Designing for AI: why & how
Conversational UI’s, bots and “new frontiers of UX” are — rightfully — hyped these days and claimed as the next big thing, wiping apps of the map by merging advances within AI and design of conversational experiences. Let’s nuance that debate.
First, introductions: Hi there 👋
Why on earth is my perspective relevant? Well, I’m a designer gone product manager at BotSupply, where we work with AI, computer vision, NLP, conversational interfaces and design on an everyday basis.
Phew. Formalities = done 👌
So let’s talk about the fun stuff, shall we? Specifically, I would like to discuss why/if “design” is critical in developing AI products, and furthermore how we can (maybe) utilize it to not “only” build AI products, but actually design the right solutions and design the solutions right.
Before we proceed, I should warn you though. This is not a “how-to-be-ever-succesfull-with-chatbots-in-2mins” kind of post. So I begin with just a tiny bit of theoretical framing to make sure we’re on the same page 🤓
First: what the f*** do I mean with design? Let’s rewind…
Historically, “design” reasonates with aesthetically pleasing objects in areas like fashion, architecture, furniture, graphics and maybe even art. This entails an idea of the designer as a creative mastermind capable of a designing “stuff”, we normal humans could not even imagine. Such a view on a design process and creativity could maybe be visually represented like this:
This may not be breaking news for designers, but I find it important to include it in my perception of design. Because while the above may also be true — and “designing the thing right” is definitely part of design — I’m still gonna claim that it’s the least important part.
In my POV, design is a multi-faceted process, in which a designer utilizes a toolbox of methods, technologies, approaches and, well, tools, to adapt to the given project.
In reality, I suggest the answer to “what is design” is: It depends. Design is both beautiful UI, and cultural artefacts, and design thinking, and UX & IxD and a way of creating knowledge through research, to name a few. In other words, it’s an ambiguous mofo and more of an approach, attitude or way of thinking and working, than it’s a craft of building something beautiful. It’s a toolbox.
So design should be applied to the creation of any product or process with the goal of making something new, whether that be services, physical objects or intangible technology like AI — it’s just a matter of picking the right tools. This also means figuring out what to build, why we build it, who it’s for and how we can build it. An easy way of navigating those questions is to use “The Triangle of Ambiguity”, a tool for product management introduced to me via cool kids Made by Many:
The triangle provides a simple way of bouncing back and forth in the search of trying to answer all of the questions and thereby create an awesome product. It also emphasizes the need to continuously reflect on why we’re creating something, what we’re even building to target that why, and lastly how we’re gonna do it.
I like that figure a lot because it’s simple, and on a superficial level it contains the essence of a design process. For instance, why and who are connected to design research, with which we can get the immensely valuable insights in users, context and pain points, which often forces us to re-think our conceptions. Additionally, playing around with what and how is to me synonymous for working agile with iterations.
Continuous and fast feedback from users, mockups, and interactive prototypes are part of the design DNA 💪
Secondly, a quick comment on AI…
… Juuust to make sure we’re also on the same page there. As you can maybe tell by now, I’m leaning towards design-side of things, so I don’t assume I’m the man for nailing the definition of AI. But for the sake of transparency, here goes a superficial version:
I believe the idea of AI can be very simple.
Hence, I consider AI as a set of tools and approaches that leverage the power of data and computation in order to able to somewhat predict and learn (aaand I’m casually skipping a definition of learning too 😬).
Relevant for my field of work are especially recent developments within AI, such as computer vision and NLP made possible by machine learning techniques such as deep learning with neural networks. Why does that matter for designing AI? So happy you asked!
The consequences of this trend/hype/research — pick your favourite label, but it’s real — are the emerging fields of chatbots, digital assistants and generally the rise of Conversational Interfaces Design (CID) or its more broad cousin Natural User Interface (NUI).
Scattered around the interwebz, these design-activities are labeled as “the new frontier of design/UX” — and it’s easy to understand why. They provide so many interesting possibilities, while at the same time being as old as human interaction (so when I say new, it’s in relation to GUI-times): New constraints, new affordances, new research required, new interaction patterns, new design skills required and new frustrations.
If you’re hungry for more, follow my publications and go follow interesting people like Joe Toscano⚡️and/or watch his presentation on “Designing Intelligence” .
On the other hand, if you’re interested in a more tech-savvy perspective on AI, check out the work of e.g. Rahul Kumar and Kumar Shridhar, both AI Scientists.
School’s over !
If we accept the premise that design is really a way of structurally working on creating new products in an iterative and continuously reflective manner, then creating of AI products fits perfect (sorry for the buzz-spam). Why?
Because design will tell us what AI product to create, how it should look or feel, who it’s for, which features are important — for both MVP and solution — and how we can create such product.
But how does it work in practice? As part of my job, I have tried to merge AI and design, and create a process blueprint, which works for an AI startup reality, while at the same time emphasizing design- and product needs to ensure that we both design the right solutions and design the solutions right. And maybe it could be something like this — I’m always open for feedback:
On the surface, this look familiar (Yes, I’ve borrowed parts and tuned them). It doesn’t include specific methods and tools for each step in the process though, because the purpose of this model is rather to create a tool for dialogue (internally and with clients) by showing an overview of a process and loops (reach out if you wanna know what’s behind).
Easy to understand? Yep✌️ As easy to actually do? nope👎
Ideally, I begin any new process with the left, but (and this is where it’s tweaked to AI-reality) the most critical part for us is what I refer to as an explicit design intention, i.e. what do we want with this project? Eventually, we obviously want to create a new product — but this may be 4–5–6 iterations down the line.
The intention is paramount, because if we, as consultants, engage in a project, where we skip to building, it means that the premise may be shaky or completely assumption-based.
This is not good if a client expect us to deliver a well-designed and finished product. Check out the examples of processes and their potential activities 👇
In the bottom case, we do not aim to create a product or even a prototype of a solution. Instead, we might aim to learn about users, to understand a technological limit, to play with a new process or method or just evaluating real challenges and needs. Tangibly, we’re maybe creating a mock up to learn or test a hypothesis, but that’s it. And no matter what, we’re gonna go back to one point in the process to begin next iteration.
I cannot state how important it is that we explicitly distinguish here — especially when playing around in a new market as conversational products still are, because many have incoherent or even straight up absurd expectations.
So, just like the case for a quality chatbot design: Always leverage expectations, and be clear on the path you’re going down to, and what the shaky parts are on that path!
Why do we even do business in such cases then?
This is where both our status and the market of AI- and chatbots are important: First, we’re a startup so business needs doesn’t always match design needs, i.e. we need money, because salary is also pretty cool. We adjust to this by focusing on the intention and iterations, as explained.
And secondly, people have crazy expectations compared to their limited budgets for AI- and chatbots, so they struggle to pay the actual costs of what we deliver — creatively and technologically. This makes MVP’s and iterations a more market-ready strategy for us, and that’s why we’re continuously shuffling business models as part of how we structure our design processes, merging sales, design and development.
What’s next?
My next article will be a case study, where I present a concrete project, where I utilized exactly this view of design and AI product design/management.
I hope that I have maybe provoked you, given you new ideas and/or maybe even inspired you to start working with design and AI yourself. Anyway — you made it to the end, so you deserve an awkward 👏
See you later, 🐊
For more information on my work/approach, please just comment here below or get in touch at simon@botsupply.ai. Don’t be a stranger — come say hi 👋
