With sites like Yelp, it's never been easier to find restaurant reviews. But it's often easier to ask a friend to recommend a place than it is to sift through dozens of reviews to figure out which, if any, restaurant suits your budget or dietary needs.
A company called 1,000 Plateaus wants to change that with Luka, an app that's a bit like a cross between Yelp and Siri. It uses artificial intelligence to learn your preferences and scan online reviews to find the perfect places for you. It's only available in San Francisco, but the company plans to expand into other cities, such as New York, soon.
You can ask Luka questions like "Is there a good coffee shop nearby?" or "Is there a cheap place to eat in SoMa?" Instead of getting a list of possibilities as you would from Yelp or Siri, you get one suggestion. If that doesn't sound good, ask for something else. You also can ask questions, like what dishes are particularly good or whether it has vegetarian or paleo options, saving you from reading reviews and menus yourself. Depending upon how the conversation goes, Luka might end up recommending someplace else.
What's most impressive about Luka is how natural the conversations feel. Sure, it's obvious you're talking to a machine. And it's not hard to stump it. But Luka can parse an impressive array of questions, phrased many different ways.
It's a striking example of what can be done with machine learning, and an indicator of the conversation-driven interfaces that are increasingly common, thanks to tools like Siri or Microsoft's Cortana. "People talk about artificial intelligence as something very science fictional," says co-founder Eugenia Kuyda. "We wanted to make it practical."
Artificial intelligence already is being used by companies like Google and Facebook to handle voice commands, photo recognition and other tasks. But tools like Siri are the closest most people come to something that tries to understand human language.
But unlike Siri and most other virtual assistants, you must still communicate with Luka through text, not voice. But while Siri is designed mostly to answer single questions, Luka is conversational. "Siri would not remember the second or third step of a conversation," Kuyda explains. "It's one question, one answer."
Kuyda says the company doesn't plan to add voice support because Luka is designed to facilitate an ongoing conversation in which you might ask something, then return to the conversation minutes or hours later. And the company plans to add a group chat feature so people you're meeting can participate in the conversation as well, and Kuyda thinks that text messaging is the model for both of those uses.
Luka's underlying technology was developed for a banking application. A Russian telco contacted Kudya's consulting company to ask about an interface for banking via text message. The telco later dropped the project, but Kudya and company were able to keep the intellectual property they'd developed, and decided to repurpose it. Kuyda had worked as a restaurant critic, and thought a restaurant recommendation engine made sense.
Co-founder Philip Dudchuk, who has a degree in computational linguistics from Moscow State University, previously worked for the Russian government news agency RIA Novosti. He helped build technology to analyze the semantic structure of news---perfect preparation for making sense of online reviews. And much of the rest of the team came from Yandex, Russia's largest search engine and had extensive experience in indexing information and processing it through artificial intelligence.
After spending a year developing Luka---formerly known as IO---in Russia, the team moved to San Francisco last year and joined the startup accelerator Y Combinator. The plan is to enable Luka to make reservations at restaurants, and for 1,000 Plateaus to receive a small fee for each reservation.
Later, the company hopes to apply the technology to other areas, such as entertainment and travel. But the underlying technology could be applied even more broadly, and that's reflected in the company's name, a reference to a strange and sprawling book A Thousand Plateaus by French philosophers Gilles Deleuze and Félix Guattari.
In the book, the authors propose a non-hierarchical mode of organization called the "rhizome," named for the underground roots that sprout from the stems of certain plants. "You never know where a conversation will go, what it will be about, or where it's going to grow," she says. "Just like a rhizome."
