Casey Hudetz on Integrating New Generative AI Models

Alex Smith: Design Leader Insights is brought to you by Fuego UX. Fuego UX is a leading UX research, strategy, and design consultancy. Hey Casey, thanks so much for joining the show today. 

Casey Hudetz: Yeah, great to be here, Alex. 

Alex Smith: Absolutely, and as we get started, can you give the audience a bit of context into your journey in design? 

Casey Hudetz: Yeah, so about close to a decade ago, I got my master's degree in human computer interaction, jumped into the agency world. Was really interested in all emerging technologies, so I worked on AR projects and voice and some machine learning, but really got fascinated about the future of AI and how that would impact our work. So I gave some talks about it, and I made a short film about it, and then slowly, Chat GPT hit, and the world changed a year ago. And thankfully, being a DocuSign, I'm a design manager. I was at a company that was taking it seriously, so over the last year, I've been very engaged with my team of how we integrate these new generative AI models into the work we do. So the one part of the journey, then the other part is really asking the question of how does the actual design process itself, how is that transformed by AI or large language models potentially? And it seems like that's really where my two interests lie at the moment. I know we'll get more into it, but that is a very, very truncated version of how we got to this call.  

Alex Smith: Yeah no, it was awesome seeing you speak at UXStrat in Boulder. But yeah, let's dive into the second half of that, which is like, how, how should, or should it, how should AI fit into the design process today? What's your perspective on that? I think a lot of designers are like, wait, this is coming. What, what's coming? How, how should I use this? 

Casey Hudetz: Yeah, you know, I, there are so many think pieces right now about  how we. Everyone is going to have to re-skill or up-skill. Every executive talks about their plan of how they're going to have to do that. And I, I think it's, it's foolish if we think as designers that's not true for us. We don't exist outside of the knowledge work, even if ours is more in the creative arts sense. If one were to break down all of the discrete tasks that a designer has in order to build a human centered project, I feel like there are so many. That could go from the double diamond discovery to, you know, choosing the right problems, prototyping to testing and all the rest, all the rest. So if I think back to my agency days where I had to be more of a generalist and I would be on one project that's consumer packaged goods one week and then the next year on automotive and then the next year on life insurance. And on each one of those projects, you are responsible for different parts of the design process. Something that we, so let me just reflect on a few of the things I had to do and then how that could be done now. When we would do audits for information architecture, we would say how do all of our competitors talk about their offerings on their website? It would be an afternoon, half a day, I don't know, coding to all of them, filling out a spreadsheet, this is what their navigation is, et cetera, et cetera. Now that these GPT is hooked to the internet, I can say here are five websites, go to them and give me a table of all of their navigation items and it does it instantly. Let's say I have to create a survey. Okay, you generate a ton of survey results. You get a bunch of survey questions. You go and execute. But when you got that information back, none of the designers on my team were data scientists. And oftentimes we'd have to wait to get resources from other people in the agency to help us make sense of it. With these new data analysis tools, which I've been experimenting with, you can put in a CSV and then just ask questions about it. What does this mean for this date? What does this mean? Can you make a chart that proves this? Can you help me understand correlations between X and Y? So if you think of not only survey data, but A B testing and how quickly You, you can get information that can help you make better decisions to look at it another way. There have been these breakthroughs now in multi modal multi modal generative AI. And what that means is you can put more than just text into it. In particular, you can put images into it. So I've been experimenting a lot with that. And if I think back on my career. And I would have to go and run a design thinking workshop. And at the end of it, you've got all these stickies and you've got all these big whiteboards and things. I would then have to go back and like on the plane I'm transcribing it and trying to make sense of it. With one photograph now you can transcribe all this and pull out the major themes that you're hearing. And it could do it in an instant. Also with that multimodal. None of these are perfect. They all require some sort of, you know, second pair of eyes and that's why if we think of them as interns, we're guiding them and we're checking their work and making sure it's right. But if you have something like that ongoing as this co part co pilot, like in Figma or something that can constantly give you MIDI feedback as you go that you can take or accept, like how can that not help us improve and move more effectively and just have Yeah, have, have better outputs and outcomes.

Alex Smith: Let's talk about the UX of AI right now. It feels clunky now, but what you just said is so true. This is the worst it'll ever be in terms of probably UX. So where do you see, I see that as a major Achilles heel right now. Like the, the actual experience is not that intuitive. And there's also a trust factor where people are getting some results from ChatCTP where they're like, that's, you know, that's not a real legal case or whatever. Whatever the source is, right? So let's talk about how that experience could evolve. 

Casey Hudetz: I see the evolution happening in two ways. So right now we go to, let's say, we'll just keep using Chat GPT. We go to Chat GPT as a separate place. We put in these questions. Mine can be literally like define ARR. You know, something as simple like that, it goes into the biggest best model in the world, and it gives me an answer. And then I take that back into my life and I figure it out. Where we're headed now, I see it in two ways. One is it's all going to be baked natively into everything we do. So if you look at Microsoft 365, they're slowly rolling out this co pilot into every product that they have. So you could say, take this spreadsheet and make it a PowerPoint and then give me these images to help make my point and refine this language X, Y, Z. So it's not going to be, you're going somewhere. It's, it's going to be baked in and the usage there. will be highly dependent on the user experience. How great is it? How easy is it? How discoverable? How usable? So I think that is something that every, every product company will have to grapple with is A do we need to bring these tools into what we offer? And then B how do we make them as easy as possible? That's something my team is doing at DocuSign. What DICE tools do, how can they solve a problem, and then how do we make them as easy to use? But then the second piece that I think will help us all is we've got this, this gigantic model in GPT 4. We're going to get smaller and hyper specialized models that are just good for legal contracts, that are just good for medical, that are just good for transcription like we've got here. Which will make them more reliable, they'll make them faster and, and way cheaper for companies to use and build upon, which will, I think, lead to a bigger proliferation.

Alex Smith: What advice do you have for new designers entering the field today? 

Casey Hudetz: Yeah, I would say two things at this moment. One is the same as I would have told when I got out of grad school. A while ago and then one I think is unique to now. The first I would say is, and I would, this is evergreen, is to focus on the human aspect of the job. And by that I mean, what does it mean to communicate a research report very clearly? That is a skill that we should develop. How do we facilitate an actual fruitful workshop that then gives us things to work on? How do we build coalitions internally so that people take design seriously and keep us in the room to have a discussion? So I don't always like the term like work on your soft skills, cause that feels so unquantifiable and how do you know if you're doing it or not? But I really think the things that make us, what makes it more beneficial are our communication skills, our ability to strategize at a high level, our ability to run effective collaborative workshops and the rest. I think those things are timeless and I think that should always be capitalized on. The second piece though, and I'm, I know I'm just a broken record here, but I think the newer ones to the, to the industry, I think have to know the tools better than anyone if they want to get their foot in that door. Because those that have been around 20 years, if they pick up the tools, not only do they have the tools, but then they also have the expertise and the judgment to know when to use them. So I, I think, I, I think that those that are new to the industry should just really, Understand the design process and then really think about how the tools can make them stand out in terms of their efficiency, in terms of their effectiveness, and then really lean into the people skills necessary to, to bring it about.

‍Alex Smith: Makes a ton of sense. Casey, thanks so much for coming on the show today. 

Casey Hudetz: Yeah, hopefully it's helpful in our lives.

Alex Smith: For sure.