Oskar Wahlbäck, Head of Global Operations

4 September 2020

No AI Software Will Replace Designers

Automation scares have long been part of our lives, and they are even more acute for jobs that use computers as the primary output tool. Recently, Art.Lebedev Studio revealed that design pieces from their cheapest price plan have all been executed by an AI solution. The results, however, are far from perfect and certainly do not spell doom for human designers.

AI Design Software

The Art.Lebedev Studio designers are somewhat vague about the software but one can still deduce the idea. The user provides the software (dubbed Ironov after the fake designer alias Nikolay Ironov) with text input, such as a brand or product name. The tool comes up with visual associations for the text and can iterate upon one or multiple associations to suggest design options. If everything goes well, they are all a finished design product.

At the core of Ironov is an unsupervised machine learning algorithm, as evident from the interviews with Art.Lebedev designers. They recall Ironov failing pretty miserably when it came to rendering animal body parts, but the software eventually got over that. The reason for such mishaps is that the software ran a computer vision solution through a ton of dog pictures but no one initially taught Ironov what a dog tail is. 

Inspiration-wise, Ironov has processed and keeps processing thousands (if not millions) of images to formulate their idea of a great design. Their source is unclear but I assume everything available on Google is a far game. Some photo banks that have previews without intrusive watermarks (like Getty Images) are an option as well. 

AI Design Perception

Art.Lebedev Studio has been “employing” Ironov for about a year with some 20 projects available in the portfolio. When the cheapest design was produced by human designers, a lot of the logos from the gallery below became news pieces larger than the companies themselves. Whenever that happened, it was not due to a positive response. 

Human designs for a burger joint and bakery

 

 


AI designs for a concrete company, PR agency, dumpling joint

Although the buzz admittedly was very much worth the low fee (some 1,300€), the design choices from Ironov have been weird. The most notable example is a commission from the Russian blogger Ruslan Usachev. His travel show is called “Пора валить”, which is a wordplay on whether one actually has to flee Russia. Usachev publicly criticized and rejected the first logo for failing to meet the brief (e.g. “employ the joy of exploration”), proving unsuitable for merchandise production and generally low-effort. When the blogger paid Art.Lebedev Studio again, they used a human designer for a more satisfying result.

Ironov and human designer logos for the travel show

The use of a human designer to meet Usachev’s requirements casts a grim light on the studio’s own perception of Ironov. On a broader scale, letters have been a popular target of critique. Judging from portfolio, it seems like the software continues to struggle with properly “drawing” letters. The Paint-esque level of detail and WordArt vibe do not really back up the advertised advantage of not being limited by a single human’s perception or style. 

AI Design Limitation

The first limitation is the software development investment that comes with using such products. Yes, a lot of the computer vision algorithms are public domain and one could argue that it’s all about weaving them. It is still, however, at least half a dozen people trying to put things together and keep the solution running, all without any output for several weeks and often on a higher payroll than designers.

Furthermore, designers are still a necessity. It’s unlikely that a programmer can choose a machine-generated template or even mix & match them to present the output that a client would expect from a design agency. You will need people to find gold in the pile of automated designs, which often exceed hundreds per brand name. 

Vision-wise, keep in mind that Ironov essentially sees the world through a database of images that someone fed it. If the ArtLebedev showcase is any indication, the software is yet to come up with new distinct styles that would define digital design for the next 6-12 months. It’s designers with a heartbeat who do that. By the very architecture, Ironov would struggle to pick up on a new trend. The samples will be scarce, so they will need to be manually weighed up against patterns formed on old images. Once again, a coder probably should not be the judge here. 

Summing up, AI-powered designs currently look like an entry-quality solution at best. I can see proprietary algorithms phasing out some Junior designers or maybe making them maintain software solutions design-wise. Most people and companies with ambitions would still want a creative person to make a new artistic product for them. The choice will be similar to over-the-counter vs prescription glasses: the second option has a wide price range while always healthier for the client.