To say technology like blockchain or AI is disruptive is a laughable understatement. At Airbnb, though, VP of engineering Mike Curtis said he’s not as concerned with being disrupted by blockchain as he is with the less technological challenge of convincing people to stay in other people’s homes.
To solve this very human problem, Curtis said, the Airbnb app and website use AI like predictive models and rank search results, which take into consideration like your preferences and rating and search history.
More than 100 signals are used to determine the search results you see, then machine learning is used to give those signals weight.
Airbnb began using AI to pair hosts and guests back in 2014. Today, all search results are personalized.
“There’s work we’re doing now building some predictive models that are able to tell us — for any listing that you’re looking at, whether you’re clicking on it or whatever — we actually have a prediction score behind the scenes of how likely you are to give that a positive review on the other end. And we’ve trained that model on all of the review data that we have from other people who have stayed there or also on your own reviews from places that you’ve stayed,” he said.
Curtis had a broad-ranging conversation about how Airbnb uses AI — as well as the future of work in the age of AI — at MB 2017, a gathering of AI and bot enthusiasts held July 11-12 at Fort Mason in San Francisco, California.
Another big factor in booking accommodations is price, of course, and Airbnb has AI for that, too.
That model, Curtis argues, has increased earnings for both hosts and Airbnb, and it furnishes an example of AI that’s giving people income from jobs that can’t be replaced with AI. Airbnb itself, he said, is an example of a job supplier that is creating a source of income for humans in the era of advancing AI and automation.
In its ambition to continue implementing artificial intelligence, Airbnb is just beginning to use unsupervised AI training to determine how best to show home listings to a particular user.
“One of the areas we’re exploring now is being able to think more about unsupervised learning on things like all of the unstructured information that we have — listing photographs, listing descriptions, everything else — such that maybe an unsupervised algorithm can get better at clustering listings together that have the same kind of feel to them,” he said. He explained that the goal is “being able to do that same kind of process on the guest population to understand where those clusters intersect in a way that’s possible or positive and then use that as part of the ranking model for search.”
Years from now, Curtis said, personalization could play a role in creating specially made itineraries and end-to-end trip planning so that Airbnb facilitates not just your stay in someone’s home but also your transportation, entertainment, and other details of your trip.