Designers’ reactions have been mixed. “AI, as developed by Wix….is no more artificial intelligence than robots on an automated assembly line constitute skilled craftsmen,” writes Ben Moss editor of WebdesignerDepot. “Wix ADI is an elaborate template browser.” For many business, however, that’s all they need. Half of small businesses still do not have a website 25 years after the arrival of the web and they cite expense as a major factor.

Wix’s first public example of an AI-designed site is an attractive if uninspired take on a personal trainer website. Acsaf predicts that today’s version of ADI, which is being rolled out to new users in the coming months, is only a hint of what’s possible as more users join. “The AI is as strong as the data,” says  Achsaf. “It’s only going to get better.”

Wix’s public example of an AI-designed website
Wix’s public example of an AI-designed website
Image: Wix

Others are not far behind. Weebly, which claims 30 million users, says while AI-designed websites are over-hyped and under-delivered to date, “we think having more intelligence behind the web design process is the future,” wrote David Rusenko by email. “It’s an area that we’re investing in.” Google and universities are researching how to make machines better designers.

Yet AI has inherent constraints to achieving great design, argues Igal Raichelgauz, CEO of image recognition company Cortica. Today’s technology excels at making good choices within narrow sets of rules, but that’s ultimately limiting because they’re not making decisions based on feelings and human understanding.

Today’s technology is reaching its limits, argues Raichelgauz. As machines grow better at unsupervised learning over massive data sets, it’s possible that self-taught systems can move beyond simple rules such as centering images or picking layouts with size ratios. He points to Google’s research into neural networks as a taste of what’s possible when we allow an artificial intelligence to make more “creative” decisions.

“In good design, breaking rules is just as important as sticking to them,” writes Raichelgauz. “When AI can learn when to do which and to make decisions accordingly, we’ll really have something.”

Neural net “dreams”— generated purely from random noise, using a network trained on places by MIT Computer Science and AI Laboratory.
Neural net “dreams”— generated purely from random noise, using a network trained on places by MIT Computer Science and AI Laboratory.
Image: Google

📬 Sign up for the Daily Brief

Our free, fast, and fun briefing on the global economy, delivered every weekday morning.