Most of the designer’s work is to reuse existing design patterns. An experienced designer knows tens if not hundreds of patterns and selects one that will, with some modifications, solve the problem at hand.
Designers will not spontaneously propose AI solutions for design problems, if they are not in their pattern toolbox. How can we get them there?
Photo by Susan Holt Simpson on Unsplash
During a design process, designers first aim to understand the problem — the target users, their needs and barriers, the use context, the business requirements, available technologies, etc — before creating alternative design proposals [1]. It is an interesting question to think about the designer’s creative process: how do they come up with the design proposals? Where do they come from?
A designer relies on patterns
Usually, a designer will first consider solutions that they have earlier witnessed to work well for a similar problem. These ready-made solution candidates are called design patterns. (For the reader interested in the history of design patterns, read more here [2].)
An experienced designer can have tens, maybe hundreds of different design patterns in their toolbox [3]. Let’s take an example: the designer needs to present a large amount of information that is spread over tens of web pages. The task is to create a user interface that the user will find easy to navigate and find the information that they need.
For presenting the information, the designer has several design patterns that they can use. For example, they can sketch nested navigation menus for finding the information, add breadcrumb navigation to help users understand where they are within the hierarchy, or provide an efficient search with faceted search filters for more complex and uniform information, and so on.
An experienced designer has a bigger toolbox
A large toolbox and complementing skills and methods of efficiently finding the most relevant solutions are what tell a senior designer apart from the designers early in their careers. With the patterns, they can create accurate and well-functioning design proposals faster.
Sometimes we want to go beyond the obvious and explore completely new kinds of design solutions. This takes more time, and there is always a possibility that the new, previously untried solution will not be feasible after all.
Is AI too risky for your designs?
Do designers have AI, machine learning and natural language processing, in their toolbox? I claim that not many designers do. When these are not in the toolbox, we don’t intuitively present them as solution proposals to problems where they would be a good fit. We rather default to using more traditional and less risky design patterns.
This is one of the challenges for AI solutions to gain wider adoption. The question is: how do we get designers to understand the potential and get first hand experience of AI solutions so that they will be included in the designer’s toolbox?
Most of the traditional design patterns have been around for decades. It seems that AI and machine learning solutions have only recently become ripe enough to be really relevant in solving users’ problems. Therefore there has not been enough time for them to filter through to become staple solutions for designers to consider.
How to get AI in your toolbox
Although there are some resources for design patterns for AI (e.g. [4]), nothing beats hands-on experience. You can speed up the process of adopting AI to your toolbox by taking the time to analyze different AI solutions and what kind of problems they are successfully solving.
We can also develop our design toolbox by intentionally encouraging exploring AI solutions for our design challenges, even if AI wouldn’t be our first design pattern of choice for that problem. It also helps tremendously if you can team up with AI experts in this: they will be able to help you to understand the potential and the limitations of the AI models.
It is interesting to follow how the design community is currently exploring the possibilities of AI and thereby trying to adopt it as one of the available solutions in their toolbox. Following different trials, trying them out yourself, and learning from their successes and failures is a key ingredient in mastering the topic.
Machine learning has limitations
I have to admit that I’m still very much in the learning phase of AI solutions. It is not yet part of my daily toolbox, but I’m getting there.
One thing that I’ve learned is that most AI systems rely on machine learning: you need to have loads of training data first. After this training, you can show fresh data to the machine, and it knows what do to with it based on the patterns that it has learned in the training.
For example, if you have a smart indoor thermostat that aims to maintain a certain room temperature in changing weather, it needs to be trained first with a large set of different parameters of the weather forecasts and sensor data in order to accurately control the valves or resistors for reaching the desired temperature.
In many cases, the challenge is that there is no relevant training data available before the system is actually up and running. Possibly, the system can learn over time during use, but this means that for the first use right out of the box, we already need to have the primary solution first that doesn’t require any training. For a designer, this will always put the AI solution as a secondary option, or an addition that will improve the first UI using a more traditional design pattern. This will raise the question if the improvement brought by AI will be worth the effort.
The new kid on the block
Using AI systems that don’t require training before taking them in use is a completely different matter. For example, many language-based models and natural language processing (NLP) algorithms are ready and trained beforehand with of existing text material.
Large Language Models is the technology that you want to have in your toolbox.
Enter OpenAI’s GPT3 and other Large Language Models, and almost overnight the rules have changed. They have already been trained with millions of sentences of text. They are there, ready for use, almost in any language you want. This is the technology you want to have in your toolbox.
More of this in the next blog post. Stay tuned!
Literature
[1] “Five Phases of the Design Thinking Process”, Interaction Design Foundation
[2] “A Pattern Language”, Wikipedia
[3] “Design patterns”, Tim Bacheller
[4] “People+AI Guidebook”, Google
[5] “AI is a solution in search of a problem”, Panu Korhonen
