Conversation is the Most Human Interface

Keyboard and mouse was never meant to be the final boss of human-computer interfaces.

How did ChatGPT become the fastest-growing app ever?

Compared to most tools, it’s remarkably easy to start using an AI chatbot. This is a big reason why ChatGPT reached 100 million monthly active users faster than any other app in history. ChatGPT reached this milestone in just 2 months, dwarfing the initial growth trajectory of products like Instagram and Tik Tok (I’m not counting Instagram’s Threads, which essentially copy/pasted Instagram’s user base).

A primary factor in the rapid rise of these products is their intuitive interface that lets users interact with them using a method that they already do every day: talking. AI chatbots bridge human and computer language like no computer before them.

Letting humans talk to computers is not a new idea. For decades, computer scientists theorized about this capability before it ever existed. In 1960, the computer scientist J.C.R. Licklider wrote about “man-computer symbiosis”, or the potential for computers to augment human skills, intelligence, and productivity. In Licklider’s view, “the basic dissimilarity between human languages and computer languages may be the most serious obstacle to true symbiosis.” In other words, while he envisioned computers rapidly acquiring the capabilities and computing power needed to help humans with complex tasks, it was the human/computer interface that would hold us from using computers to its full potential. For Licklider and his colleagues who spent hours translating natural-language instructions into 0’s and 1’s for computers to read, the ability to type (or speak) a normal sentence into a computer and have it respond with the same would have been the stuff of dreams.

AI chatbots combine two major innovations in one

An oversimplified model of an AI chatbot is that it has two components, which tech people often call the “frontend” and “backend”:

a problem-solving and content-creation engine operating behind the scenes
a chat user interface, which is what users see and interact with

The best-designed products tend to have user interfaces that obscure the technical complexity happening behind the scenes, making advanced technology accessible to everyday users.

I talk about them as distinct because language recognition (a human’s ability to type a regular sentence into a box and have a computer understand it) and language generation (a computer’s ability to generate new, unique sentences) are distinct tasks. A computer can do one without the other.

For example, imagine a horoscope-generating bot where you select your name, birthday, and gender from dropdowns, and the bot then generates a unique horoscope for you. In this example, the bot can still dynamically generate unique content on command, like a chatbot. But it only accepts inputs in the form of values selected from dropdowns, not natural language prompts like “Write a horoscope for me”.

Both the backend and frontend of AI chatbots would be considered significant innovations even if they were introduced on their own, and both have been crucial to the growth of these products so far. Whether they incorporate AI or not, the best-designed products tend to have user interfaces that obscure the technical complexity happening behind the scenes, making advanced technology accessible to everyday users. We may even come to see the current AI chatbot interfaces as too simple. Chatbots like ChatGPT are increasingly incorporating dropdowns, forms, and other dynamic UI elements as they reckon with the challenge of teaching users how to use a totally new type of product.

Backend vs. frontend

The backend content-generation engine is what puts the generative in generative AI, and is the part of AI that captures' most of the public’s attention. This is understandable, since it is the part of the system where decisions about content generation, data handling, privacy, and potential biases are made. As a result, this part of the system carries more of the ethical baggage and concerns about economic and social impact. In addition, this is the part that AI skeptics are usually referring to when they point out the limits of the technology or argue that it is overhyped.

With a more rigid user experience, ChatGPT would probably not have grown as quickly as it did.

As a product designer, though, I’m more interested in that second innovation: the user experience (UX) and user interface (UI) of how the human and computer talk to each other. Designers and product managers know that no matter how innovative a product might be behind the scenes, if it’s too confusing to get started, the product probably won’t get very popular. For example, imagine if OpenAI had launched a version that required users to enter prompts in a specific format or syntax, like a coding language. With a more rigid user experience, ChatGPT would probably not have grown as quickly as it did.

Hard to master, easy to learn

Sure, it takes some time to get used to the idea that you can just talk to it like you talk to a person. It also takes practice to learn the techniques that can make it really useful. But ultimately, those techniques are about optimization, not getting started and learning the basics. You can get pretty far just by talking, and without any special training or understanding of how it works behind the scenes.

There are real barriers to mass adoption of chatbots, but I don’t think that difficulty getting started is a big one. As a result, chatbots have a brilliant on-ramp that most products would kill for.