There’s no question that machine learning and AI can pull off some creative feats, as exemplified by something like 2016 Cannes Lions darling “The Next Rembrandt,” which analyzed every work the painter ever did to create what a totally new painting by him would look like.
But was that really creativity? The algorithm did create something completely new, but it was within the context of very strict parameters.
“I would prefer to define creativity as an ability to observe and perceive the world in new ways, then building connections and patterns within those observed concepts to generate a solution,” says Vitaly Pecherskiy.
Pecherskiy is the co-founder and chief operating officer of StackAdapt. As a demand-side native advertising platform, the company already uses machine learning to get ads in front of the most relevant audience, and is very familiar with the capabilities of data-powered algorithms.
During a speech at Dx3 Wednesday, Pecherskiy examined whether or not those same algorithms could be used not just to target ads, but come up with creative solutions to business problems.
Ultimately, the idea of a single machine-powered source for creative solutions seems unlikely, he says. But that doesn’t mean there aren’t benefits in going down that path.
The strength that computers and machines have over human intelligence comes down to raw processing power, he says.
“We tend to define intelligence based on how closely it resembles the way humans think,” Pecherskiy says. “I don’t think that’s a fair way to look at it, because humans are not the pinnacle of intelligence. If you define intelligence as being able to hide from predators really well, we might not be that intelligent compared to any animal. So I don’t want to judge computers based on things they are inherently bad at, like empathy or emotions.
“Instead, if we look at something that computers are good at, like processing and navigating massive amounts of data, they are way more intelligent than we are.”
That strength can be used to process volumes of data and information humans could never accomplish, finding patterns, connections and trends. But what machines have in processing power, they lack in the kind of reasoning, empathy and practical concepts humans use in the creative process.
As an example, Pecherskiy says a machine could be told that images with bright colours help drive consumer engagement and conversion. But it might have no idea why a photo of someone with pink hair would be appropriate, but someone whose entire body has been painted pink isn’t.
“Over the course of time, a human can reinforce the learning and get the machine pretty good at spitting out ideas that are pretty good,” he says. “But fundamentally, it’ll have no idea why some ideas work and others don’t. It will just know that it should stop generating ideas that use certain elements.”
As another example, Pecherskiy cited airline’s ongoing problem with seating – maximizing the number of passengers it can fit onto a plane while keeping boarding quick and the experience tolerable. He referenced a AI design where seats were turned to face each other, not realizing how the lack of leg room or the prospect of being face-to-face with a stranger during a four-hour flight might not be a desirable experience.
“We are seeing machine learning being able to pick up on topics and patterns that humans could not identify, and it might be able to create content that should influence purchase behaviour,” Pecherskiy says. “But a machine will not be able to qualify these ideas as feasible because they don’t have awareness or empathy. For them to come up with one good solution on their own is hard to imagine.”
So machines can be creative; they just might not be that good at it. Being able to generate ideas outside of the parameters of human limitations can even be an advantage when it comes to outside-of-the-box ideas. But the bottom line is that there will still need to be a human there to sort out when that lack of human concepts lead to bold new thinking and when it creates something that is completely unfeasible.
“In the context of what we do, if you go back ten years, you’d have to approach an individual site with an ad campaign, have a brainstorming session around images to use based on a survey or panel, and it’s all very unsophisticated,” he says. “So it’s a huge value just to spit out a lot of solutions to a problem really quickly. In the future, humans will move further and further towards qualification of ideas and machines will be in charge of processing the data.”
