The idea of a conversational interface is one that has always fascinated me, long before AI. From conversational forms to chatbots, I have been writing about this subject since 2012.
However, interesting though they were, they came with a lot of limitations until recently. The combination of large language models and the incredible advances in natural sounding, emotive and fast text-to-speech makes it nearly impossible to distinguish between speaking with an app or a person.
If you haven’t already I would encourage you to check out the latest improvements with OpenAI to see just how natural a conversational interface can be.
An Exciting Time in Conversational Interfaces
All of these advances have left me very excited and I am not alone. I am working with several clients to supplement or even replace existing user interfaces with chat and voice alternatives.
After all, as is argued in “The Best Interface is No Interface,” most of us would be quite content to look less at our screens as long as we can be just as productive.
They also have the potential to be considerably more accessible, personal and intuitive than a screen full of fields and buttons. Conversations are the most natural of communication tools available to us and yet until recently we haven’t be able to “chat” with our tech.
There Are Still Challenges
Of course, conversational interfaces are not always the best solution. Verbal conversations in particular are not always appropriate. I do not anticipate a world where we are all wandering around having the equivalent of public phone calls with our tech.
There are less obvious challenges too. A conversation is not always the fastest way of completing a task. For example, it is often quicker to enter credit card details into a form (especially if they have been saved by the browser) than to do it as part of a conversation.
Then, there is what these models are trained upon. An ‘out of the box’ large language model will not have all of the required knowledge to do a good job and probably won’t adopt an appropriate tone of voice in all circumstances without training.
Finally there is the elephant in the room, AI tends to hallucinate and can be manipulated. At best this can be abused, such as the man who convinced a chevy dealer bot to sell him a Chevy Tahoe for a $1. However, at worst it can be dangerous.
For example, I am working on a chatbot that could be used on a mental health charity website. You don’t want an AI giving advice to somebody who is suicidal!
You Need To Pay Attention To Conversational Interfaces
I realize this newsletter has been a bit of a ramble with no point. However, I have written it because I want to draw your attention to what could be a transformational change in how we interact with our technology. It has the possibility to be as profound a change for the web as mobile devices were.
As a result it falls to all of us who are advising our clients and organizations to understand the impact these technologies will have on the user experience. This is not a technology that can be just rolled out by IT with no consideration of how users will interact with it. Just because it doesn’t have a graphic user interface, doesn’t mean it is not part of the user experience.
