The Unique Challenges of UX Research for AI Experiences
Based on our experience, we’ll be writing a series of articles on UX research for AI experiences for UXmatters. In our series, we’ll dive into some core aspects of how to adjust our approach to conducting UX research for AI experiences and consider what it means to be a UX researcher studying AI experiences.
We won’t focus on AI tools that might help you do your job as a UX researcher or how AI could impact the usefulness of humans working in User Experience. There are many smart people and great resources out there covering these topics. Instead, we want to fill a knowledge gap that exists, regarding something that has become apparent to us from working in this space, but is not well understood or often discussed: Conducting UX research for AI experiences is different from conducting UX research for non-AI experiences, especially GenAI experiences.
If you didn’t know there is a difference—let alone a big enough difference for us to write a series of articles about this topic—don’t worry. Through this series, you can learn about these differences. Plus, adjusting to them requires something that UX researchers usually have in spades: empathy and user-centric views.
Understanding AI: Classical Versus Generative
What’s the root cause of the differences between classical AI versus generative AI? Many of these differences exist because of the AI technologies themselves in their current state. We expect these differences to evolve over time as the technologies mature. Although it may seem like AI came out of nowhere with ChatGPT, some have called AI the “oldest new technology.”
This is an oversimplified view of how we think about these differences:
- Classical AI—Trained on a lot of data—though often specific, limited data sets—this type of AI is excellent at identifying patterns. Think of this as similar to completing a jigsaw puzzle: classical AI can take all the puzzle pieces, forecast the best next piece placement, and assemble all the pieces into a known piece of art. This type of AI includes machine-learning (ML) classification models, and early natural-language processing (NLP). We’ve all likely used this technology through predictive text when texting and writing email messages, allocating certain email messages to spam, or even limited chatbot experiences.
- Generative AI—Trained on truly massive amounts of data, this type of AI is set apart from classical AI in that it can generate new content that mimics human creativity. Instead of assembling a jigsaw puzzle into a known piece of art, it can create something novel. However, GenAI still learns and generates based on the data on which it was trained—just as humans experiences and environments influence them. While there are various GenAI models—for example, large language models (LLMs)—one of the biggest shifts is this technology’a ability to mimic human dialogue. It can create a conversational experience using more natural language and greater expression and can create text, images, videos, and audio. Some examples of GenAI include ChatGPT, Gemini, and Midjourney, and there many more tools that leverage this capability—for example, the way Google and Amazon now generate AI summaries of search results or product reviews. Because GenAI is much more accessible and applicable to numerous use cases, its potential is causing a lot of excitement!
We’ve simplified these descriptions for the sake of brevity, but if you want to learn more about them, there are many good resources available online. We have included some links to them at the end of this article, in the section “Some Useful AI Resources.”
Different use cases require different solutions, which might be rules-based solutions, classical AI, GenAI, or a combination of these. The key is to use the correct technology to solve a real problem, not to assume that GenAI is the right solution. Implementing AI doesn’t automatically address users’ needs. Our UX expertise plays a crucial role in ensuring that the solution is the right fit for users.
