GenUI: Google and Claude Transform the User Interface

The Rise of Generative Interfaces: A Turning Point for AI

Discussions surrounding artificial intelligence have recently highlighted a fascinating development: generative interfaces, or genUI. In March 2024, Sarah Gibbons and a collaborator published one of the first formal definitions of this concept. The genUI is distinguished by its ability to generate real-time interfaces, customized for each user, unlike AI-assisted design, which helps designers create traditional interfaces more quickly. The main difference lies in the timing of the AI's intervention and the end beneficiary: AI-assisted design occurs during development, while genUI directly shapes the user experience.

Two years after this publication, genUI is no longer just a theory. Although there are many examples of ongoing experiments, the most significant advancements are found in AI chat interfaces. Simple interactive elements, such as buttons, form fields, and checkboxes, now appear contextually within conversations, generated by the AI when deemed useful.

Google and Checkbox Innovation

Take the example of a traveler using Google’s AI mode to search for hotels in London. Google’s AI mode offers several recommendations, each accompanied by thumbnail images of the hotels and, more interestingly, checkboxes next to each option. These checkboxes allow the user to directly select the hotels they are interested in, thereby simplifying the decision-making process.

The selected hotels then appear as "chips" in a bar located above the chat input, ready to be used in follow-up questions without the need to retype the information. This feature demonstrates how a simple reduction of friction in the user interface can represent a significant enhancement of the overall experience.

Buttons and Forms in Claude

In another example, Claude, another AI chat interface, sometimes generates a set of interactive form fields to gather additional information from the user before providing a response. This process is managed by a predefined module called AskUserQuestion, which Claude can decide to activate when additional information is needed.

For instance, Claude may ask the user for their budget for a trip to London. This question is part of a maximum set of four questions, illustrating how safeguards are in place to ensure a consistent user experience. Without these limits, a model could, for example, ask a user to respond to an endless series of questions.

Claude also uses a multiple-choice clarification widget to understand the user's accommodation preferences before making recommendations. The questions are tailored to fill in the gaps in the system's knowledge regarding the user's specific trip.

The Importance of Context in AI Interactions

Just like the checkboxes in Google’s AI mode, Claude’s AskUserQuestion module brings together simple yet effective design elements to gather the necessary details and advance the conversation. This has become an essential component of modern AI chat experiences.

It has never been realistic to expect users to become prompt experts, capable of exhaustively providing the context needed to achieve their goals. Most users, except for AI enthusiasts, use very simple prompts, such as "Help me plan a trip to London." Without additional information, the system can provide useful general information, but this approach leaves the potential for personalized responses untapped.

Thus, these follow-up questions are crucial, but without interactive genUI elements, they can become very tedious to provide. The integration of these interactive elements into modern AI interfaces is therefore essential for making interactions smoother and more relevant.