Real-World GenAI Applications You Can Use in Your Research Today

GenAI's capabilities continue to evolve in new and surprising ways. But for people who conduct user research to use these tools to the best of their advantage, we need to evolve our thought processes and approaches as well.

In this conversation, Nathan Reiff and Yaron Cohen go into innovative and tactical ways to leverage GenAI today, the difference between savvy AI users and citizen AI developers, and two deep-dive use cases you can apply to your own research.

Let’s dive in!

Nathan Reiff is a UX Researcher at Dscout focused on the development and deployment of AI features.

Yaron Cohen is a Lead Design Researcher working for a major international bank.

Jump to…

• A crash course on AI
• How UXR could shape AI development
• How UXR can benefit from AI augmentation
  • Harvesting knowledge around previous research
  • Designing test studies and research activities
  • Synthesizing and analyzing data
• The difference between AI savvy users versus citizen AI developers
• Deep dive use cases with strategic foresight and storyboarding
  • Use case 1: Strategic foresight
  • Use case 2: Storyboarding
• Final pieces of wisdom around responsible AI usage
• Conclusion

Nathan: Can you give us a little crash course on AI?

Yaron: There are two main types of AI. There's predictive AI, which has been around for quite some time, and there's the younger version, Generative AI—or GenAI in short.

To understand predictive AI use cases, think of any recommendation engine on platforms like Spotify or Amazon that gives you suggestions for songs or products you might like. That is a common way organizations utilize this tool.

Usually the algorithms power tidbits of insights, but they don't necessarily create anything novel. That's where GenAI enters the picture, it's much more exciting.

GenAI is an algorithmic creation of novel content such as text, images, and music. Good examples are ChatGPT, Midjourney, and DALL-E.

What's common to both types of AI is the fact that the creation is done through pattern identification in large data sets that essentially enable the creation of other insights or novel content.

What we have to remember is that both types can be applied in several different ways in UX research and there are ways to leverage both.

Nathan: Now that we have a baseline understanding of what we mean when we say AI, let's start talking about AI as it relates to UX research.

Nathan: Here at Dscout, we see a lot of opportunity in both AI's ability to help researchers execute research, as well as researchers’ ability to aid AI development. Most of today's chat will be about how AI can help researchers execute research.

Let’s touch on how research can help aid AI development. At Dscout, we're approaching AI development with a research mindset, versus the typical move-at-a-breakneck-speed tech approach.

Our AI team is infusing every step of our AI development process with both technical research and UX research to ensure that our AI features will truly help teams run faster, [conduct] better research, and [stay] in the spirit of centering our users, their needs, and our development process.

We're placing high importance on compliance with zero data retention policies with third-party services, and making sure that all AI-generated content is clearly labeled and available on an opt-in basis.

Nathan: I'm curious to hear if you have any additional thoughts on how UXR should shape AI development.

Yaron: It's exciting to see these developments happening on your end. I'd like to first of all mention a book that I liked by Ben Shneiderman on Human-Centered AI.

Some of the principles that he mentioned are essential to make AI trustworthy from a user standpoint.

The first principle is transparency, which means from a user standpoint, I want to know how the AI system got to the decision that they decided to make.

UXRs can help mimic this in prototypes and tests with users and their understanding of—not just the output that they see—but also the steps listed by the system on how it makes decisions, and how that makes them feel from a trust standpoint.

The second way mentioned in the book is that UXRs can help validate accountability. The idea is to create some sort of audit trail for the operators and the developers of the AI system to be able to troubleshoot if something goes wrong.

You can mimic this in a prototype and test it with a group of people who have these kinds of roles, understand their comprehension of the audit trails, and also their level of trust in what they see. If something goes wrong, they are the one who needs to know why.

The third principle is about privacy. UX researchers can help test If our users can read and understand privacy policies and their right to opt in and out of data collection.

