AI as pair design
These days, there’s always a constant influx of new things to learn as a designer. We had the great “should designers code” debate; we also had accessible 3D crop up with new Figma-like tools such as Spline to learn; we had NFTs where you needed to grasp the concept of tokens and fungibility; and finally we have AI, where you need to learn how to integrate it into your workflow to avoid “losing your job”.
Granted, no, you don’t have to actually learn all these tools or ways of working. But as Designers and Developers, we’re often curious by nature and keen to be on the bleeding edge of technologies. Naturally that means we want to try things out, but importantly we also sometimes have to because our company (or companies we want to work for) expect us to.
Recently there’s been a lot of discussion about how AI could finally be after our design or development jobs. Like always with AI, it’s seen as an inevitability and that maybe we need to start looking for new things to be good at. Kyle Lambert wrote a great piece on this recently.
In reality, yes, AI may soon get good enough to do some of the tasks of a designer. But critically, it’s still very much at the early stage where it simply copies things from the internet and attempts to sometimes create something slightly new, but that normally always has flaws.
The more I’ve been playing with Chat-GPT which is OpenAI’s playground for its GPT-3 Large Language Model (LLM), the more I’ve gotten interested in treating it more like a companion than a threat. More like a pair designer or pair developer, helping me get around thorny problems or simply just kick of a project.
Instead of assuming AI will take our jobs, we can assume it’ll become the best pair designer/developer to help us get started.
I recently spent some time using it as a copywriting assistant when designing mocks.
I can ask it a quick phrase like “Write a headline that helps explain the importance of investing compared with saving into an interest-bearing account over the long term” which gives me the response “Investing for the Future: A Look at How Smart Investing Beats Out Saving for Retirement”. Not bad really!
What about “Write three bullet points for an app that helps explain the importance of investing compared with saving into an interest-bearing account over the long term” instead? This gives me:
Investing in the stock market, mutual funds or other types of investments can result in higher returns than the interest rate on a traditional savings account.
Investing allows your money to grow faster over the long term, allowing you to take advantage of compound interest.
Investing allows you to diversify your portfolio and protect against market fluctuations, providing valuable protection for your savings.
Also really not bad at all. And if anything, a helpful jumping off point to refine based on our copywriting style.
AI pair design will also provide a great mechanism for fast iterations on an idea. Think of it as the ultimate Crazy 8’s companion. Feed it the initial concept and let it go wild for you. These tools will also be able to quickly evaluate hundreds of design solutions and narrow them down as quickly as possible. This means that the design process can be completed much faster than if it were done by humans. Equally, they can also be programmed to automatically adjust parameters and learn from their mistakes, allowing them to continuously improve the design and produce more reliable results with added accuracy.
But this is more a future thinking concept when tools powered by DALL-E, Midjourney or other new AI models improve. Right now, the image generation is incredible but it doesn’t really hit the mark for consistently good app ideas, it’s much more useful for fantasy images.
Diagram have started to play with some interesting ideas here via their Magician plugin for Figma. As a way to generate quick icon iterations, UI and marketing copy or image ideas, it’s a fantastic tool to speed up your workflow. But it can’t replace an amazing icon designer, illustrator or copywriter right now.
Ultimately we’ll still need the human in the loop to ensure designs connect well with the people who’ll interact with them. It’s naive to think AI will be capable of human-level empathy any time soon or to make real, “conscious” decisions off the back of internal feedback. But one day we’ll likely get there.
I think we’ll see an entire industry boom around the design of great prompts soon too. Imagine feeding an AI with the perfect prompt to generate amazing outputs. We’re seeing this happen in the form of shared prompt spreadsheets or other websites, but there’ll likely be entire businesses that help you generate great prompts that generate great outputs. Then, eventually, we won’t need those anymore and we’ll just skip straight to great outputs from any prompt.
I think it’ll be important for us to embrace this new reality. Eventually every product will be powered in some way by AI, some are even already betting on it as part of their product strategy [1, 2]. So if you’ve not tried out GPT-3/3.5-turbo/4 and want to give it a go on iOS or macOS then you can try out an app I built for myself but decided to put out there. It’s a simple native companion that uses OpenAIs API to provide responses based on text input. You can read more about here.
I encourage you as a designer to embrace it to shortcut the more tricky part of mocking designs—coming up with fake but realistic copy—and let me know if Arti proves useful. As a developer, I’d encourage you to play with GitHub Copilot again if you’ve not in a while, it’s incredible and powered by Codex from OpenAI (which you can also use in Arti’s Chat tab).
Update: a great article from Geoffrey Litt about thinking of ChatGPT as the ultimate muse.
Update: another great take from Matthew Ström comparing AI in design to the notion of flying-by-wire.
Update: GitHub recently announce Copilot X which is the GPT-4-powered version of their prior Copilot offering.