Just one of DARPA’s many AI programs cost taxpayers $150M. Now basically considered a success, SRI who built “Siri” has now sold it to Apple for up to $250M. In effect, we’ve all paid $1 to help build a Skynet prototype. Since I don’t agree with this type of research, for many reasons, I want my tax refund.
Siri is the product of DARPA’s “PAL” program, which stands for Personal Assistant that Learns, or Perceptive Assistant that Learns, Cognitive Assistant that Learns (CALO), or even Reflective Agents with Distributed Adapted Reasoning (RADAR), depending on which DARPA person you talk to. Here’s DARPA’s “PAL” promo video:
They’ve in effect completed the program:
The software, which learns by interacting with and being advised by its users, will handle a broad range of interrelated decision-making tasks that have in the past been resistant to automation. A CALO will have the capability to engage in and lead routine tasks, and to assist when the unexpected happens. To focus the research on real problems and ensure the software meets requirements such as privacy, security, and trust, the CALO project researchers themselves are using the technology during its development.
Venture Beat just detailed article on the transaction. Along with telling people “you want the virtual assistant to be as smart as someone talking to you on the phone and doing Google searches to answer your questions”, they also mentioned some important details:
Curt Carlson, the president and chief executive of SRI International, is very excited about the Siri intelligent agent technology that his organization sold to Apple in April for a rumored $150 – $250 million. He expects it will show up in future iPhones, but he also believes that SRI’s additional technologies from virtual assistant research will become part of even more rich applications in the future.
SRI was spun out of Stanford University more than 60 years ago to commercialize research. First named Stanford Research Institute and later renamed SRI International, it became independent from Stanford University in 1970 and now has more than 2,200 researchers working on things such as the cool artificial intelligence, speech recognition, natural language processing, location information, and agent technology that is part of Siri.
Siri, spun out of SRI 18 months ago, is a virtual personal assistant technology. Its first application launched in February as a free iPhone app that lets you perform tasks such as making dinner reservations by speaking into your phone and letting a virtual assistant do the rest.
In 2003, SRI got a $150 million grant to start CALO — the Cognitive Assistant that Learns and Organizes — to develop virtual assistant technology over five years. The effort included 25 world class partners and it incubated ideas from government and commercial researchers.
One of the projects that the military is interested in is the Command Post for the Future. It is an assistant for generals in a command post that helps make decisions about how to fight wars or skirmishes.
Commercialization of ideas started four or five years ago under the lead of Norman Winarsky, vice president of ventures, licensing and strategic programs at SRI. In that process, SRI takes its ideas and finds an entrepreneur in residence to make the idea into a startup. If they can figure out a big market opportunity and a compelling business model, they launch it. That’s what happened with Siri. Upon being spun out, Siri got another $24 million from Menlo Ventures and Morgenthaler Ventures.
But Siri isn’t the end of SRI’s research. It is just one part of it. Another startup using the SRI technology is Chattertrap, a Menlo Park, Calif.-based startup that will apply the personal assistant concept to online content. The idea is to create a personal information service by figuring out what kind of news you like and delivering a personalized newspaper to you.
In 2007 WIRED Danger Room interviewed Tony Tether, then DARPA chief, and he covered both PAL and CPOF:
NS: Do you know of anything that Darpa’s working on right now that’s really game changing?
TT: Yes — our cognitive program. The cognitive program’s whole purpose in life is really to increase the tooth-to-tail ratio [military-speak for the number of combat troops to the number of support troops].
… Our cognitive program’s whole aim is to have a computer “learn you,” as opposed to you having to learn the computer. We’ve got the technology to the point where we can now apply it in Iraq to a system that we also developed called CPOF, Command Post of the Future. It is a distributed command and control system.
NS: I mean, I don’t have to tell you that people have been promising cognitive computers [since..]
TT: Well, a lot of time has passed. … Since the ’90s to now, our ability to create algorithms that can reason — can more abstractly reason — about a problem and come up with answers, and also remember what they did using Bayesian techniques and changing values, has really advanced. I mean, it tremendously advanced in the past — from the ’90s to, say, the early 2000s. At the same time, computers became more powerful. We’re on the verge of having computers with densities approaching a monkey’s brain, and it won’t be long before we’ll have a computer with the density of transistors, or equivalent to neurons and almost human. What we’re missing is the architecture. So it seemed like it was time. We had great advances in algorithms for reasoning and in algorithms that learned in general. At the same time, the computers, the actual intrinsic hardware, was really approaching the density of a human brain. And so it seemed like it was time to try again. We’ve had some great success. This cognitive program I told you about is actually showing that it is learning, and it is learning in a very difficult environment. This is the program Stanford Research runs for us.
NS: Which program is this?
TT: It’s PAL [Perceptive Assistant that Learns]. And we have other related programs. One major research issue has to do with learning. If you and I learn something, like baseball, and then we go play another sport, say golf, we somehow transfer that –?? we are able to transfer some of what we learned in baseball to golf. That’s what makes humans very resilient and flexible. We have some research programs trying to come up with the same technique –?? that if you had something tackle a problem and then gave it another problem, it would do better on that second problem than if it had not had the previous experience.
PAL & CPOF are but 2 AI and Strong AI (AGI) programs DARPA is working on. In 2006, they had two offices that dealt with AI type research. IPTO dealt with “cognitive” programs, while iXo delt with “exploitative” surveillance type programs. The iXo section of the DARPA website had an animated interactive feature, for a time. I just so happened to video capture the entire thing. About 9 months later I noticed they had removed the interactive section, and immediately began work on a new video with that interface as the framework. From there I used all quotes from DARPA and related military documents, and all animations and graphics from the same sources along with the defense contractors doing the work.
I titled it “DARPA’s iXo AI Control Grid: The Official Version”, as it technically is official, being sourced entirely from their own words and displays.
What sucks is by about a year later I noticed that they had merged the iXo office into the IPTO office, which kind of ‘debunked’ my title as iXo was no longer a part of the website.
So at that point the IPTO office had all of their AI programs under one banner.
Today AI programs have spilled out of the IPTO office, and are now found across most of their other 5 offices. Considering that Obama made Zachary Lemnios DoD Director of Defense Research & Engineering, which makes him the boos of all DoD sciece R&D. Zack is a major AI pioneer, and former Deputy Director of DARPA’s IPTO office.
AI, including Strong AI, is quite literally the overarching agenda in Obama’s Pentagon, and the $150 million spent on the Siri project is only the tip of the budget iceberg. I’m working on a massive “AGI Manhattan Project” post that details every facet, and with that I intend to finish a film project I started several years ago now.
‘Skynet’ is an easy to use popular culture reference. The thing is, what they’re trying to build, makes Skynet look like played out Sega 8-bit.