AI + UX: Product Design for Intelligent Experiences

Hi. I'm super excited to be here. I'm always part nervous, part very excited to talk about design. Since I'm a product designer, I'm going to start with some problem scoping. I just want to start by saying that I don't consider myself an expert in AI and design. I don't know who is at this stage of the industry, but I'm a collector of interesting ideas about design and I add a layer of personal experience on how I interpret those and then communicate them.

And then secondly, typically I do this talk in 40 minutes, one hour, but I've cut it down to 20 minutes, so, if it feels a bit rushed, just bear with me. It's going to be like looking at a 1.2 speed video, but, yes, I just want to make as many points as possible, so thank you for bearing with me.

Now, my interest in AI started at UiPath which is a company that does robotic process automation. I worked on a product called Clipboard AI, which actually won Time Magazine Best Invention of 2023 award, and it got me excited about the industry and the new problem space and the challenges and opportunity it's presenting us with, and then I joined Miro, and now I'm designing for the collaboration space, which is very exciting for a designer. I'm basically designing for myself and, yes, this is the background.

And now let's see what AI is doing right now. It's losing steam. It's losing hype. It's less exciting. There aren't as many valuable use cases as expected. Companies are kind of losing trust. They were investing a lot of money in these technologies and they're not yet seeing a return on investments, and so there's generally, if we look at the Gartner's hype cycle, we're entering the era where the peak of inflated expectations has ended and people are generally in this trough of disillusionment around what AI can do for us.

But what's interesting when you look at this graph is that sure, we're going to say AI is doing nothing as promised and it's just infant technology, super immature, but it's going to get there, and we're going to see finally standards and best practices and frameworks that we can reference as designers, as builders, and then it's going to stabilize, so it's here to stay. Let's see how we can design for it since it's going to be with us from this moment on.

To frame my thinking around how we should design for AI, I'm starting from the definition of what AI is. And of course, there are multiple ways in which you can define AI. One way to think about it is as a collection of tools that we now have available, and if you think about these tools that we're using to shape products, to shape experiences, to shape the world, it kind of makes me be a bit poetic or philosophical and think that if we're now shaping these tools, we're essentially shaping ourselves, or at least our future, the future of human experience, and so essentially how we will work and live. So this is a quote by John Culking, we shape our tools, thereafter our tools shape us. What we design ends up designing us. If you think about the world in general, the furniture we design, the architecture, cities, workplaces, products, experiences, conversations, everything we design contributes to the way we experience the world. And so it essentially shapes us. Which is exciting, but it's also a call to responsibility in the age of AI where everything is ambiguous, not regulated, there are no clear standards or rules, guidelines, guardrails. So we have to be even more responsible than before because the way we set up this foundation for these technologies will essentially shape who we become as humanity.

Now, if we want to get that right, we have to start from the challenges, at least this is now putting my designer hat on and from the design perspective. We start we should start from the challenges that AI brings for designers. One of them is working with bias. We all know it by now, AI systems are inherently biased, which is not because they're evil, they're just a reflection of something that unfortunately exists in society, but that's also an opportunity, right? So if we want to change the way these systems mirror us and reflect the world we live in, we can feed them more representative data, we can interject that process and kind of change the way they reflect us and it's actually an opportunity. Safety and security, that's not just about AGI becoming sentient and killing us all, but it's also about data privacy concerns, even the psychological safety that we have when we're having conversations with AI systems or interacting with them in other mediums. Yeah, there's also the aspect of these systems being easily deceptive and so they can be manipulated in the wrong hands. Then there's the aspect of designing for probability, which is my favorite one. As a designer, it's really interesting to think that in conventional system, you control the experience. You know what's going to happen because you decided, because you get to design it. So you always know what's going to happen next or a couple of options of what can happen next when a user takes an action. With AI systems, you have no idea what's going to happen. Maybe it's going to be good.

Maybe it's going to be bad. You have not a lot of tools or heuristics to judge whether what happened is the right thing that should have happened. Is the output what the user was looking for? So that's why we see these systems rely a lot on thumbs up, thumbs down mechanisms to kind of, is this good? Did we do a good job? But yeah, as a design industry, we have to build frameworks that accommodate this lack of predictability.

The aspect of trust and transparency, people can trust something that they have no idea how it works and so there's always this, even in recent neuro studies, we've learned that people feel the need, the urge to triple check something that AI outputs so it's really interesting to see that we can trust where this information is coming from but there are also opportunities for that. Like this concept of demonstrated thinking. Show what goes on under the hood or explain a high level path that the AI system took to get to a particular answer, citations, references and so on.

