Why chatbots fail
The novelty of AI does not make up for a bad UX
In the tech industry, we want the next big thing to be here so badly that we jump very quickly to pronouncing its arrival. Driven partly by genuine enthusiasm, partly by inbuilt capitalist incentives, we generate wave after wave of hype. And there is a special flavor of this zeal reserved for technologies that realize that pinnacle of sci-fi fantasy: human-like artificial intelligence.
One problem with the term “artificial intelligence” is that it gets tossed around so carelessly. The current AI chatbot narrative centers around the use of natural language processing (NLP), for example, but Google Search has been using NLP in its search results for a long time. AI is a marketing term used to generate hype, and the tech media is buying right into it.
This is not to say that a broad range of “AI technologies” won’t create user value. But as product designers we must be able to distinguish that value from the sales pitch. It wasn’t too long ago that Microsoft’s Cortana, Amazon’s Alexa, and the Google Assistant were ramping up to change the world. The way people talked back then, it seemed inevitable that these digital assistants would become copilots in our lives, unburdening us of its tedium.
Big tech companies have claimed victory, but five years on the trend is clear: although useful for certain things, the robots have not changed society in any meaningful way. This low adoption isn’t a result of pushback from luddite activists, restrictive government regulation, or fears about evil machines. The issue is one of inertia — there simply aren’t that many reasons to go out of your way to use the thing, as it turns out.
Changing human behavior is no trivial feat. There is both an academic discipline and a self-help industry dedicated to understanding its nuances. Sometimes tech companies strike gold and create products compelling enough to work their way into our everyday lives — personal computers, smartphones, and social media — but generally speaking, it takes a lot for people to change their habits. In order for a new technology to catch on, it needs to hit a specific note at just the right pitch, at just the right time. Billions of dollars have been spent trying to figure out how to hit this note consistently, but it remains elusive.
The industry has had some time now to test and iterate on chatbots, and there are now some excellent design guidelines out there. Most of these guidelines acknowledge the limitations of chatbots as well as their strengths. There are a few key UX issues with conversational interfaces that make it hard to support the claim that chatbots like ChatGPT or Bard are the next big thing in technology.
Text only
With a chat interface, the text is presented in a linear way. Imagine holding a pencil in your hand and trying to discern its color. If you’re staring down the nose of a pencil, it’s going to be difficult. If you can move the pencil around and look at it from various angles, it’s much easier.
Using a chatbot is like staring down the nose of a pencil. When you ask it a question, and get a response in the form of text, you’re getting one piece of information at a time. Instead of typing a query into a search box and getting a list of answers in the form of text and other UI elements like buttons and tabs, you get one text response that may or may not be helpful.
This is why I’m highly skeptical about Bing’s new ChatGPT-powered search. When we’re searching for something on the internet, there’s not necessarily one right answer. ChatGPT sometimes returns a list of results, which helps a bit. But a simple ordered list still lacks interactivity.
Over the years, even Google Search has evolved away from lists to a much more robust UI with tabs, expandable cards, and summarizing elements known as knowledge panels. Thesevinteractive elements put more control in the user’s hands, allowing them to find the answer amidst the results, rather than relying on a chatbot to hit the nail on the head with a conversational response.
Ambiguous input
Another issue with chatbots is that you never know exactly what to say in order to get the response you’re looking for. This is why if you’ve ever used a digital assistant it will first offer up suggestions about things it can do. Without those prompts, users have no idea how to engage with it.
Admittedly, ChatGPT helps solve this problem by having an incredibly sophisticated model that can respond very naturalistically to most user-generated messages. But the fact remains: if you’re looking for a nuanced response to a complex answer, it’s frustrating to go back-and-forth with a chatbot until you hit on what you’re looking for. Much better to just get all the information in an instant and quickly find for yourself what’s most relevant.
Graphical user interfaces were actually invented to solve this exact problem. In the early days of computing, you needed command line interfaces to execute functions. When the people at Xerox PARC invented GUI, they empowered an entire population of non-programmers to use computers by recreating familiar models (e.g. files, folders) on a screen.
Novelty isn’t enough
The main value of a chatbot or any AI is its novelty factor. People love to play around with the thing — it’s cool! But if the coolness is the main reason people are using it, that’s a problem for the product’s longevity.
Calling something “artificially intelligent” is hubristic. And the public’s response to this will be to immediately test the limits of this supposedly intelligent thing, hoping to break it. (Need I remind you of the infamous Tay bot?)
We’ve already seen journalists generate negative news cycles for ChatGPT by doing this. Aside from the harmful rhetoric unfettered chatbots could potentially disseminate, it’s probably not good for business if a main way people are engaging with your product is by trying to prove it doesn’t work very well.
ChatGPT is being heralded as a breakthrough technology, but chatbots are in fact a stale idea. Our attachment to the fantasy of a sentient AI is what makes this a big news story, while the user experience is glossed over and the hype train keeps rolling. We need to break this mold to unlock technology that’s actually going to provide value to users. What excites me personally is not artificial intelligence (whatever that means), but what it might enable. This requires a process of trial and error based on user needs and behavior, not hype. Strong human-centered thinking can steer technology in the right direction – and that’s where the design community comes in.
If you’re someone who likes to read a bunch of semi-useful information in text form, might I suggest simply subscribing to my Substack where I write about a range of tech and design topics. I promise you will never be darkly manipulated into ending your marriage. 😊