The last thing that I'd like to touch upon is bias. Everybody wants to know that the system that they work with is not biased, or has a very minimal bias. It's the organization's responsibility to do testing on this and create guard rails against biases.

From a UXR perspective, once you get some good, explainable systems that tell you the limits of the system and what it can and cannot do, then you can test it from a user standpoint and see that your users can understand all of this.

One last thing I’d like to mention, if you want to make AI solutions even better and more useful, UXRs can further help by understanding the persona that's more likely to benefit from AI solutions.

Based on what I've done in one of my previous positions, I've learned that there are personas who would like to outsource their responsibilities to an AI-based solution. The more of an expert you become, you want to take some control over the output and collaborate with the AI.

This is where you can, as a UX researcher, step in and understand the job to be done by the different personas and understand, once you start mapping the customer journey, at what point a user would like to take control from the AI-based solution, and what kind of parameters they would like to adjust themselves. At the end of the day, AI-based solutions are tools, and everyone wants and needs to use them slightly differently.

You can use this output generated by research activities to develop better solutions. These can help anybody who designs for AI know where they need to improve their work.

Can you give us a high-level overview of how you think UX researchers can benefit from AI augmentation?

Yaron: To understand the big question of how UXRs can leverage AI, we have to think about the holistic research journey and the different stages where AI can be useful to a UX researcher.

Most of us start our work by talking to stakeholders and doing some requirement gathering. At this stage, it's better to focus on the human element and develop your soft skills of talking to humans because here AI might not be so useful.

✔ Harvesting knowledge around previous research

Once you understand the problem you're going to be working on, this is where you can start engaging AI in a knowledge harvest, and there are a few solutions that you can already leverage out in the market.

For example, ScholarAI is essentially a customized ChatGPT that helps you understand trends and recent research that was done in academia about the topic that you're curious about.

This way, you can identify knowledge gaps and understand where you, as a researcher, can start to dig in a little bit more. Several organizations already have research repositories in place.

In the future, we're more likely to see some GenAI solutions that can be plugged straight into your org's repository and help you start understanding what researchers before you worked on, and how you can leverage this research. This way you don’t have to reinvent the wheel.

✔ Designing test studies and research activities

Moving to the next step, you’ll dig into designing the test studies and research activities. This is where AI can give you a divergent thinking muscle. I know that UX researchers are usually focused on the problem space—on understanding what the problem is and defining it—not so much on divergent thinking, ideation, and brainstorming.

This is where, especially if you work alone and you don't necessarily have a thought partner, AI can help you come up with ideas. This is where GenAI can be leveraged to come up with a more interesting research plan.

✔ Synthesizing and analyzing data

Another area where this could be interesting is the area of analysis. In the future, once we have safe and reliable systems, we can start using GenAI to synthesize raw data to understand what's happening and even leverage other data sources. When it comes to analysis and synthesis, both predictive AI and GenAI can be useful.

As a researcher who has worked with a system that leveraged predictive AI back in the day, we did some of the manual work ourselves and then ran it through an algorithm that gave us some clusters of thoughts. That was cool to work with and interesting to see. In the future, we're probably going to see more of this coming to our reality.

When it comes to the last step, which is usually to share the results of our research presentation, I don't think that there are very good solutions right now, but I think in the future, we might see something along the lines of an AI coach that can help give you recommendations on how to improve your presentation and cater it to different audiences.

Can you speak to your idea of AI-savvy users versus citizen AI developers?

Yaron: First of all, we have to clarify these terms and understand what they mean. An AI-savvy user I would consider somebody who knows how to work with AI to produce a better output, rather than working completely alone.

A Citizen AI developer is a step up from this. It's somebody who understands AI to some degree and understands how to build a customized application.

To become an AI-savvy user, you have to have good communication skills because you have to do a lot of prompting, at least with today's tools. You have to be able to communicate to the AI what you want it to do for you and you have to communicate to your stakeholders what you've done with the AI.

