Fast Company

Siri-ously DARPA

Apple's voice-control technology for its upcoming iPhone 4S was spun out of DARPA.

Don't let her dulcet voice and easygoing, eager-to-please manner fool you. Behind Siri, the voice-controlled personal assistant app destined to power Apple's iPhone 4S, lies the heart of a hardened combat veteran. That's because the technology was spun out of the Defense Advanced Research Projects Agency (DARPA), the Pentagon's high-tech research and development arm.

While Siri was developed as cognitive software with artificial intelligence baked into its code--designed to learn from experience and respond to ever-changing threats on the battle field--iPhone users will find more prosaic uses for it as a hands-free replacement for the touch screen. It'll even decipher drunken slurs, so you can order your phone to order a pizza for delivery, text a friend, read email back to you, turn on music, search for the nearest Starbucks bathroom, and who knows: Maybe your phone will get so smart it will know what you want before you ask. Once there was Ask Jeeves. Now there's the ever mobile Siri.

For now it can only respond to simple commands, but the technology underlying it is anything but. The problem with most speech recognition technology has been that it has a hell of a time with all-too human variations in speech--accents, dialects, intonation, enunciation, and slang. Tell it you want to hide under "a rock" and it might tell you about "Iraq." Like the dream of the paperless office, which the advent of the personal computer was supposed to herald, speech recognition often makes more work than it saves. Siri promises to change all that, and you should thank the wizards at DARPA. While they didn't create the technology, they incubated it.

I visited the agency in August to interview DARPA director Regina Dugan and heard a common expression tossed around its hallways. For a project to be worthy it has to be "DARPA hard." It describes an agency that's unafraid of thinking big and sometimes failing spectacularly. Over the last half century, DARPA technology has led to predator drones, night vision goggles, GPS, and even the Internet. It's invested heavily in telesurgery, exoskeletons that function like wearable robots, landmine detection (including one technique that relies on honeybees), a flying car, handheld speech translation, and biosensors, as well the usual deadly array of guns, bombs, stealth planes and missiles. One intriguing DARPA-funded invention in development seems right out of a James Bond movie: a miniature robot that flies like a hummingbird and is equipped with a built-in camera capable of transmitting live video. 

 

Over the years DARPA has conjured up other far out, comic book worthy schemes, many of which never made it out of the laboratory: Steam-powered robots that would fuel themselves by eating everything in their path, ripping trees, shrubs and other biomass and stuffing them into built-in furnaces; telepathic spies who the agency hoped could use their psychic powers to spy on remote targets; a futures market to predict terrorist attacks, which was largely a victim of timing, coming not long after 9/11 and leading to public outcry; and, developed during the Vietnam War, a mechanical elephant to cut through dense jungle.

In August the launch of a $320 million [Ed.Note: See clarification below] unmanned hypersonic vehicle designed to fly at 20 times the speed of sound (13,000 mph)--it moves so fast its skin burns away--disappeared into the Pacific Ocean 45 minutes after launch, a situation Air Force Maj. Chris Schulz, program manager for the project, called "vexing." This prompted the Foxnews.com headline: "DARPA Loses Hypersonic Vehicle, Goes From $320M to Zero in 2,700 Seconds." (It was found a week later.) 

As Dugan told me, "You can’t lose your nerve for the big failure because the nerve you need for the big success is exactly the same nerve."

I can't wait to tell that to my Siri-powered iPhone, although I doubt it'll know how to respond--not yet, anyhow.

Adam L. Penenberg is a journalism professor at NYU and a contributing writer to Fast Company. Follow him on Twitter: @penenberg.

UPDATE: Eric T. Mazzacone, DARPA Public Affairs Officer asked us to include the following comment, which we have agreed to append in its entirety for clarity.

The August launch did not cost $320M. The cost of flight test two was $120M (which includes the cost of more than 20 test range assets and the craft itself). The total cost of the program since its inception in 2003 is $320M. Some of the significant technical advances that stemmed from the second test flight may be found here.

The use of Maj. Schulz's “vexing” quote is taken rather significantly out of context and is misleading. The actual quote from the DARPA release is “Here’s what we know,” said Air Force Maj. Chris Schulz, DARPA HTV-2 program manager and PhD in aerospace engineering. “We know how to boost the aircraft to near space. We know how to insert the aircraft into atmospheric hypersonic flight. We do not yet know how to achieve the desired control during the aerodynamic phase of flight. It’s vexing; I’m confident there is a solution. We have to find it.” Therefore, what is vexing is “knowing how to achieve the desired control during the aerodynamic phase of flight.” It is not that the aircraft engaged its autonomous flight safety system, which used the craft’s aerodynamic systems to make a controlled descent and splash down into the ocean. Controlled descent is a term typically associated with a human-in-the-loop directing or guiding the unscheduled landing of an aircraft. For DARPA’s Hypersonic Technology Vehicle 2 (HTV-2) controlled descent takes on new meaning thanks to the vehicle’s safety system.

Additionally, the vehicle was tracked continuously during its entire flight. Due to security requirements and release approval processes however, we were not able to publicly confirm splash down until three days after the test flight. Even without that knowledge of the release process, the idea that it was “found a week later” is incorrect since DARPA announced splashdown confirmation on August 14 (three days following the test flight) here.

Add New Comment

2 Comments

  • katie hatch

    Does the US Government retain the Intellectual Property rights to it's innovations? It seems like DARPA could fund itself a million times over if it were collecting royalties on patents.

  • SpeechGuy

    As a veteran of the speech recognition business, I can attest that DARPA was instrumental in the advance of speech recognition. Early speech recognition research was difficult to assess, since results were reported on different tasks and different speech data, making them difficult to compare. DARPA held an innovative annual conference, where companies and research organizations were challenged to test their methods on a speech database supplied by DARPA for the test, making the results comparable. The winning technology approach, Hidden Markov Models and Statistical Language Modelling, is at the heart of today's technology.

    As to Apple's Siri, the natural language understanding technology comes from SRI International, hence Siri. SRI has a number of DARPA contracts, I believe, and I suspect one of them may have been in natural language understanding.