Embracing AI: 5 Principles for the UX Frontier
AI won’t take your job (*yet), but it sure makes your job a whole lot easier.
“AI is taking over.” We’ve all heard it. The takeover is inevitable. It’s coming for our jobs– destroying the design and tech industries. It dominates news headlines in 2023– sandwiched between articles on Taylor Swift’s love life and Russia’s occupation of Ukraine. As inevitable as this takeover may seem– the days of Skynet and the Matrix– they may not happen within our lifetime, if at all.
After spending two months as a humble graduate student researching and experimenting with Generative AI tools for UX design, I can confidently say there’s no need to stock up the apocalypse bunker. At the end of the day, AI is a tool. It is a tool that is only as good as the user operating it and the data behind it. This seems to be the reality. It could also be a way for me to blissfully cope with the fact that I am transitioning to a career where robot takeover is imminent. Who is to say?
With that said, there is no disputing the unprecedented power and possibilities generative AI offers. Regardless of if it spells out the extinction of UX as we know it, it is important for UX professionals to understand how to integrate these technologies into their everyday workflow, if anything from an efficiency and up-skill standpoint.
So here are 5 principles I have found most important during my pilgrimage exploring the uncharted wilds of generative AI:
1. Don’t Use Generative AI to Reinvent the Wheel
If you know exactly what your project needs, don’t use generative AI.
If you already know exactly what your want make — whether it be a persona, wireframe, style guide, UI component, or product summary — it might be faster to craft it by hand. Generative AI is great for rapid ideation, but if you or your business partners know the granular details of what you want, put that UX-know-how to good use and make it yourself.
Say I want to generate the name of a book called Fellowship of the Rings. If I ask ChatGPT to generate a list of book names (conveniently using the full description of the actual book by Tolkien pulled from Wikipedia) it will give me everything but “Fellowship of the Rings.” If that is the exact title that I want, I could have saved time but just typing out the title at the top of my hypothetical-but-most-likely-plagiarized fantasy novel.
This is, of course an anecdotal example, let’s take a look at an example more closely related to UX:
If I am generating icons within an already existing design system (in the above case, Material Designs), it is probably best to construct those icons on my own. As you can see from the above example from the Magician AI tool, it doesn’t matter how specific I get, I cannot reliably consistently match the styling of Material Designs, it just looks a bit off. The styling, level of detail, and stroke weight just don’t fit. Generative AI is highly efficient at producing content, but if you already have said content in mind- why bother having ChatGPT reinvent the wheel?
2. It Helps to Know What You’re Talking About
When engineering prompts, paint a detailed picture.
In other words, be specific. Of course you don’t have to be an omniscient sage in your designated product or industry, but it certainly helps to know what you’re talking about.
Asking Whimsical AI to develop a user flow diagram for an online audit software for accountant without any industry specific terms or details within the prompt results in… a pretty generic user flow. Formatting issues aside (we’ll get to that in a minute) there is much left to be desired here:
This is where you want to leverage your user research, interviews, and stakeholder meetings to include some of those little details that assist gen AI in building a more robust and realistic output — especially if you’re working within internal systems or documentation that is not readily available online. I am far from an accountant, but throwing in some terminology (that I ironically pulled from Chat GPT) to spice up my user flow diagram vastly improved the output.
Now, rather than produce a generic user flow, Whimsical, built something that — at least on the surface — seems a bit more well thought out and context specific. If you’re going for an output that is more high-level, then this shouldn’t be a problem! However, if you want generative AI to be specific, you also have to be specific. Generative AI isn’t trying to feed a family of little robots at home. It isn’t going to preemptively impress you with highly detailed reports in hopes of garnering a promotion. It is a tool and, as such, you’ll get out of it exactly what you put in.
3. Generative AI is a Side Dish, not the Main Course
Use generative AI in sparing steps, don’t let it dominate your entire workflow.
Once again, generative AI is only a tool. It is a highly powerful, efficient, and unprecedented tool but it is still just a tool at the end of the day. Don’t use ChatGPT or its counterparts to be the primary force driving your UX process. Generative AI is best used as a small part in the system, and it shouldn’t become the entire system itself. Using Whimsical again, here is what happens when I ask it to build me a food ordering app using the following prompt:
If this is my starting point for mapping out a user flow then great! But in a world where this tool completely replaces mapping out a user flow diagram entirely, the ‘final result’ is a bit lackluster, even with a more detailed prompt. Proceeding from this flow to wire framing and subsequent development would probably result in a half-baked product with a lot of missing context and features.
