Designing Bots for 2020

In the previous instalments of the series, I talked about how early-stage startups can leverage messaging bot to find their product-market fit. I also covered the dilemma of whether bots are here to stay or disappear, and why we still don’t know the answer.

This time, I approach the question of what needs to happen if bots are to become game-changing technology like the iPhone. I’m going to bring up the vision of the future we’re betting on at Ada, which is an AI-driven real estate renting platform.

Thanks to bots, for the first time ever, we might see seamless integrations between services at scale. By seamless integrations I mean apps exchanging data with other apps in order to build user experiences that couldn’t happen if each of the apps worked alone.

Imagine ordering an Uber and asking it to drive you to your Airbnb apartment, because Uber bot can easily talk to Airbnb bot and exchange information.

Building self-defining interfaces…

Both voice and text are universal interfaces. I prefer to call them self-defining.

Unlike the designers of mobile apps who sketch out every interaction in advance, engineers who program bots don’t have to design any predefined screens and interfaces. Just talk to the bot and find out what it can do.

This is how Steve Jobs pitched using a keyboard on the iPhone back in 2007: “If you don’t need it, it’s not there.” Now imagine what happens when you don’t even need a screen.

Screen-based interfaces are less rigid than hardware interfaces, but they’re still quite rigid. If Uber were to integrate with Airbnb, both apps would need two matching interaction flows and present the results to the user on the screen. Two advanced bots would simply explain what’s happening to the user in two self-generated sentences. That’s seamless.

…and self-defining interactions

Self-defining interfaces can present any type of data they received by simply talking about it to the user.

Yes, explaining a dataset using words isn’t always the most effective way of presenting it. Displaying a chart, for instance, works better than talking about the numbers organized in it. Customers, however, have repeatedly shown willingness to tolerate solutions that are less powerful if they’re more convenient or cheaper.

When we have an interface that can present any rudimentary dataset, the way to achieve seamless integrations is self-defining interactions between multiple bots. And Viv is already doing something like this.

The secret sauce behind Viv is a solution called dynamic program generation. Generating dynamic programs allows the AI-powered assistant to understand intent and generate a program itself to best answer the query, integrating multiple third-party services along the way.

Right now, apps can only connect to each other if their development teams manually match the data structures generated by the API behind each app. Text, on the other hand, is universal. A bot that understands grammar and intent can automatically extract all the information it needs from a response generated by another bot.

Take a look at the conversation between two AIs, Clara and Amy, personal assistants that schedule meetings. Two users wanted to meet for coffee, so they decided to let their AIs work it out—and it worked.

That, of course, is a fortunate use case. Both Amy and Clara work within the same domain and are still supervised by humans.

What AI engineers need to figure out next is how to build bots from different domains that are able to communicate with each other.

Now imagine what will happen if technology progresses to the point when every bot can do something like this. It’ll be a world of services seamlessly talking to each other without predefined screens and rigid API endpoints.

I think we’re getting closer every day.