To get to the point where you're a Citizen AI developer, you have to have a few more skills. I would recommend having a basic understanding of how AI works, at least at the conceptual level, because you want to understand things like the editor at ChatGPT that enables you to essentially build the application.

You don't have to understand coding. You do have to understand prompting at a higher level. You also have to have some good critical thinking skills, because you're going to get output from your GenAI tools that sometimes is not the best output or what you wanted right away.

You need to be able to understand how to evaluate this and what you need to do next to get to the place where you get a good output. Nowadays, GenAI is still pretty young, and because of this, some of the tools that we have in the market right now are susceptible to errors and hallucinations.

I aimed for use cases where we can use AI almost as an extra muscle to get inspiration, but not necessarily very accurate answers. I would refrain from using it for these kinds of cases.

Can you chat through your deep-dive use cases that you've explored using AI for?

Yaron: I'm going to talk about two use cases where I try to envision a future where some of us become Citizen AI developers and we leverage them. The large language models (LLMs) do some things for us.

Use case 1: Strategic foresight

The first use cases are about strategic foresight. Strategic foresight is a set of research methods that are used to consider possible and plausible future scenarios. It's usually part of the family of generative research and done at the beginning of the process. Researchers can use it to build scenarios and narratives based on signals that can be found in different places.

The reason why this could be very helpful is that if you work on any strategic research project that’s in the realm of service design, systemic design, and strategy design, you can use these scenarios to understand the reality that you're designing a solution for.

The solution can be a strategy, a user interface, or even a service blueprint. The most important idea is to understand the future possible and plausible scenarios, use them as a starting point, and then start designing for the desired future reality.

The main challenge is that it's a very specialized niche of research methods. It's not a very intuitive way of thinking for most folks. Even if you're trained in this, we have a lot of human biases. We're not wired to think about the future in certain ways. This is where AI tools like ChatGPT can help a lot and give you extra muscle to teach you how to think this way.

ChatGPT Editor is where you build a customized ChatGPT for any use case. To get started, you have to think about the steps you want a user to go through once they use your application. To do this, I started breaking down the scenario that I wanted to use into high-level steps.

I had to describe to ChatGPT what I wanted it to do. Let's say we find a signal—that means a piece of information—could be news or could be something in an academic article, but it's usually a piece of information about a phenomenon that is a bit out of the ordinary.

It's not mainstream media, but the idea is to go from this to building a scenario about what happens if this little signal becomes a reality. The first step is to write the signal in a few words. Sometimes you can even add a link to the article, then you choose the type of scenario that you want ChatGPT to generate for you or several scenarios.

Then you're getting an output. To make the scenario a bit more rich, you can ask ChatGPT to develop a narrative around it. Then you can use this narrative to do an ideation session or group discussion. It's usually one of these boundary objects that you can take into a workshop and then work with your stakeholders, clients, and coworkers on developing solutions for it.

My idea was to use the customized ChatGPT to generate them for me. So on the right-hand side, you would see the graphical representation of what I'm going to touch upon in a second scenario. The arrow that goes up is an extrapolation of trends that continues with minimal disruption. Whatever we find, whatever we read about becomes the norm. It grows over time.

Scenario number two is a transformation scenario where society or systems fundamentally change or reorganize around a new paradigm. It could be a new technology coming up or maybe just a new social norm and then, everything changes accordingly.

Scenario number three is a scenario of constraint, which is a core guiding value or purpose organizing society and governance behavior. This usually happens when a government passes a new law when a professional organization decides to regulate the entrance to a profession, and then you have this sort of constraint.

Scenario number four is about collapse, which is a rapid catastrophic system and infrastructure breakdown, usually just a decline of what we've seen before. So how can ChatGPT help us as researchers understand these scenarios a little bit better?

Looking through the lens of we're a user using one of the signals that I mentioned. So, Science Alert, this is by all means, not mainstream media. I selected an interesting signal that's related to the field of environmental impact. We're disrupting the Earth’s salt cycle through activities such as mining and construction, and this poses a significant risk to freshwater.