Throughout my time using Gen AI, I have found the process chunking tasks to be a highly beneficial on improving the overall quality of the final deliverable. Rather than just have Whimsical generate an entire diagram to use, I use the first prompt to generate smaller flows and interactions to compile into a larger, more detailed user flow diagram.
It might seem quicker to generate this user flow — like the first prompt — all in one go and move on to the next step in the design process. However, moving to the next step and passing this user flow on to other designers or product managers would probably result in lengthy questions in email chains and meetings to clarify additional or missing details. Allowing AI to be the central driver in your UX process may complicate and cost more time for your product team. Using it as a part of the process — integrating it with your own work and research — will save you a tremendous amount of time, energy, and clarity.
4. Always Add A Human Touch
Use generative AI in sparing steps, don’t let it dominate your entire workflow.
This might seem like an obvious one, but it is still worth mentioning: always review whatever generative AI outputs. It is not an infallible tool that some of the aforementioned doomsday headlines make it out to be — at least not yet. It lacks context, and is only as good as the prompts and data it has to work with. It can still make mistakes. Take a look at some of these user personas I generated using ChatGPT:
There are two things worth mentioning here. Firstly, all students listed in the program attend public research institutions. There is no mention of smaller private schools or commuter schools. Secondly, even more concerning, Michigan doesn’t actually have mountains to climb. Using these user personas would be problematic for additional R&D later on down the line. However, an easy fix is to 1) ensure we include a student persona from another universities in the area and 2) rework any section that describes users who using outdoor climbing trails in Michigan, focusing more on climbing gyms.
Here’s another example using Wireframe Designer Plug-in within Figma:
At a glance, the low fidelity model looks great, but when you stop to look at the page there are a few things that need to be fixed. A landing page probably doesn’t need a back button, there’s nothing to go back to — so I deleted it. The platform heading might not be the most important piece of information to include, so I swapped it with the game title. It is important to give the user some visibility of system status between pages, so I added a bit of styling to indicate the user is on the “recently played” tab.
Making these adjustments, adding additional context, value, and some high fidelity changes makes the final result negate some of those mistakes Wireframe Designer made on the first iteration and properly showcase the efficiency and power of Wireframe Designer. What Wireframe Designer does really, along with most other AI tools, it start the UX process. I didn’t need to spend hours adjusting the spacings or constructing auto layout frames within the design, but there were still some things that needed adjustments — that added human touch. Because, at the end of the day UX is about human experience and, as such, sometimes requires that extra human touch.
5. Gen AI’s Lesson on Insanity
Keep trying with generative AI, you might get different results. Try everything!
Albert Einstein allegedly once said, “Insanity is doing the same thing over and over again and expecting different results.” If only he could’ve seen generative AI. Oddly enough with most generative AI platforms doing the something over again usually yields different results and sometimes those results are difficult to replicated.
There are two sides to this. On the one hand, it makes it difficult to replicate an exact output (See principle number 1). However, it also encourages users to explore the system and try out different prompts; see what works, what doesn’t, what is missing. Is there a method to ChatGPT’s madness? Or is it truly just luck of the draw?
As you can see, I inputted the exact same prompt 3 different times. All three times yielded different outputs. By inputting the same prompt multiple times I discovered a few patterns: (1) All outputted names are a concatenation of Sushi and a second word. (2) All outputted names follow the same structure of a name followed by a brief slogan. I can use this in later prompts to isolate variables I want to keep, and those I want to change.
Generative AI adds a sprinkle of magic to the user experience mix. Sure, it’s got its quirks and limitations, like not always understanding our human contexts or nailing that creative spark. But, it’s also a game-changer.
Think about it: this tech can whip up interfaces faster than you can say “sushi roll.” Need to generate content? Bam, it’s done. Want to predict user behavior? Easy peasy. And let’s not forget about automating those repetitive tasks.
Generative AI and design is something I am still figuring out. Nevertheless, I can say with certainty that, in order to take full advantage of this tech, you need to continue to try things out; reuse prompts, change minute details, find different use cases — even if they lead to failures or dead ends.
For all the UX professionals who have read through until this point, don’t think of yourself as the last dinosaurs waiting for a robotic meteor to destroy your livelihood. Think of yourself as one of the lucky few pioneer — budding experts — in generative AI and UX. Just do what all UXers do best; stay curious and keep asking questions.