When Airbnb started 10 years ago, it began with its co-founders Brian Chesky and Joe Gebbia inflating three airbeds in their San Francisco apartment.
Since then, guests have checked in 300 million+ times into Airbnb properties, in 4.5 million places across 81,000 cities.
How does a start-up go from three airbeds to a $31 billion company? With the help of artificial intelligence (AI).
AI is central to everything that Airbnb hopes to achieve as a company. We caught up with the company’s VP of engineering, Mike Curtis to talk about AI, data and the next steps for Airbnb.
AI is intergral to Airbnb's mission to change travel
Curtis joined Airbnb around five years ago, after working for several tech companies including Facebook.
“[Facebook] was really fun, but I felt like Airbnb was doing something that I really wanted to be a part of,” he explains. “It was enabling people to go out there and see the world, as opposed to spend more time on their screen.”
The AI terms to know
AI — computer intelligence, with machines doing things that we associate with humans, such as problem-solving and learning
Machine learning — often used as a synonym for AI but actually referring to a specific type of process where a computer can get better at something, often by using lots of data to train the system to get smarter
Algorithm — a process or a set of rules to be followed by a computer in order to solve a problem
As the VP of engineering, Curtis manages the 1,000-strong engineering team, across four offices, including one in Beijing, China. When there is a new launch, such as the recent introduction of Airbnb Plus, the team builds all the digital products: the new sections for the website and the apps.
“When we think about what we’re going to build, we think about what is possible with technology right now to inform what we decide to do.”
A really important aspect of this is managing the data teams at Airbnb, who build the AI and machine learning algorithms that keep the platform moving.
“Every time you interact with an Airbnb app or the website, you’re interacting with machine learning in some way or another,” explains Curtis.
For instance, AI is in search rank. When you’re searching for a place to stay on Airbnb, such as Windsor, you don’t see listings in alphabetical order.
Instead, an algorithm is seeing the similarities in the places you click on, how long you look at them and the places you look at in the most depth. Then, it re-ranks search results to find the places where Airbnb thinks you’re most likely to stay.
“We can show you the listings first that are more likely to give you the best experience offline,” explains Curtis. “It works for us because we don’t just want to get a booking for the sake of a booking; we want you to book something which is a great experience.”
Then, that great experience will lead to a great review. You’ll probably leave a 150-word review at the end of a stay and never think about it again. But reviews are integral to the way Airbnb functions.
“They’re really critical for when you’re thinking about making a booking. A few experiments we did showed reviews first on listings from people who spoke the same language, or were from the same area as you. The second ended up being really helpful for people to complete bookings.”
And, did you realise that Airbnb uses machine learning translation to automatically translate the messages between you and a host if you speak different languages?
“The fact we can do reliable machine learning translation of reviews and messages is evidence of the advancement of AI in the past few years,” says Curtis.
AI and the problem with bias
When it comes to AI, one of the big focuses is on how to prevent bias from teaching machine learning algorithms.
“It’s great that this is an industry topic now,” says Curtis. “One of the ways we attempt to fight bias is by having prompts in the product that help raise awareness for our users for potential moments of bias. So they’re aware of it and so their behaviour, in turn, doesn’t train the algorithms in a way that appropriates bias.”
For instance, when hosts sign up to Airbnb, they sign a community agreement to be welcoming of people from all backgrounds. “This is setting expectations up front about what it means to be a part of our community,” he says.
If Airbnb sees a host reject a booking for a set of dates, it will automatically block the host’s calendar for those dates.
“We assume that if they can’t host the person for those dates, then the place is not available. And if they choose to go back and open those dates we ask them why they rejected the guest.”
This is interesting because it shows that Airbnb is taking a responsibility for its community and how its hosts act. In the argument between tech companies as a platform or a service, such as the argument over whether Uber is a taxi company or merely a platform, the tech company usually argues it doesn’t have a responsibility.
However, Airbnb isn’t taking this line. “We take the view that we’re building a community. That means we have community guidelines, we keep track of them and that’s the stance we take,” explains Curtis.
AI and chat: is the Airbnb concierge service coming soon?
As Airbnb continues its plans to change how the world travels, China has become one of its fastest-growing markets.
“I’ve been to Beijing many times, the first maybe 10 years ago,” says Curtis. “I expected it to be less developed but it’s one of the most urban places in the world. Seeing how much development has happened over that time is amazing.”
China has its own technology ecosystem, based on platforms such as WeChat, which is full of chatbots. These are AI computer programmes that simulate conversation.
You can use bots in WeChat to book an Uber, find information about a product, or find a restaurant to visit.
“Apps became incredibly complicated. But dialogue is this amazing way that you can traverse so much different technology in an intuitive way.”
And this makes it relevant for travel, he says. “When you’re planning a trip, how many apps and websites do you look at to figure out the logistics? Dialogue could be interesting for us to make it easier to plan travel and provide a concierge service whilst you’re away.”
Could an Airbnb concierge bot be on its way? “We’ll have to see,” says Curtis.
But if one does materialise, it won’t be limited to text. Curtis says there is potential in voice interfaces as well, such as Amazon’s Alexa or the Google Assistant.
“You can imagine a voice interface [would work] especially for travel. If you’re out in the world, wondering [where] you should go for dinner tonight, being able to invoke [recommendations] via voice command could be really powerful.”
For now, Airbnb is concentrating on Plus and what makes the platform work so well: expanding the ability to live like a local when you’re abroad.
“Travel is such a huge part of the human experience and for a long time, it felt very mass-produced and not unique to the person that is travelling. We want to totally change that – and the ambition and push for that has always been my favourite thing about being part of Airbnb.”