And on the topic of ownership and intellectual property, I think this deserves a conference in itself but I just want to share something, a sidetrack, which is I believe that in the age of AI as designers, as a society, we will essentially shift what we value and we will move from valuing art because it's scarce, art is definitely not scarce anymore, it's its last problem today, but because it's human made. So I like to give the example of Mona Lisa. If you look at Mona Lisa created by AI, and you know it's created by AI, especially since now we have these labels made with AI. Okay, Mona Lisa, but it's like AI made so whatever. But if you're in the Louvre and you see people frantically taking pictures and being emotional and they're in front of this work of art that we collectively agreed has value for humanity, it's important, we worship it together, then there's a different kind of experience of art. So we will probably shift as a society to admiring more what comes from the human experience.

But if you're in the Louvre and you see people frantically taking pictures and being emotional and they're in front of this work of art that we collectively agreed has value for humanity, it's important, we worship it together, then there's a different kind of experience of art. So we will probably shift as a society to admiring more what comes from the human experience.

The principles that we can, I think I'm not doing very well on time, the principles that we can use and apply and keep in the back of our mind as designers, but not just as designers, as anybody that's contributing to the systems in general, start with people. This is the biggest rule that designers have to consider whenever they do something. But especially in the age of AI, when we're all going to live in this hybrid model of constantly collaborating with our AI assistant, wherever that is surfaced, we have to make sure that the user is the one that has control. We have to make sure that the user is the one that we're designing for when we're creating this hybrid model of partnering, augmenting our life with AI. Fighting bias, give safety and control, mechanisms to report or feel safe or undo or just feel that the user is in a safe space. For everyone, I'm keen on this, it's not just about accessibility and inclusivity, but it's also about making sure that these technologies are affordable and that everyone can use them. Because otherwise, we're just deepening the disparities in society, and so the wealthy will have more power, and the underprivileged will be less competitive, and so these technologies should be used by anyone. Be accountable, admit when you're wrong, gracefully, classic UX rules, promote good every time you make a decision for your product, think about the harm that that decision could potentially do, and if you see it, then don't do it or reconsider it. Help people grow. This is an aspect of mutual learning that goes on in the system, so if the system is designed in a way that it empowers me to use it at its best potential, then it learns from I'm using it, and we're both doing better. So there's this feedback loop, and we have to think about it when we're designing. Build trust, expose what goes on under the hood, explainable AI, demonstrated thinking, and age nicely. Now, this means that we're essentially going to design for completely different experiences and interactions and mediums. We have started with batch processing. We're in the very long age of command-based interaction, so if I want to obtain a picture of a mountain at sunset, I would, until now, open Photoshop, one command, create a file, create a triangle, add texture, add fill, so one step over the other at my command, so I'm in control, I have the locus of control. I'm building, I'm getting closer to my goal. In this new paradigm for design, which is called intent-based outcome specification, we are essentially giving the computer our goal, it understands, it deconstructs the intent behind it, and then it makes decisions, and by that creating the output we are looking for so it completely reverses the internal locus of control, and it completely changes the way we should design this product. So what are experiences going to look like in this new world? All the same. I used to be, like, criticise what's going on, that all the products are essentially apparently copying each other, everybody's using sparkles, purple, but I believe that we do need a visual paradigm and a visual vocabulary for how these actions are surfaced, and we all know when we see the sparkles or purple, oh, this is AI, and that's a good thing because it makes things more transparent. And I'm also thinking quite a lot about what are the right interactions in like the surfaces, the right surfaces in which these capabilities should be surfaced, and most of the products everyone is scrambling around in the industry trying to figure out how to have AI in their product. This is called Pilot, and they chose the retrofitting of AI in a different surface that essentially is disconnected from where the user is working at, so that doesn't really feel like the best model. I'm a huge fan of meaningful AI, so AI surfaced that moment where the user actually needs it, like, for example, when you record a loom, it automatically generates chapters, you don't have to worry about the structure, it generates titles, they're pretty good. Same for art, oops, for how long has this been there? Yes, so, art browser, I don't know if you used it, but it generates AI-generated previews of what's on on a website when you go through the links. So I think this aspect of invisible AI, implicit AI, that essentially helps you with a task, it's not just, here's AI, we have AI in our product but it does something meaningful for you. That's the right path forward. How can AI enhance, sorry, I'm really running, I know, but I want to say all the things. How can AI enhance experiences? I found four main pillars in which AI can help us build better products. They can make them more relevant, which means AI can personalise the experience, it can have context awareness, understand what's the particular goal you have at a particular moment in time, and optimise for that, predict what that goal might be based on who you are.