So if you think about it, this is not something that is talked about very often compared to other environmental impacts. So if I wanted to understand, okay, what are the implications of this? I tried to run this by the application that I built.

The customized ChatGPT is called a foresight facilitator. As you can see, I typed the link and gave a very short description of this. My application is specifically for Canada, which is where I live. I wanted to understand the implications something like this could have on life in Canada a few years from now.

The first thing that happens is ChatGPT asks you if you want to generate any of these scenarios that we touched upon, it gives you a better description of what that means in terms of life in Canada. We are going to see an example of what a growth scenario looks like.

It's basically how Canada might adapt or benefit from this situation. I could also ask ChatGPT to look into different scenarios, but I wanted to focus on one.

In this scenario, Canada experiences significant advancements in environmental technology and policy making and the government and private sector collaborate to develop innovative solutions for maintaining balanced salt levels in water bodies.

These solutions not only prevent damage to freshwater ecosystems but also foster economic growth. All positive things, if you think about it from a high-level perspective. That gives me a few potential areas to explore.

But what if I wanted to make this something a bit more interesting for a workshop, or something that can spark an interesting conversation, debate, or an interesting ideation solution?

This is where the next step—which is creating narratives—comes into place. We can start understanding what the future could look like 10 years from now. What people are excited about when they hear about all these things, while the green tech sector booms.

There are more international investments in talent and more courses in universities and all of this is great, but there are also some worries about the rapid pace of industrial growth in the green sector, potentially leading to new forms of environmental impacts. Economic disparities might arise favoring certain regions where this industry is located.

This is a way to understand and leverage the process of working together with AI—understanding how researchers can collaborate with it in an area that requires creative thinking rather than accurate answers.

As you can see, some of the scenarios and the narratives are pretty generic. This is why I mentioned right from the beginning, that inspiration is the name of the game.

You want to ideally use something like this or at least the output you're getting from ChatGPT as a starting point before you continue to dig further and consolidate the scenarios with more knowledge.

It's cool though to be able to get some help in the thinking process and understand how to go from theory to actual applications of this field called strategic foresight.

In the future, we're probably going to see more options to connect these tools to curated data sources about signals, about the research that UX researchers do within our organizations. All of this can become signals that can feed into the creation of scenarios and narratives like this. We're probably going to see more options to do more interesting things. So this is case number one.

Use case 2: Storyboarding

The other scenario that's more straightforward is the case of storyboarding. As you can see, the image itself is generated with DALL-E because I wanted to depict the idea of AI helping us to come up with a storyboard.

Storyboarding is pivotal in bringing narratives to life. It allows us to almost live through situations and ideate solutions to address what we see. As UX researchers, we could leverage the power of storyboarding when we run workshops.

Co-creation, ideation, or brainstorming research activities that require some sort of visual artifacts, boundary objects, or visual articulations of ideas to then spark ideation, debates, conversations, you name it. The main challenge is that once again, a tool can help us here.

Some of us do not necessarily come from visual art backgrounds. Because of this, we have limited graphic design skills and limited sketching skills. Some of us don't use storyboards because of this. In the world of 2024, I'm also mindful of the fact that the budget to hire dedicated resources is limited.

Sometimes people skip it altogether. What I wanted to try was leveraging the power of AI to create a storyboard for us. Once we see how easy it is, we start leveraging it more in our research.

I used ChatGPT Editor and I took advantage of DALL-E, the engine that creates images that are embedded within the ChatGPT4, and I tried to think about the steps that the user would have to go through to generate a storyboard.

The first thing is to think about the big picture storyline, the theme, what is the experience that we're talking about here. So I have to describe this. Then you have to start thinking about the individual steps, almost like the customer journey that we want to describe visually in images.

