As machine intelligence seeped into the mainstream psyche of humankind through the last decade, one lingering thought that trickled down the information pyramid (and still propagates) was that artificial intelligence is going to kill jobs and make swathes of us useless essentially via automation coupled with vastly different cost structures taking both time and money preference in a business model). As we have seen, quite a few mundane jobs are now actually getting automated and we aren’t needed for them anymore. However much like any other revolution that changes the industry, this one (assuming this is one) is likely to create jobs of different kinds than simply replace us. The key thought behind that is whether we are willing to skill up accordingly or not.
The field of design hasn’t been insulated from this industrial movement either. In recent times, at the bleeding edge of tech, we’ve seen code that converts designs to websites, generates designs, paints pictures, transfers styles etc. It has caused some insecurity in the minds of designers, but not a lot, I’d say.
We are all most likely here to stay
As a species, we definitely have major creativity chops. It gives us joy to make things, see them function, solve problems etc. Irrespective of what any of us does, we like to spend some time to do something out of the mundane that explores a part of our brain which we aspire to regularly exercise and possibly be known for. And yes, we all have mundane stuff to do every day, repetitively. In Design, tools have been an integral part of the job for a while now. Photoshop, for instance, has existed for 30ish years now, while a lot of new media tools have cropped up in the last 15 yrs as well, creating companies worth billions of dollars. A lot of the recent ones have been utilizing machine intelligence to jump through hoops. And that’s it. Designers have realized the potential of machine intelligence to actually aide them in their tasks. Quite naturally, design tools have started coming out in the markets that help designers expedite their work and put their time into things that really matter. As a UI designer, if I were to remove the backgrounds of images, I could do that on photoshop for 20 odd minutes or just have a program do it for me. Again, as a UX designer, I could wait for my product to go out in the market and check for usability stats or I could rather take care of the obvious problems right off the bat using pre-existing data and established insights before the product actually went to market. AI is increasingly becoming a natural aide to us Designers. Yes, there are products that help you build sites without actual code, but start setting up one such product and you’ll find yourself searching for templates and calling your designer friends for validation. AI is essentially becoming an enabler of design processes. A powerful exoskeleton for a designer if you may.
Opinions aside, here are some of the applications of machines and their computative abilities in design:
Computational Design: Computational design takes some steps ahead of standard design practices and allows algorithms to generate models based on a specific set of input instructions and sometimes predefined pathways. This creates the need for tools that deal with generative work built through multiple design iterations based on the input parameters. Tools that need the definition of process to work on the iterations. Instead of using tools to define a form, we use them to define a set of scalable instructions that propagate through various design iterations and hence create a large sample set of output possibilities. John Maeda, a very popular design and tech practitioner has been speaking about computational design quite a lot in the last few years and is definitely a voice to track in this field.
GANArt: The GAN (Generative Adversarial Network) was invented by deep learning legend Ian Goodfellow recently in 2014. GANs utilize two neural nets to generate visuals. One of them is the generator and one is the discriminator. Essentially, one builds something and the other classifies it to be the right thing or not. GANs have been recently utilized by artists in the juncture of art and tech to build GANArt. Some popular GANArt creatives have been Helena Sarin, Robbie Barat, Mario Klingemann etc. Some works generated using GANs are style transfers while some are purely generative works that seem abstract to the human eye.
Typography: The concept of fonts and how they impact the picture painted by a visual the font dominates, isn’t new. The way we perceive one font vs the other changes the perception of the visual representation of a word. Fonts have been shown to have multiple feature dimensions to them and hence can be mapped onto a multidimensional vector space. This allows the comparison between two fonts by measuring the Euclidean Distances between them. FontJoy and Fontmap(IDEO) are two such products that do that.
A significant mention would also be that of Responsive Typography via Interpolation. Designer/Techie Florian Shulz in one of his blog posts(2016) had written at length about how he had built a typography system that scales responsively based on the relationships that components have with each other, for instance, the mutual contrast of the text areas determined by their sizes.
Personalizing content and brand: Netflix’s been at it for a while now. Netflix engineers have described on their engineering blog how subtle changes in aspect ratios, contrast, hues, text placement etc on very similar images have massive implications on clickability of show previews. They essentially dynamically generate different show previews based on user personas. There has been a considerable amount of work around generating logos and brand assets via machines as well. Similarly in the branding space, there has been a considerable amount of work in generating colour palettes and even design systems for that matter.
Of course, none of these would have been possible without the actual inputs of design professionals.
Tools: There’s been a considerable amount of disruption in utilities in design via AI as well. Tools have been built that help in removing image backgrounds automatically, reducing image sizes massively, predicting eye-tracking on creatives, suggesting font pairs etc. Most of these tools use computer vision. Font pairing, as mentioned above, uses machine learning methods to explore the dimensionality of fonts and find their mutual distinction/similarity often via their euclidean separation.
Videos
Artificial intelligence in video formats has seen disruption only very recently especially given the challenges arising due to the contextually threaded nature of most videos, and the sheer amount of vision data that needs to be processed. In 2017 google open-sourced an app, “Storyboard”, that converts a real short clip into a series of comic strip panels with various styles. Recently, Nvidia had released an AI that interpolates through existing frames in videos to give you super slow motion. Similarly, Adobe has been making strides in video editing via its platform Sensei. Some of Adobe’s work has been around creating masks to type behind object inside images, semantic image segmentation, font recognition, scene stitching etc.
While major strides are being made as we’ve moved into the new decade there is a lot of literature being written around the impact of data-driven and algorithm-driven design practices. Machine learning is great at doing one standalone job extremely well, which is why we have with us AIs which we tend to call narrow AIs because they do one thing very well. This allows us to superpower certain aspects of our lives. While that happens there’s obviously a parallel stream of dialogues happening on the ethical impact of AI as well. Fields that can benefit from it in the immediate time frame are educating themselves more about the applications and limits of AI. It’s imperative to know the bounds of the machine and to understand the ethical limitations of employing them in tasks that affect human lives and behaviour. Design sits fairly in the middle of that.