Decluttered, we're going to see more and more apps essentially being replaced, like Grammarly the most recent example, iOS now has Grammarly's features in it, so I don't know what's going to happen with that product. FML apps, this is a concept that essentially means whenever you have a need, you have one interface where you communicate that need, and then there's a solution that gets created on the spot, and you can discard it and never use it again. So it's the concept of having one interface that solves for all your needs, and you don't need 100 apps, and so on. Some people are experimenting with moving all of it outside the digital space and more like tangible products, but that's not going very well.

Of course, AI can also help us make products more accessible. This is something that we should all be excited about when we think how AI is bad, it can also be good already. And then helpful, just explain itself better to people. Now, if you want to design for AI with AI, I don't know if there are many designers in the room, this is the more design-heavy part. We have to start by understanding what are the things that AI can do for us as designers, and what are the parts that it can't replace or shouldn't replace. So if we make that clear distinction, we can now understand how to leverage those parts that can be replaced, and augment, like, work for being better critical thinkers, and whatever AI will not replace any time soon. AI will not replace us for a couple of reasons. It can't figure out the problems to solve yet, so it's pretty hard for any AI system right now to grasp the ecosystem around the person, like this person's background, trauma, history, personality, family relationships, work, and everything. This is something that we are uniquely positioned to be able to connect and unpack. Design solutions require multidisciplinary efforts. That means that we're not just designers. We have to also employ psychology, interaction design, visual design, research, sociology, and whatever. So the idea is that you have to bring a lot of disciplines at the table to be able to deliver a compelling and meaningful solution. AI needs a lot of guidance. It gets easily lost if it loses context, and of course it lacks empathy and a true understanding of human psychology. What does true actually mean? So AI knows a lot of things about human psychology, probably more than us, but it just knows them. It doesn't really understand them because it can't feel, so if I'm in front of a person, I can imagine what that person feels. I can attempt to build empathy and compassion because I've maybe experienced something similar. AI obviously doesn't feel, so, for now.

Yeah, so this is the slide that people are always most excited about because it's very tactical. Like how do you bring up AI in your work as a designer? But I just want to invite everyone to do this exercise of what's my current process in whatever work I'm doing? What are the parts that can be outsourced to AI, and what are the parts that I enjoy doing so I don't want to outsource? And what are the parts of which AI is not very good yet, so better not? The most important thing when you look at this slide is the fact that at the center of it, it's the person that's deciding what the process will be. We're governing this process, right? We decide what needs to happen next, what we need to find out next, what are the missing pieces in the process of building a solution. So we're orchestrating all of it, and I think that that's something that will essentially not change for a long time. Plus, products like, I don't know if you've seen, you know, Wizards. This is a tool in which you essentially prompt text to generate design. So this means that producing design will not be what will make us special in the future or not replaceable.

But it's always going to be the thinking behind it, right? So you can now have design gems with products like this. You just tell them, I want this kind of mockup for I don't know what website and you iterate, you get where you want it to. And so that's not something that will differentiate us in the future. But what will always be the case, I hope, at least. I don't know. Nobody knows. But we can employ these technologies to solve the right problems. And so understanding what are the right problems to solve? What is important? What matters? What should I do next to make the world better? That's on us. And so this aspect of critical thinking and making connections and putting all the things together to understand what to do, and then for whom, and so on, that's something that's still in our hands.

And yeah. AI can only enable us to make some parts of that process better. But it won't replace the process. So I have a newsletter about AI. I just launched it, End Design, AI End Design. And yeah. I want to invite everyone to subscribe. There's a lot of noise in the industry. So what I'm trying to do is to be a gatekeeper between noise and things that are actually meaningful. And I'm on time. This is so exciting. Yay.

What's your favorite part about AI that you spoke about? Yeah. My favorite part about AI. I don't like AI. No. I mean, I think it's just this aspect of, this is something I've been reflecting in my work for the past week a lot because we had this internal workshop about what AI should do for our users and so on. And one of the things that kept coming up for most people, and I'm personally a fan of, is that there are so many moments of frustration and procrastination, and I just don't want to do this task. I don't want to move that pixel, or I don't want to write that PRD, or I don't want to write that code, and so you just won't do it because it makes you sick that you have to do it. And then AI can be there and kickstart that process. Help you make that leap into starting, by giving you a probably silly draft of what you're about to do, but it's a starting point. And so for me, it's this aspect of enabling us to move with less friction through the workflows that we have every day. So just leave the parts we don't like to this entity, and we can't leave them completely, which is also good, because you want to think about everything you're doing. But yeah, it's just removing the friction from the things I don't want to do.

