Editor’s note: The following story is the first in a series of articles on Sandia’s Cognitive Science and Technology Program. Articles in future issues of the Lab News will discuss various individual research projects underway at Sandia in this growing area.
* * *
Imagine a world where a machine creates a “virtual you” by modeling how you think and your expertise on a subject. Or one where your car’s computer assesses your driving skills and compensates for your limitations.
That’s the world Sandia has entered full throttle through its Cognitive Science and Technology Program (CS&T).
A revolution is at hand, says Chris Forsythe (6341), member of the Labs’ cognition research team. It’s not one of just better guns and weapons for national security. Instead, “It’s a revolution of the mind — of how people think and how machines can help people work better.”
“I believe that a fundamental level, science-based understanding of human cognition will, indeed, revolutionize the world,” says Steve Roehrig, director of Energy, Resources, & Systems Analysis Center 6300 and Sandia’s CS&T program lead. “Our long-term CS&T program emphasizes teaming with outside institutions to research and apply the spectrum of science underlying human cognition — from neural basics to the integrated brain, to understanding the mind and individual behavior. With an emphasis on national security, Sandia can have major impacts ranging from aiding neural scientists with new technologies and computational approaches to developing cognitive aids for humans on the other end of the spectrum.”
Focus on the individual
A large portion of Sandia’s program today focuses on the uniqueness of the individual interacting with others and with machines. It involves using machines to help humans perform more efficiently and embedding cognitive models in machines so they interact with users more like people interact with one another. The result is the ability for researchers to take advantage of the basic strengths of humans and machines while mitigating the weaknesses of each.
Cognitive projects and research at Sandia span the gamut from student training to assisting with Yucca Mountain licensing, from designing “smart” cars to using video-like games to train military personnel, and from determining how neurons give rise to memory to global terrorist threat detection.
Funding for the research has come from the Office of Naval Research,
Sandia’s Laboratory Directed Research and Development (LDRD) program, DOE, the Defense Advanced Research Projects Agency (DARPA), and other government agencies. The CS&T program also benefits from collaborations with the University of New Mexico, the MIND Imaging Center in Albuquerque, and most recently the University of Illinois at Urbana-Champaign.
Cognition and national security
The initial decision for Sandia to develop cognitive technologies is based on the belief that “there are numerous positive impacts cognitive systems technologies can have on our national security,” says Russ Skocypec, senior manager of the Human, Systems, and Simulation Technologies Dept. 6340.
“Everything in the world is becoming more individualized, including conflicts,” Russ says. “You see this all the time in Iraq and other places where terrorists are taking their toll. Security threats are more personalized.”
Russ believes that today’s conflicts are unlike others over in the past century. He says that although all wars are driven by humans, major influences on the outcomes have differed. World War I was a chemists’ war, World War II a physicists’ war, and the Cold War an economic war. Today, he believes, “we are engaged in a human war that is influenced primarily by individual human beings rather than technology or bureaucracy.”
That is why he considers it appropriate for Sandia, a laboratory with national security as its mission, to use its resources to better understand the minds of this country’s adversaries, as well as to use machines to enhance the Labs’ abilities to recognize patterns, deal with massive amounts of data, solve perplexing problems, and perform complex activities.
While Sandia dipped its toes in cognitive research in the late 1990s, the Labs’ real effort in the area started in 2002 when the program won an internally funded LDRD grand challenge. Based in part on the success and path set by the grand challenge in 2005, the former Mission Council — a group that consisted of senior Sandia vice presidents — selected cognitive science and technology (CS&T) as a research focus area for the Labs.
Most people who are well-versed on cognitive issues — like Danny Rintoul, manager of Computational Biology Dept. 1412, whose department has several people involved in cognitive research efforts — believe that the “near-term high-impact effect of cognitive science research would be to enable people to do complex, dangerous tasks in a safer manner.”
“This includes, of course, protecting and augmenting the warfighter, but eventually could be used to help people with a broader set of tasks that are not just dangerous, but unpleasant and tedious,” Danny says. “In the long-term, the real goal is understanding the inner workings of the mind. This is probably a 100-year goal, but the payoff for this would probably revolutionize human life as we know it.”