We have to decide how many steps we would like to show in the storyboards. The steps and the little images are called vignettes. The next thing is to decide on the look and the feel and select a visual style that resonates with the storyline. The storyline [hones in on] users’ perspectives and then designs the vignettes by articulating the vision.

What do we want to happen in each scene from an artistic standpoint, understand, let's say, the angle we want to show the people, how many people we want to have in each scene, things like that.

I wanted to think about a real use case, something that's becoming more and more common in intermodal travel. So intermodal travel. We're going between point A and point B, and you're doing one leg of the trip by train and bus, and then the other one by airplane. So two types of transportation modes and the interconnection between airplanes and trains are becoming very common in Europe.

There are some good cases in France, Switzerland, and Austria that I know about. I'm sure in other places too. Not so much in North America. It's a mental model. It's still a little foreign to most of us. I wanted to understand—how would an experience like this look from a user perspective?

If I needed to design a solution for this, how would I describe this whole thing in a storyboard? I developed a customized ChatGPT to help me do this. I ran this use case through the application. On the left-hand side, you can see when I started to create the baseline image.

I wanted to create a story about a young student who goes to school in a small university town in Italy. She has to travel back home to another European country for the holidays. She starts her journey at a small train station where she has to go to a major airport and then she has to fly to another country.

All these steps have to appear here on the storyboard. The baseline image depicts the opening scene of getting into the train station. As you can see, the text is a little bit off. This is one of the main problems that I identified right away with some of the solutions that exist nowadays to generate images. They do not always depict small details very well.

Once I understood it, I knew what I needed to do to improve the image, even though I liked this sketch-like style. I had to tweak it a little bit and ask the AI to add more people to the train station’s space.

So as you can see, we're going into the train station. We have to check in, and then this is a great place in an ideation session to ask, what do we want to happen here? Do we want people to check in their luggage and collect it at the airport, or only after they get to their final destination? Is the boarding pass going to be applicable for both the train and the plane journey? Is this something that’s going to be enabled by an app? An email?

These kinds of questions can come up then of course the actual journey by train and plane. Then there's landing at the final destination, picking up luggage, and leaving the airport.

To be able to generate the storyboard, you have to do some final tweaks. You have to bring all the individual images to one canvas. It can be something as simple as PowerPoint or something more advanced that you can create using a graphic design tool. In my case, I used Affinity Designer, but the main thing to remember is that the tweaks that you usually have to do are minimalistic. You don't have to go into graphic design or anything like that.

The main conclusion points to keep in mind when it comes to using AI tools for storyboarding are:

• Use your imagination, because you have to come up with the storyline and breakpoints within the story. You can always ask ChatGPT or another GenAI to help you further develop the narrative if you struggle to do it.
• Learn how to prompt or build prompts that are related to visual language, such as the kind of style you would like to see in each vignette. There are websites such as Dallelist that can help you with this.

Some people like to bring wireframes into storyboards. DALL-E, Midjourney, and this family of tools are not so good for wireframe creation. It's better to use tools like DALL-E for the storyboard and then other tools such as Galileo AI for wireframes.

What final wisdom can you give us as it relates to risk mitigation and responsible AI usage?

Yaron: The key is to know the limits of the AI system that you are working with or developing in your company. Understand what it can generate, and also make sure that these limits are clear to your users—then they can make the most out of the solution.

Wrapping it up

As GenAI continues to develop and evolve, it's important to take a critical look at where and how the tools source their information. Take a good look at how you can use GenAI to gather knowledge about previous studies, design research activities, and even synthesize and analyze data.

Use cases like strategic foresight and storyboarding can really help your team flesh out more complex scenarios with a lower lift. As always, it's important to understand the limits of its capabilities—and keep a humanistic perspective in mind.

You may also like…

• GenAI Adds Tension to UXR—But Can Also Bring Opportunity
• How to Train GenAI to Work as Your Personal Research Assistant
• 15 Tips for Navigating GenAI as a UXR