Generative Design: AI-Driven UX Paradigm Shift
Today I had the pleasure of attending the STRAT conference in Amsterdam, where there were many fascinating talks about AI in the current space of product design and how it will evolve. However, there was one that stood out the most out of all of them, one from the Design Manager Kay Hofmeester from META Reality Labs, who was also previously in the original design team behind Alexa from Amazon. He shared a new framework used internally at META currently only used for secret prototypes. We are talking about Probabilistic Design, which completely shifts the way GUI and product will work for end users and are created for.
Before we start let me break it down for you into digestible chunks of how we going to talk about it. I am going to talk about the origin of probabilistic design, what is deterministic design, what is probabilistic design, the language of generative design, the role of UX-er going forward,
What is Deterministic Design?
The current digital landscape of products that we use is 99% deterministic, which means there was a design team tasked to design a certain experience for the most common denominator of the target group, and it works. However with many products going to millions of users, and sometimes billions like Instagram, having the most common denominator no longer is viable, and adding more target groups with many functions to the apps makes them even more bloated and harder to use for everyone. I recently saw a thread on Twitter(X) about how user interfaces over the last decade got more complicated and buried down the simplicity with infinity scrolls and a lot of dopamine-inducing features. But what if I just want to search for something or the system knows I am usually going to the office at 8 am and just shows me UI for that instead of all the possible options.
What is Generative Design?
When it comes to probabilistic design, for simplicity here will be called Generative, it's more focused on the context-based creation of a layout exactly to your personal needs and use cases. Let’s not misinterpret it to the one from Rabbit R1 as it wasn’t generated every time on the fly for you, but concept-wise was close enough. In Generative Design, each user interface would slightly differ from one another and would be based on the design system set by designers with rules on how to position elements based on the defined logic. In deterministic design, as we usually create it would otherwise be impossible with the number of users and personality types. However, with the help of LLMs, it all becomes much more possible, using information stored locally on your device to determine what layout for a given situation would work best for you and your mental model of interacting with the device.
How does generative design operate?
The way design would be generated would depend on the input from the user, context, and general knowledge of the LLM, and the interaction itself would depend on a few types listed above. By no means is its final language of how generative design will operate, however, it gives an idea of how it would gather data and use it to communicate with use tailoring experiences to our needs on the fly.
When it comes to input and output it would be chosen based on the best path for the user so sometimes the interface would be audio only, but sometimes there would be no response at all, it would just know for example when you come back home from the work at 6 pm during winter time it would turn on the lights and pre-heat home based on the entry in your calendar and schedule. That’s why the magic is not only personalization but automation at many layers of our daily lives.
The most interesting and useful part is proactivity, which means apart from reacting to your commands and giving you data, it would also suggest certain things based on your situation or just do it for you without asking because it would know that in this scenario like meeting with you clients, it would just take notes for you without needing your confirmation.
Deterministic Design vs Generative Design
Which one is better you might ask, well there are a few circumstances where one is better over another, so let’s dive deep.
Things to consider about Deterministic design:
- Lower initial resources: cumulative costs for every additional template
- Cheaper to operate: LLM require costly tokens to run on the cloud, however the prices as well as methods of operating change drastically
- More control: only if we prefer a lowest common denominator design
- Lower latency: generation of every time custom UI might be resource-heavy be it token-wise or battery-wise for mobile device
Things to consider about Generative design:
- Personalized UI: tailored to your unique personality
- Contextualized UI: appropriate for each situation
- Unlimited layouts: enabling flexible responses
- Cross-app scenarios: enabling complex tasks and reducing the overhead of switching from app to app for given tasks
The role of the Designer and impact of GenUX
The designers as we know it will change drastically already in the coming year, from manual execution to curation of experiences for all of us. So that’s why being a designer never was, and in the coming future even more so will be more about what you can strategically bring on the project rather than raw knowledge of layouts or Figma. That’s why, if you’re not the best at making beautiful mockups, but have great critical thinking skills, project management, and great organizational abilities then you’re at the right place. That doesn’t mean visual skills won’t matter anymore, because the experiences will continue to be branded with the predefined design systems that will continue to be designed by humans. Also, certain experiences will be hard-coded because not every scenario will require a hyper-personalized experience tailored to your psyche and life.
So what you can do going forward?
- Explore more new tools and approaches
- Learn about generative design and its principles
- Experiment the new ways of interaction design beyond websites and apps
- Actively observe the changes in our discipline landscape with excitement and curiosity instead of imminent doom and gloom!
If you enjoyed this article, don’t forget to smash that “clap” button, and follow for more stories like that, and share this story with friends! 😎
*Due to NDA agreements between the organizers of the conference and Meta Inc. all rights are reserved to the original author of the principles. I am not allowed to share original slides, names, specific prompts, and examples of specific internal implementations.