The largest complex problem
He adds, “Sandia is one of the premier places in the world where disparate groups of researchers have been able to work together to solve large complex problems. Cognitive science is probably the largest complex problem that scientists are considering right now, and our ability to bring together computational scientists, engineers, and neural scientists to tackle this problem is critical in solving it.”
Deputy Labs Director for Integrated Technologies and Systems Al Romig says Sandia’s move to emphasize cognitive research is “not unlike what we did when we decided to go into biotechnology in 1999.”
“Many people believe cognitive science will be a major focus of scientific research and technology development over the next several years,” he says. “Just like biotechnology is not the biology of our grandparents, cognitive science is not the science of even your older sibling.”
He calls this a “new frontier” that will grow rapidly over the next 10 to 20 years.
“Right now we can’t really know what will come out of a radical new area like cognition,” he says. “We may not see an impact tomorrow. Ten years from now may even be optimistic. But it’s our responsibility to be at the forefront.”
Strategic planning for Cognitive Science and Technology
During the spring and summer of 2006, the cognition team conducted two investigations. The first looked at what cognitive capabilities exist at Sandia.
The second examined opportunities involving the convergence of Sandia’s initiatives in the areas of cognition, biotechnology, and nanotechnology. This led to a Cognitive Science and Technology Plan with three technical objectives — a basic science understanding of the human brain, mind, and behavior; improved human performance; and advanced human-machine systems at all scales.
“The plan is at the level of ‘send a man to the moon’ — beyond the scope of what any one institution can possibly do,” Chris says. “It’s a synthesis of ideas. Now, our intent is to home in on a few areas in which the labs can make a unique and profound contribution.”
Chris says there are two elements to Sandia’s strategic planning for cognition.
“What makes most sense is for Sandia to select areas where we have unique, collective technical strengths, areas that few others in the world can do as well,” Chris says. “These include such capabilities as high-performance computing, nanotech, physics-based modeling and simulation, and surety.”
The human factor
The other element involves a focus on opportunities where specific national security problems have a human factor.
“As a national security lab, we have a responsibility to apply this emerging science and technology base to provide innovative new solutions to these problems,” he says.
In speaking about the importance of understanding the brain, Sandia VP and Principal Scientist Gerry Yonas notes, “Our best role is to harness the power of our computing and modeling, as well as sensors, to assist in the rapidly advancing field of neuroscience for national security applications. Sandia can apply neuroscience to national security problems in a variety of ways — by improving our understanding of brain functions and of the implications of stress, confusion, and ambiguity on decision making, all in an effort to enhance our own performance and degrade the performance of our adversaries.”
John Wagner, manager of Cognitive and Exploratory Systems and Simulations Dept. 6341, says the new area of research means “profound opportunities exist for the Labs.”
“CS&T’s ambitious direction may not be realized for many decades, but the information required for progress is emerging today,” he says. “It is reasonable to expect future discoveries will become the Nobel-class achievements for the cognitive and neuroscience communities at large in the years to come.”
What is a cognitive system?
The term “cognitive systems” has been used worldwide to identify a variety of programs, initiatives, and technologies. However, so many varied uses have led to ambiguity of meaning. Sandia has established its own definition: “Cognitive systems consist of technologies that utilize as an essential component one or more computational models of human cognitive processes or the knowledge of specific experts, users, or other individuals.”
“It is important to note in this definition and in all of Sandia’s work in the area of cognition that the human is always part of the equation,” Chris says. “Technology might try to take the human out of the system, but the human element never goes away. A machine might be able to do the work 10 humans can do. But a human has to be in the loop — there was a human who designed it, most likely, a human who uses and maintains it, and in the end, a human will probably dispose of it.”
To Russ, “the goal is to have seamless and sophisticated unions between humans and computers.”
“Extending human cognition in such a way offers people the unprecedented ability to increase their productivity as well as improve the quality of their work,” Russ says.
Further research, he says, on how behavior and the mind work together is needed before research can go much farther. knowledge of the brain is deep in specific areas but much is still unknown, particularly in understanding brain function at different scales and as integrated systems.
John says that cognitive research at Sandia — like most worldwide — is in its infancy. He anticipates that within the next decade research that seems like science fiction today will be a daily part of everyone’s lives. The cognitive revolution will be in full bloom.
