Skip to Content
Uncategorized

Computers Can’t Answer Everything

A startup says natural language processing works best with human intelligence.
November 19, 2009

Providing answers to tricky questions has become big business online. But community question-and-answer sites can get clogged up with outdated answers, and it’s fiendishly difficult to create software that can automatically understand a question and provide the best answer.

Information hookup: The social search engine Aardvark helps users find other people who can answer questions for them.

Damon Horowitz, chief technology officer and cofounder of the San Francisco-based Aardvark, will outline a different approach when he speaks at the Web 2.0 Expo in New York today. Horowitz believes that the real power of natural language processing can only be unlocked by acknowledging its limitations and filling in the gaps with human intelligence. The company launched its product to the public last month.

Aardvark has done extensive research into using artificial intelligence techniques to answer questions, but the company’s focus has shifted away from training machines to respond. “We wanted to let another human being answer and have the machine do the heavy lifting of indexing everybody–the tens of thousands of people who are in your extended network and all of the things that those people know,” Horowitz says.

The difficulty of having machines interpret meaning has forced many “semantic Web” companies to focus on niche areas, such as answers to questions about medicine. “There’s a reason why all of our artificial intelligence systems only do so well with language processing tasks,” Horowitz adds. “Language has much more to do with live interaction with another person–understanding context and forming a connection.”

When a new user signs up to use Aardvark, she is asked to enter her Facebook login information and list of topics she is knowledgeable about. When a she asks a new question–for example, “What’s the name of a good restaurant in Cambridge, MA?”–the system tries to find other users who can answer it.

Aardvark has to parse the question to determine its topic before hunting for users with related interests. But the system also looks for potential answerers who are connected socially. It does this by gathering data from Facebook connections, and searching Twitter messages and relevant blog posts. The result, Horowitz says, is artificial intelligence that facilitates a human connection, helping users find someone who “is going to look you in the virtual eye.”

Aardvark forwards the question to chosen users and funnels the replies back to the asker. Users can ask questions through the Aardvark website, through an iPhone app, through e-mail, or by instant message.

Horowitz says users are motivated to answer questions through a desire to help out another person, pride in their own knowledge, and basic goodwill. In surveys the company has done, most users liked to receive questions at least once every couple of weeks. The site’s statistics also suggest this is true–Horowitz says that 50 percent of users who’ve signed up for Aardvark answer questions regularly.

Pedro Domingos, an associate professor of computer science at the University of Washington, says that having a human answer questions isn’t always necessary. He thinks it’s wasted effort to get a new answer every time a user asks, for example, about a standard physics equation.

Domingos also says that we shouldn’t give up on the idea of getting machines to answer questions. Data-mining and natural language processing systems have the potential to pull together data from a variety of obscure sources to respond to questions that no single human could answer, he says.

However, N. Sadat Shami, an IBM researcher who studies the way people search for expert information online, says Aardvark’s approach may be a good one for the consumer market. The questions asked through Aardvark may not need a single expert capable of replying. “You just need a response,” he says. “You need someone willing to put in the time to answer.”

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

This baby with a head camera helped teach an AI how kids learn language

A neural network trained on the experiences of a single young child managed to learn one of the core components of language: how to match words to the objects they represent.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.