In the nascent world of artificial intelligence on the Internet, companies like Google, Facebook, and Netflix have built a foundation, and everyone else is trying to pile on. That, it seems, is the consensus of the Silicon Valley startups harnessing AI to offer businesses better ways to connect with consumers.
Continue Reading Below
A company could use AI in virtually unlimited ways, but it's best not to overdo it, according to Chris Monberg. He's the co-founder and CEO of Boomtrain, which bills itself as AI for marketers, and he belongs to the class of entrepreneurs who struggle to make ends meet while slaving away on a grand but remarkably simple vision: reinvent the same algorithms that Google came up with to recommend YouTube videos so that any website could recommend anything—from a pair of shoes to an airline ticket—to its visitors with remarkable accuracy.
"What if you don't have longitudinal user behavior like YouTube and Netflix have?" Monberg wondered aloud at the AI World Expo, a gathering of machine learning startups and experts in San Francisco this week. After all, Netflix controls 35 percent of peak Internet traffic in North America, which gives its prediction algorithms a massive amount of data from which to learn. For Monberg and his colleagues and competitors, the holy grail is AI that is just as simple and precise as Netflix's recommendations, but that can be easily adapted to a website of any size.
Like many tech entrepreneurs, they are driven by a desire to make the Internet a better reflection of what they perceive as the diversity and authenticity of the real world. Every Web surfer is different, so why can't every website adapt to those differences like Google and Netflix do?
"Personalization in general is critical for every single digital property," Adam Spector said at the AI Expo. The company he co-founded, Liftigniter, has several success stories under its belt. After graduating from the prestigious Y Combinator startup competition in 2014, it has applied its algorithms to everything from sports scores (teaching a soccer website to know the difference between live and historical game scores) to travel booking (ordering flight results by price and connection time).
Liftigniter and Boomtrain are competitors, and their leaders unsurprisingly describe a shared vision to spruce up the ordinary Web—the sports fansites, deli menu pages, and fashion blogs that record hundreds or thousands, not millions, of pageviews per day.
Continue Reading Below
For Spector, the vision is motivated largely by admiration for what Google achieved with YouTube. True website personalization, he argues, began at Google four years ago, when the search giant "productized" machine learning and began to record YouTube viewership metrics in real time so that video addicts would always have something relevant to watch next.
"We've been trained by Google and other services to get what we want the first time," he said. "You trust it to give you the answers as fast as possible every time."
Besides Java, there are a litany of other open-source software platforms that AI startups use. Companies like AIBrain and DataBricks harness Apache Spark to crunch huge datasets in the cloud. Google's own TensorFlow is also popular, and even powers oddball ideas like Intraspexion, whose ominous but still-in-development product scans employees' work emails for words like "discrimination," "lawsuit," and other litigious sentiments, automatically alerting company lawyers.
Like Spector, Monberg also extols the virtues of open source and simplicity. After a near-bankruptcy, though, his company has expanded its product offering to include other AI niceties such as messenger bots and a Google Analytics-like dashboard.
Besides Google and Netflix, Monberg is inspired by the online behavior of millennials, especially those who eschew big social-networking sites like Facebook (which, admittedly, is the king of delivering personalized content).
"Millennials are looking for authentic conversations," he said. One of the worst mistakes a marketer could commit is to assume that because someone liked several pictures of a friend's newborn baby on Facebook, he or she also wants to buy baby food.
Although they toil in the shadow of the giants that inspired them, Silicon Valley AI startups aren't likely to improve marketing to the point where Web surfers will happily disable their ad blockers and stumble upon their own personal consumer utopia. But their hope is that by applying the same algorithms that recommend cat videos, the mom-and-pop corners of the Internet will become slightly more welcoming and relevant.