How to Tell AI Hype from Reality

January 11, 2018 — by Todd Wasserman    

In 2017 it was hard to escape talk of artificial intelligence. Gartner claimed that AI had reached the peak of its hype cycle and deep learning and machine learning were at the “peak of inflated expectation.”

Such excitement is par for the course in tech. Some have compared AI in 2017 to cloud computing a decade or so before. Though the cloud was hyped relentlessly, actual adoption was slow. AI may be on a similar path.  A 2017 SAS survey of European business execs, for instance, showed just 24% had the right infrastructure in place for AI.

Is the market ready for AI? That’s a moot question because marketers are already using many AI technologies. Some have been using for years without calling it AI. But other AI applications are still years or even decades away. Here’s how to tell the difference:

AI reality

Most AI applications are improved versions of what we would have called “data mining” a few years ago. The difference is that there is often a predictive layer. Marketers are now using these technologies in varying degrees:

• Content curation/recommendations: IBM Watson teamed up with UnderArmour in 2016 to create a personal health consultant that uses a particular consumer’s data to craft a personalized training routine. For instance, a 45 year-old man training for a 10K could use UA’s Cognitive Coaching system to get advice on diet and exercise goals, which are based on data from other consumers in the same demo. More recently, IBM conducted a pilot that involved integrating its Watson Cognitive Bidder with MediaMath’s DSP to incorporate more unstructured data like the weather or location to fine-tune real-time buying decisions that address consumers’ current states of mind.

• Natural language processing: Consumers telegraph their needs via search queries and social media chatter. But because of the way we express ourselves, translating that language into data can be tricky. For instance, as Stanford computer scientist Chris Manning recently pointed out, “terrific” used to mean “to terrify” and has only fairly recently taken on a positive meaning. Thanks to wide daily exposure to natural language, machines are getting much better at picking up on such nuances. Some companies even analyze incoming sales calls to unlock patterns and opportunities.

• Image recognition: This is another AI technology that has become so common that it’s mundane. We take it for granted, for example, that Facebook will be able to “tag” people in photos we post. For marketers, image recognition provides analysis of visual imagery. U.S. consumers now post some 3.25 billion photos a day. Marketers can analyze such images to get clues about behavior and brand affinity.

• Ad targeting. Of course, one of the most popular deep learning applications right now is ad targeting. By analyzing data, machines can optimize ad performance via targeting. Marketers use this tool to improve their ROAS. In theory at least, your ad targeting keeps improving over time.

• Customer service/chatbots. One bad customer service experience can undermine millions in advertising. That’s why some brands are turning to chatbots to handle such inquiries. Experiments with voice-based chatbots have found that they can solve consumers’ queries 82% of the time and leave humans to handle the more complex queries. In addition, brands like 1-800 Flowers and Whole Foods have used chatbots to establish a brand presence on Facebook’s Messenger app.

AI hype

While those technologies are in use right now, some AI being discussed won’t be of use for years — maybe even decades. Those include:

• General intelligence: This is the type of AI usually portrayed in Hollywood movies and sci-fi novels. In reality, today’s AI is “narrow AI,” meaning the intelligence is focused on a single task, like playing chess or understanding spoken language. Exhibiting a well-rounded intelligence — one that can make quick connections and exhibit wisdom — is still a long way away, if it is ever possible.

• Complete automation: Another fear about AI is that it will monopolize all jobs handled by human workers. While some tasks — like customer service — are increasingly going this way, they are often shifting the work of human workers, rather than replacing it completely. As a corollary, look at the spread of ATMs. While you might think that they would have replaced human tellers, instead there were more because ATMs made it cheaper to open new branches and more branches meant more jobs for human tellers.

• Creativity: In advertising, AI can be used for “dynamic creative,” which swaps in marketing messages based on available data. (For instance, you might see an ad for a sunny vacation in Rio if the temperature has just plummeted near you.) But what machines are really doing is grabbing elements that a human has created. While AI systems can now compose music and even write Harry Potter fan-fiction and create movie trailers though, humans are still required to exercise judgment. An AI system is great at presenting random combinations based on exposure, but a human is still needed to make sense of it. Making unusual connections (like Jay-Z’s sampling of “It’s a Hard Knocks Life” from Annie) is still beyond even the most advanced AI system.

As we move on from a year in which AI hype has peaked, 2018 will likely be a building year in which businesses and marketers pragmatically apply the technologies that affect real-world outcomes. Meanwhile, AI itself will keep improving, expanding the palette of what’s possible and what will be possible in decades to come.