“Once that happens, the best of both worlds can happen,” John says. “If we understand human cognition better, we can work together as a nation to reduce tension, find problems before they turn into armed conflict, and work toward actions that establish and maintain peace worldwide.”
Ethics and cognitive research
The new cognitive technologies promise a great future — machines helping people learn faster, better; new ways for experts to share information; and more efficient ways for people to work using machines. But potential danger lurks in the shadows if these technologies are misused or misunderstood.
“We are talking about a whole new level of ethical and legal considerations that may have impacts in ways we can’t yet even comprehend,” says Wendy Shaneyfelt (6341), member of the cognition research team who began thinking about these issues early in the research.
Russ Skocypec, senior manager of the Human, Systems, and Simulation Technologies Dept. 6340, says he credits staff members, like Wendy, who recognized the potential for ethical and legal concerns.
“They brought their concerns to the attention of management, which has been very supportive of advancing these objectives directly and objectively,” he says. “It is our hope to educate ourselves and society early on these issues to mitigate the potential for purely or highly emotional responses later.”
Many questions emerge when thinking about these systems, Wendy says. Who owns a cognitive model that represents a human (cognitive agent) and is responsible for its behavior? Can a cognitive agent be licensed? Can a cognitive agent legally act on a person’s behalf? How can access to a cognitive agent be controlled? What if a cognitive agent performs a business transaction on behalf of a deceased person? Who is responsible for erroneous information supplied by a cognitive agent? What if a cognitive agent shares personal information it is not authorized to share?
In an effort to be ahead of the game and be prepared with answers before the cognitive revolution reaches fruition, Wendy led a grassroots effort to explore how other emerging and well-established scientific disciplines confronted similar issues and what it would take to develop cognitive systems responsibly.
“We wanted to learn from the experiences that the biotechnology and nanotechnology communities have had as they confront ethical and legal issues in their respective rapidly expanding fields,” Russ says.
The result was the evolution of seven ethical principles and guidelines (see below) for the development of cognitive systems.
Working with the cognitive science researchers on her team and consulting with ethicists and psychology professors from the University of New Mexico, a framework was established to provide guidance and practical strategies to proactively address both real and potential ethical issues.
“In general, the principles vary in terms of the specificity of their content,” Wendy says. “Some are broad, like respect for persons, while others are more limited in scope — protecting the confidentiality of a research participant’s data.”Ethical principles for the development of cognitive systems
• Cognitive system developers will remain knowledgeable about and apply their respective established professional/ethical codes and guidelines as appropriate.
• Cognitive systems developers will proactively consider the intended uses and impacts of their specific technologies, as well as the potential for accidental use, misuse, and abuse.
• Cognitive systems developers will provide inherent safety features to the extent considered reasonably possible to maximize the prevention of accidents, misuse, and abuse.
• Cognitive systems developers will proactively initiate ethical discussions among themselves and support public engagements as these technologies develop.
• Cognitive systems developers will provide human test subjects with a clear understanding of the personal information acquired and how it will be stored, analyzed, and applied.
• Cognitive systems developers will responsibly handle any personal information obtained from test subjects.
• Cognitive systems developers will respect the limitations of a cognitive model as an imperfect representation of a test subject.
Investigating surety methodologies for cognitive systems
To further study the legal and ethical ramifications of cognitive research, Wendy Shaneyfelt (6341) and David Peercy of Sandia’s System and Software Quality Engineering Dept. 12341 established a multidisciplinary team of researchers from Sandia and the University of New Mexico. The team was charged with characterizing a range of actual, potential, perceived technical, and nontechnical risks associated with the development of cognitive systems.
The team participated in a series of workshops and conducted several focus groups to help understand what threats both scientists and the public perceived to be involved in cognitive research. Focus group participants included nontechnical people, such as University of New Mexico professors who had no scientific and technical expertise in this field and a well-educated public from the Albuquerque community; technical people including UNM, The MIND Institute, and Sandia personnel who were researchers in the field of cognitive sciences; and Sandia employees working in the area of surety.
“All participants agreed there was some threat with the cognitive technologies, in particular in the areas of privacy, legal, ownership, and reliability of systems,” Wendy says. “This was across the board — something unusual in this type of survey. The concerns of the scientists were not that dissimilar from the concerns of the public.”
