The Invisible IA of Conversational AI
Conversational AI poses an interesting challenge to the traditional way we think about information architecture.
Take Cleo, for example. Cleo is an “intelligent assistant for your money”. Its closest equivalent is probably something like Mint. Once you connect your bank account, Cleo allows you to review your spending, manage your budgets and set savings goals (among other things).
The app uses Facebook Messenger as a base; users interact with it by asking ‘Cleo’ questions, much like you would with a real person.
Apps like Cleo make the information architecture invisible to the user, in a way. Much like Voice User Interfaces (VUI), there is no visible hierarchy, taxonomy or navigation to give the user a structure within which to find the information that they seek. The interplay of ontology, taxonomy and choreography seems to happen “behind the scenes”.
Let’s imagine you want to find out how much you spent on Uber’s last month.
On a GUI based platform, you would log-in and land on your dashboard. From here, you would most likely look for a section, button or link labelled ‘Transaction History’ or something similar. You would then scan the page for some sort of search bar and type in “Uber”. You would then see that you have results from the past 6 months, but you only want to see last month, so you look around the page for some type of filtering tool, labelled something to do with dates or date ranges. You select your dates, click on the filter button and voila — behold all of your Uber transactions from the last month, with individual amounts, tallied into a total at the bottom.
You completed this task by relying on the interplay of ontology, taxonomy and choreography. You navigated from point A-to-B by reading the signs, the waypoints, the map (e.g. labels).
Let’s do the same with Cleo.
You open up the Cleo app, and you type in: “How much did I spend on Uber last month?”. About one second later, you get a reply: “You’ve spent $104 on 9 Uber trips”.
Simple as that. No ‘navigation’. Rather, you provided Cleo with the destination (“how much did I spend”) and the waypoints (“Uber”, “last month”) and Cleo got you from point A-to-B instantly.
It’s like asking where something is in a foreign country and some kind chap gives you directions AND gives you a lift there.
But, what if you don’t have a map and you don’t speak the language?
Here’s where there’s still some work to be done. Although there have been massive developments in AI in the past few years, it’s still not perfect. Getting conversational AI to understand what you really want tends to be an iterative process of trial an error.
A lot of interactions with conversational UIs still feel something like this:
You walk into a library and there are no shelves, no signs, no books anywhere. There’s just a front desk with a friendly looking librarian. You go to the front desk and tell the librarian you’re looking for “Book A, by Author Z”.
The librarian at the front desk disappears into the back room for a bit and reemerges with one of three answers for you:
- “Here it is.” (he hands you Book A, by Author Z). Success.
- “I don’t know what that is.” (he stares at you blankly). Fail.
- “Here it is.” (he hands you Book M, by Author K and it’s totally not what you asked for, but he smiles like he expects you to be on your merry way anyway). Definitely a fail.
While there are more and more conversational AIs capable of dealing with scenarios 2 and 3, usually the user ends up having to ask the same question in several different ways before getting the answer they seek.
With no map to find your waypoints and no one to understand you, you’ll have a hard time getting from point A to B.
I won’t pretend to understand all the complexities behind conversational AI/UI, but I look forward to learning more. I’m curious to see how conversational AI will continue to challenge the way we think about Information Architecture. Kudos to the conversational AI designers out there, already making waves and challenging the status quo with complex data models, content models and interdependencies.
References:
- Designing Conversational UI with Information Architecture (4 part series from Chatbotsmagazine.com)
- The future of Information Architecture: the Voice User Interface
