Designers: having an impact in the era of AI
Explore how modern UX and Product Designers will harness their superpowers to innovate in the evolving landscape of AI and design.
AI will take your job
If you believe AI is the defining technology of our era, read on. If not, this article probably isn’t for you.
I want to tell you, definitively, one designer to another: AI is going to take our jobs, and why that’s a good thing.
As you’re tired of hearing already, generative AI’s like ChatGPT, Google’s Gemini, and Midjourney are changing our lives and the way we work at an accelerating rate. As the foundational models that power these tools become exponentially more vast and complex, new use cases emerge that were never before possible, for better and for worse. History shows us that this is true for any new technology that reaches the tipping point of widespread adoption, and in turn this new way of interacting with the world brings about new needs and opportunities to be addressed. The internet and the mobile revolution are the first examples that spring to mind.
We’re designers: the people who care the most about bridging the gap between technology and people. When either one of those undergoes a revolution—technology or people—so does our profession.
In the same way that mobile changed the way that we think about UI, and when web designer stopped meaning “the person who writes HTML”; technological revolutions transform the way we work as designers, and the work that we do. To that effect, I propose that the internet also took our jobs. So did desktop publishing, and the printing press before that. The same is equally present for cultural revolutions, look at how important equity and inclusion are to our work today, as is accessibility, and how the advent of social media has influenced design through marketing to security.
AI is the next technological revolution that will have a profound impact on culture and society, and that fact means it will take (over) our jobs as we know them.
So, what do we do about it?
The role of Design in the AI era
Each revolution ushers in new interaction paradigms for designers to learn and new use cases to design for. Transitioning from clicks to taps and from printing to sharing are modern adaptations that we’re mostly all familiar with, but I think these specifics aren’t important.
Empathy, creativity, and curiosity will always be the core of what makes designers valuable.
We don’t need to do anything fundamentally different to stay relevant in the AI era as long as we keep our focus on people as creators of culture and as the users of the technology.
As people integrate more AI into their day-to-day lives, we’ll build an understanding of their use cases over time, and we’ll use the same tools to find creative ways to solve users’ problems as they arise. That’s how design has remained relevant until now, and how it will continue to be relevant in the future. Just stay curious and open minded, and keep bridging the gap.
However, there are already some emergent problems arising from today’s state of AI use that could really use your help. Let me give you some examples of problems that users of generative AI face today, how design can be impactful, and where the trends look to be moving.
How design can steer the progress of AI today
Interaction trends are shifting towards chat and conversational interfaces, sometimes referred to as the Assistant model. Large Language Models like ChatGPT are so powerful (and Google is so bloated with useless SEO-stuffed results) that they’ve already completely transformed how we access information. However, the size and complexity of LLMs come with major drawbacks that designers can have a significant impact on solving without the need to understand the math behind the models.
1. Latency
A response from generative AI is slower than we’re used to getting from a machine. The output is derived from a number of possibilities that’s almost impossible to truly grasp, and with language models specifically, the output is produced word-by-word (technically chunks smaller than words are produced, called tokens). While this might be acceptable when you’re waiting for an image of A Still of Kermit the Frog in Total Recall (1990) to generate, it’s far less tolerable when you’re waiting for the model to regenerate an entire 2-page document just to update a minor factual error somewhere in the middle.
Waiting time is more impactful when viewed as a human perception than an objective truth, which is why designers have been so influential in solving this problem in the past. There are many stories of designers reducing perceived waiting times (“psychological time”) with smart solutions and this is a great opportunity area for designers to define a new set of practices to solve this now and into the future.
2. Hallucinations
When generative AI makes factual errors or veers off-topic, we call these hallucinations. This issue will decrease over time as we implement better self-checks into the models, but inevitably, it will never go away entirely.
We don’t have a best practice for letting users recover from hallucinations yet. We’re not even close. We need designers to figure it out. We don’t even know how to make users aware of the potential for hallucinations, so when they encounter a hallucination for the first, second, tenth, or hundredth time, it’s a huge erosion of trust. It gets people into big trouble.
How do we design error reporting that allows users to flag hallucinations and improve the model? What level of up-front communication do we need to give users about the possibility of factual inaccuracies? And perhaps most of all: how do all of these things change as new interaction paradigms (conversational) become the norm? Who’s going to stay on top of the human element? You are.
3. Bias and fairness
Machine Learning models of all types are biased by the data they’re built around. In the era of generative AI, that’s content from the internet, with its own biases and distortions baked in. With generative AI the bias is obvious because we’re interacting with the outputs directly, but it is still a very present and impactful problem in hidden AI applications.
The current solutions for fighting bias for generative AI are manual processes like prompt engineering and data cleaning. For example, both Google Gemini and DALL-E 3 have representation requirements present in their image post-prompts. If you ask Gemini to generate a photo of an astronaut, your prompt is reinterpreted and the model is told to randomly select the cultural representation of your character to ensure fair cultural and identity representation across the board. The intentions are good, and this is a fair way to correct for some of the cultural biases inherent in our society, but it fails horribly when your aim is factual accuracy.
Is that really the best we can do? There should be more designers in the room when we’re making decisions about post-prompting, or when cleaning data and adding and removing parameters in training data to fight bias and improve fairness. There’s two-way traffic on the bridge between technology and people, and designers are the people in the room who understand people best.
Conclusion
In the face of the AI revolution, we’re not losing our value as designers; we’re finding new ways to make a difference in the same way we always have. History has shown us time and time again, from the printing press to the internet, that designers adapt and thrive by retaining the core of what makes our profession valuable. AI’s emergence is another chapter in this story and offers us an amazing opportunity to use our empathy, creativity, and curiosity to help shape the future.
The new challenges that AI will bring aren’t just problems. They’re chances for us to innovate and work together with scientists and engineers to guide AI towards being a more human-friendly, fair, and inclusive contribution to society and the human experience, and we need designers to be engaged in these issues sooner rather than later.
With AI’s potential to have such world-transforming impact, our role as designers is more crucial now than ever.
