Skip to Content

Intelligent, Chatty Machines

A startup hopes to help computers have meaningful conversations with people.
September 19, 2007

A new company called Cognitive Code has built software that it believes will let everyday gadgets talk with humans. At the Techcrunch40 conference in San Francisco on Monday, the startup unveiled a developer’s studio with a set of algorithms that convert strings of words into concepts and formulate a wordy response. The developer’s studio could let businesses, such as cell-phone manufacturers and toy makers, use the technology to add conversational abilities to a product.

Gabby gadgets: SILVIA is a new platform designed to let computers, cell phones, toys, and other gadgets carry on realistic conversations with people. Cognitive Code, the company that created SILVIA, is initially targeting toy manufacturers. Here, the interface for SILVIA is represented by a woman’s face on a laptop screen.

Instead of composing an e-mail on a PDA, says Leslie Spring, the company’s chief technology officer, imagine instructing a handheld to “send an e-mail to Tom and tell him ‘I’ll be there in 10 minutes.’” Spring says that such a feature could be possible with the algorithms–based on 15 patents–that Cognitive Code has developed.

The problem that the company is tackling is called natural-language processing, and it’s been the subject of intense research at world-renowned research labs for decades. Some computer programs are already able to parse basic information from inputs that don’t match exact commands. Well-known examples are chatbots such as Alice and Jabberwacky, programs that simulate a conversation via text input.

Spring claims that Cognitive Code’s product, SILVIA (which stands for symbolically isolated, linguistically variable intelligence algorithm), is more advanced than chatbots for a couple of reasons. First, SILVIA remembers and understands the context of a conversation. For instance, if you’re talking about the movie Star Wars and ask what the plot is, the system refers to earlier pieces of the conversation to retrieve an explanation of the movie’s plot instead of giving a general definition of plot, or the plot of some other movie or book that was discussed before Star Wars. The other key aspect of SILVIA that makes it different, says Spring, is its ability to comprehend concepts that are worded in a variety of ways and produce uniquely worded responses. “You can speak to SILVIA using whatever phrase you want,” says Spring, “and it extracts meaning. And on the reverse end, we have algorithms that can put [responses] back into human language. Sometimes we’re surprised at the way SILVIA creates these things.”

Multimedia

  • Click here to see an early version of SILVIA interact with Cognitive Code chief technology officer Leslie Spring.

The system works like this: during a conversation, words are turned into conceptual data, Spring explains. SILVIA takes these concepts and mixes them with other conceptual data that’s stored in short-term memory (information from the current discussion) or long-term memory (information that has been established through prior training sessions). Then SILVIA transforms the resulting concepts back into human language. Sometimes the software might trigger programs to run on a computer or perform another task required to interact with the outside world. For example, it could save a file, query a search engine, or send an e-mail.

This high-level explainer could describe a number of conversational artificial-intelligence systems, but Spring claims that the major advance is that SILVIA’s core algorithms can be implemented into a variety of applications and devices. At the Techcrunch40 conference demo, for instance, Spring showed a couple of the tricks that SILVIA can do when installed on a computer. For example, using SILVIA, Spring was able to open and close Word documents simply by making the request verbally in a conversational manner. (There are some programs that let people perform operations on computers with spoken words, but they require specific verbal commands.)

Spring says that SILVIA could work with Windows, Mac, or Linux operating systems. Also, Cognitive Code has shown that the algorithms can be compressed enough to run on a smart phone. Initially, however, Spring says that Cognitive Code is targeting the toy market, and he hopes to see products, akin to Teddy Ruxpin, with “SILVIA Inside” within a couple of years.

It’s still unclear, however, exactly how easy it will be to plug SILVIA into any system. Vasant Honavar, a professor of computer science at Iowa State University, says he expects that it would take a fair amount of work by engineers to put SILVIA into each platform. Honavar adds that in some of the applications that Cognitive Code is targeting, the responses will need to be much more accurate than the technology behind chatbots. “If you want it to delete a file and it deletes the wrong one, that wouldn’t be good,” he says.

And while Honavar is pleased to see Cognitive Code exploring business models for natural-language processing technology, he is hesitant to get too excited. “What I worry about [with] all this AI in general is that sometimes it’s oversold, and it’s hard to deliver on what’s promised.”

This is part of the reason that Cognitive Code is initially targeting the toy market, says Spring. There’s a lot of flexibility in a teddy bear’s conversation with a child, and the outcome isn’t necessarily as critical as it could be on a computer used for business.

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.