20,000. In addition, there are opportunities to win a number of $2,000 bonus prizes for categories ranging from productivity to audio, visual, and games. That should be enough incentive to get a lot of participation.

Amazon Lays Out Alexa’s Future, Adds 31 New Features for Voice App Developers at Alexa Live

Alexa for Apps Wants to Launch TikTok and Other Apps for You

Alexa Conversations to Automate Elements of Skill Building Using AI and Make User Experiences More Natural While Boosting Discovery

Amazon to Roll Out Cross Skill Goal Completion that Bundles Skills to Better Fulfill Alexa User Needs and is Not to Be Confused with Alexa Conversations