Yeah, thanks. I saw one view in Twitter, I think, that people don't want that AI help them with coding, they want that AI will help them with home chores like laundry, washing dishes, etc. So okay, we have questions. Not all products benefit from AI, but still, investors want to see AI solutions due to the hype. Any suggestions? Yeah, just say you have AI. You're not gonna get funding if you don't have AI.

It's something that I've noticed first hand. I have a lot of friends who have their own startups where they're trying to get funding, and it's always that. But where is AI? Every investor wants to see AI in there. I think it's really easy to honestly add AI in your slice. Just do something that has some AI aspect to it. I think it's also quite related to the hype. I think that products in the future, we won't have to say we use AI. It's gonna be implicit. Yes, it's a way, it's a part of the solution building for products. It's not gonna sound special. It's just gonna be another instrument like coding or whatever. But for the time being, just say you have AI, and experiment with these technologies. So it's not just about saying you have it, but also trying to look for opportunities to use them. They're increasingly accessible. So I think that it's interesting, just an honest experiment, not just for founders, for investors.

Thanks. Next question regarding my previous comment. How do you think we can separate our human need to create an AI, slowly taking that away from us with AI, art, music, etc.? I think about that a lot as well. I think what's interesting is that so many people across the world had no medium or skills or tools to express themselves. Like everyone is creative. But some people can draw, other people can sing. Most people have none of that available.

And so for a lot of people who were not talented, or just didn't invest time in growing these skills, now AI is a, it's a channel for expression. It's a space where they can say, I have this image in my mind and now something can create it for me. And those people are not gonna be the million dollar artists, right? Artists will still be artists. But the regular person who's not talented, who's working as a nurse or whatever, they will be able to also express themselves through these technologies. And you hear that a lot when you have interviews with people in the AI space with users, like, I don't know how to write. So now GPT's helping me write. And it's not, it's not the next Kafka, right? It's not gonna be super talented. Because there's this human aspect of suffering and everything that needs to go into something to make it valuable. But it's just enabling people to express themselves, which I think is something that could make the way we look at AI a bit nicer. It's just empowering people to take that internal creativity that had no outlet, and use it for something. Even if it's something that will never be commercial.

Yeah. I saw how some people create amazing things with prompting now. They just, like, put words in correct order. Okay, I think maybe last question. My PM and UX designer treat AI as magic wand. But I can't use it that way as developer because of my myriad limitations. How to get them back to earth? What? Which is the question? The top one. I have the same problem as a designer. So everybody who's kind of... Yeah, it's interesting. Because as designers, and I think PMs have the same problem. Now everybody in the leadership levels is expecting us to magically do stuff with AI, both internally, externally.

But we're all navigating design systems, and all sorts of technical constraints, and security issues. And so it's not just that easy. You can't magically integrate with GPT overnight in your product and do stuff. I think that this is a perpetual conversation. Like what needs to happen at this stage of the story is for us to kind of continuously discuss these expectations. Like unpack them. Are they realistic? Why they're not realistic? It's just, in a way, a communication problem. Because it's about setting expectations within your team and just... Yes, discussing about the art of possible, all the things that we could be doing with AI today. But also discussing the technical limitations and yeah. Some companies don't have access to the internet. So yeah.

Oh, it's not... I think we have maybe time for one more question. What is the biggest trap in building apps that create content with AI? The biggest trap? Trap. Problem. Yeah. Well, I have a lot of very direct insights from our users. At Miro, we had essentially a surface where we summoned GPT as our own internal assistant that could enable them to do stuff. And the problem is that the technology for many parts is not there yet. Which, it's there for many parts. It's doing amazing things. Like 4.0 is amazing and so on. Should I shut up now? But the idea is that it's still infant technology. Not so infant. It's also... Yeah. I don't want to say it's completely immature because it's not the case anymore. But it's not there. Whatever the proverbial there means.

So users are disappointed every time they use it in most products. They try, they experiment and they're gonna try one, two, three times and then they're not gonna try again if they get disappointing results or if they can't use whatever AI was produced. Then they're gonna drop it and maybe try again in one year.

So most products trying to embed AI, infuse AI in their experience are losing users. So I've talked with people from Notion, from all the cool tech companies doing AI. They don't have good adoption. People experiment with it and then they drop it because it's not good. The question was, I think, how can you tackle that? I don't overpromise. I think most products overpromise. Write your blog posts with AI. Sure. You can write something. But is it gonna be the... But you still need to put work into it. So maybe just a more honest setting of expectations. We can help you do 20% of your work and then you have to do it on your own. It's also this disconnect between what the entire discourse about AI is and what it can actually deliver, which causes this problem. So more honesty, guardrails, and just better systems, I think. Whatever that means.

Yeah. Thank you very much, Johanna.