Conversational UX - The AI Revolution Continues ...

Will the real AI revolution be Conversational UX?

The surge of AI media attention has me convinced we are occupying unusual real estate on the famous Gartner “Hype Cycle”.? We seem to find ourselves both in the “Peak of Inflated Expectations” and the “Plateau of Productivity” simultaneously. ? The emergence of LLM’s has surely disrupted content generation, but its virtues are being extrapolated (by some) in ways that are a little far-fetched.? Media outlets will have us believe that AI is about to reach the fabled, sci-fi singularity and that we’re soon to be extinct.

While I’m not qualified to answer the question of how close we are to an AI apocalypse (though if you care, I’m firmly in the camp of “unworried”),? I do think I may be qualified to venture a guess at another big disruption for LLM’s.? One I don’t see folks talking about enough yet.? The end of the classic user interface.

In the beginning …

I grew up in the dawn of the Internet, and marveled as we built new and novel ways of presenting data and user controls for web and desktop software.? We “oohed” and “ahhed” at things like Tree Views, Multi-select Drop-Downs, Image Galleries, and the“Infinite Scroll” of sites like Pinterest.? The evolution of UI controls increasingly made giving users access to large volumes of data feasible.

When mature search and filtering capabilities came along, Search quickly earned a prominent place in the UI of nearly any and all sites.? While traditional ways of drilling through volumes of data remained as a fallback, users adopted search with gusto as the fastest way to get to the information , product, or service they needed.? Over time, as search got more predictive and refinable, its prominence grew.? Now you’d be hard pressed to find any retailer that doesn’t feature search as the first and most prominent call to action on its site.

Conversational UX - The next disruption

It feels imminent to me that the next major disruption in UX is going to come from LLM’s.? If the modest Search box disrupted UX in the 90’s and 00’s, the LLM-powered Chat box is surely poised to do so in the mid 20’s.? But why??

Traditional Chatbots are great at providing information for users to enable them to complete a software task.? If we ask a Chatbot to help us with our electronic Tax filing, or where to find the routing number for our bank, it can give answers (video, text, etc.) or navigate us to the proper part of the site to complete the task.? LLM’s are surely making these models even more reliable and useful, ensuring that when they provide help it is increasingly accurate.

But that’s not the disruption.? The disruption comes when that Bot can actually take the action for me.? Imagine, for example, visiting your favorite retailer and your first action is to say “buy two boxes of Ethiopian coffee with at least 4 stars and deliverable by Monday”.? A screen pops to confirm , you click “Yes” and you’re done. ? Or if you’re an i-Ready customer (the products I work on), imagine asking the system to “Assign the best lesson on finding main idea to everyone in my classroom who has scored higher than 80% on the last quiz”.? That’s the power of Conversational UX.? I don’t need to know anything about how your site is designed to do what I need to do , I just need to express it in plain English and confirm it.?

Sounds interesting, but is this really a game changer?

Hearing this you might wonder how much this differs from existing voice assistants like Siri, Alexa, etc.? You’d be be right to see some similarities, but the breakthrough that powers this is very real.

Today , to build something like an Alexa app (known as a “skill”) a developer maps user intents to actions.? This is what connects what you say to Alexa to the proper code/API call when you say it.? Example :

This means, in general, that everything an assistant like Alexa can do within your app has to be determined up front, coded for, tuned, and deployed. ? It is just another form of UX, albeit one with your voice.

The gamechanger for Conversational UX is this:? there is no longer any need for a product owner, designer, and engineer to anticipate and build a map from “intent” to “action”.? The AI does this for us.? The mapping from intent to action comes “free” when we train the LLM with our API documentation.? In essence, we train the LLM to know how to translate human language to our web application’s API, we then perpetually do so such that all new features are available via chat from day one, with no extra effort.

Is this really possible?? How would it work?

Microsoft is definitely moving in this direction with the Office Co-Pilots, providing not just a super-charged version of the fated “Clippy”, but letting it take actions for your “Make me a pivot table with …” as well.? The concept seems sound but the real test will be how effective we are at getting an LLM to be effective working with the smaller samples of input (text questions and requests) and output (software commands, likely API calls) that our applications have.? Unlike Google, Amazon, etc. most of us won’t have access to TB’s of training data so the models will need to be able to get good on smaller samples of data and have plenty of safeguards so a user doesn’t inadvertently cause havoc.

So what’s next?

I predict it won’t be long before you’re on Amazon, TicketMaster, DoorDash, or any other major site and find yourself able to ask an AI to do things you would’ve clicked, swiped, or searched for in the past.? In the meantime, smaller organizations (like mine, CA) will experiment with Conversational UX ourselves while, I suspect, 3rd parties begin to build simple offerings that make doing so much simpler.? With luck, making your site “Conversational UX” ready will become something that can be done with a short integration project with your LLM of choice, and some common programming and documentation techniques that let the LLM figure out how to map intent to your site’s actions.

Before long, I think we’ll all be asking our favorite sites and applications to do things for us that previously would’ve taken time to type, click, and swipe our way to in the past.? With any luck, we can use the time we save to prepare for that AI apocalypse.