What AI will mean to marketing (when it works)

Artificial intelligence (AI) has a lot to offer over human beings as a brand representative. AI never gets tired. It never forgets. It doesn’t need incentives, bonuses, or stock options. However, just like your junior brand manager, it can sometimes tweet abhorrent content you would rather forget.

One crisp spring Wednesday, Microsoft unveiled Tay, an artificial intelligence chatbot meant to simulate an energetic young woman with “zero chill.” The experiment ended quickly, and poorly, when Tay became a crude, racist monster.

Perhaps if Tay had been a slower burn and took longer to have her nervous breakdown, we would have had more time to consider the strategic implications of handing over a big part of brand identity to an artificial intelligence. Tay went nuclear so quickly and was pulled from service so decisively that the fallout seems to be limited, and people seem to understand that the biggest sin was that Tay was unprepared to consider the quality and reliability of her input before adapting and adopting certain attitudes and modes of expression.

“For everything you do in AI, you need to think about malicious users and what they would do, and how would their behavior influence the system?” says Amos Azaria, postdoctoral researcher in the Machine Learning department at Carnegie Mellon University.

(Microsoft’s position is that it is in fact very good at making chatbots, thank you, and that Tay’s problem was due to a “coordinated attack.”)

Somebody is going to try this again, and sooner rather than later. Here’s what AI can mean to marketing, and how to prepare.


Agents of persuasion

Tay was designed more to win hearts than to change minds, but Azaria says that AI that actively cares about influencing decisions is coming. His research focuses on the intersection of persuasion, deception, and artificial intelligence, and looks at how AI can embrace the goal of encouraging a human to make different choices based on reasoned appeals, evidence, and incentive.

The marketing applications are obvious. Marketing arts are already designed to address needs and overcome objections. Typical marketing automation script flows rely on nodes with pre-programmed content, discounts, or other incentives based on a specific point in a flowchart. AI can look at the entire conversation, as well as a wealth of other buyer data in real-time and devise a new approach to get around every roadblock.

And they’ll never run out of material. “AIs can read massive amounts of material and give each person the argument which it believes would most convince them,” Azaria says. “There’s a limit to how much people can read throughout their lives, but AI can read all of it in a fraction of the time.”

(In the future, people who love pithy quotations from famous people will be the easiest marks in the world. AI bots can look up all of the pithy quotations in human history very easily and pick the most obscure yet compelling winners.)


Agents of immunity

Marketers will have artificial intelligence on their side, but buyers will have AI allies, too. And a buyer-side AI agent is immune to Jedi mind tricks, and doesn’t particularly care about your brand halo. All it cares about is getting the best deal possible for its human sponsor, and it won’t be distracted by pop-ups or high-margin upsell offers that add little value.

Buyers could use the help, too, particularly when it comes to navigating cluttered markets where offerings are so diverse that no layperson can reasonably process all of the options. Enter MyWave, which has developed a buyer-side agent dubbed Frank. “There are 27,000 potential energy deals you can be on in New Zealand. Nobody can possibly sort through all that information and get the best deal,” says Wally Brill, VP at MyWave. “Frank goes out, finds the best deal, and changes it for you, because you’ve given him agency to take that best deal.”

Buy-side agents aren’t all bad news for brands. MyWave is working on helping brands adapt to the future by providing intent-casting data, which shows when an intelligent shopping agent is searching for deals, and reveals the parameters and constraints of the desired deal.

“When you see that real-time information about the item specifications and price point buyers are looking for, you get to decide if you can match it, or if you need to adjust what you charge,” Brill says.


Learning from Tay’s shame

The Tay implosion reinforced that computers are better than most people at a handful of important tasks. These include:

  • Brute-force calculation.
  • Playing chess, and lately, Go.
  • Cooking rice.

They are not, at present, guaranteed to be better at thinking on their feet while staying on brand, a lesson the next pioneer will have to work to overcome.

“My advice is to be careful, because these [AI experiments] can really cause damage,” Azaria says. “When you use a bot, especially one which learns from its environment, you don’t know exactly how the interaction will end up.”

Post by Jason Compton

Jason Compton is a writer with over 15 years of experience covering marketing, sales, and service. Based in Madison, WI, he is a regular contributor to Direct Marketing News, previously served as executive editor of CRM Magazine, and has been published in over 50 outlets.