Based on the findings from the focus groups, the team started looking at Sandia’s surety model as a possible way of mitigating risks.
“The model has been dispersed around the world to prevent the mishandling of nuclear weapons,” Wendy says. “We wanted to do to the same thing with cognitive technologies to avoid potential future abuses and accidents.”
Sandia’s surety model — which centers around safety, reliability, security, human factors, quality, and surveillance — has been extensively studied and applied within the Labs’ weapon/weapon-related areas as well as for other technologies.
“Our exterior advisory board believed Sandia has a unique ability to utilize surety techniques to help mitigate risks in this field,” Russ says. “They strongly encouraged us to apply them.”
Automated knowledge capture
Sandia initiated its cognitive modeling efforts in 1999. It involved many hours interviewing people and reviewing records.
“We realized from the beginning that cognitive modeling may be a wonderful technology, but doing it by traditional methods required too much time and cost too much money for the technology to ever be broadly applicable or to realize its true potential for national security,” says Chris Forsythe (6341), member of Sandia’s cognition research team. “There was a concern that our technology would only be applied for a handful of exotic applications where cost was not an issue. It was apparent that we had to automate the development and updating of models of individuals.”
As a result, research spearheaded into three different research arenas for building models — text-based models, training applications, and machine transactions. These became the basis for building models through “automated knowledge capture.”
Text-based models involve taking documents, papers, and emails and looking for information, including key words and word patterns — anything that might provide insight — to what a person is thinking or interested in. This also gives researchers the opportunity to look at changes across time. When they see new key words emerge, something might have changed. Heading up this effort is Travis Bauer (6341).
Training applications, headed by Robert Abbott and Elaine Raybourn (both 6341), are another major part of automated knowledge capture. Sandia has been working with the US Navy to develop a tool that can be part of its aviation training exercises. This involves building a model that can record students’ actions during training exercises. Computers monitor large numbers of people at once, detect when someone is doing something inappropriate, and bring it to the attention of the student or instructor.
The machine transactions segment of automated knowledge capture uses sensors to gather information about how a person drives. This information is put into a computer model. While a person is driving, the computer can detect when the driver is in a demanding driving situation (e.g., entering a high-speed roadway or preparing to pass another vehicle). This recognition allows the car to delay nonurgent communications such as cell phone calls until the driver can better handle the extra activity. Leading the machine transitions research are Chris and Kevin Dixon (6341).
Validation and verification play important role
Tied closely to the development of cognitive computer models are the validation and verification techniques to ensure the models are correct.
That’s where Tim Trucano (1411), validation and verification expert, steps in. He is in the early stages of developing ways to make sure the cognitive models are working as they should.
“I work on developing methods to generate and use information that allow cognitive modelers to validate their models,” says Tim, who has spent many years working with the NNSA Advanced Simulation & Computing Program.
He and his team, consisting of George Backus (1433), Alex Slepoy (4014), Laura McNamara (1433), and Scott Mitchell (1412), are seven months into a Laboratory Directed Research and Development (LDRD) project. They are attempting to create a validation methodology that “makes sense for cognitive and group interaction modeling,” Tim says. In particular, they are researching mathematical methods and computational procedures for scientifically comparing cognitive models with empirical observations as part of the validation process.
Ultimately they hope to develop a verification and validation strategy for the CS&T program, as well as specific insight into one or more projects.
Cognition, neuroscience poised for major breakthroughs
The fields of cognition and neuroscience are poised for the same major unifying breakthroughs experienced by the physics and biology communities of the last century and a half, says John Wagner, manager of Cognitive and Exploratory Systems Dept. 6341.
“In physics, complex disparate observations were reconciled based on simple unifying fundamental principles. Examples include electricity and magnetism, thermodynamics and statistical mechanics, and Newtonian dynamics and relativity,” he says. “Similarly, the biological revolution started when the extreme complexity and variability of the biological world was unified based on the replication and application of simple rules embodied in the geometry of DNA.”
Today, John says, there is some understanding of the brain (neuroscience) and mind (cognitive psychology and behavior) at various levels, but there is no unifying basic understanding of them or the fundamental principles of structure and function that apply to them. “Seeking this fundamental understanding from first principles, not just empirical observations, helps set the scientific direction for the CS&T program,” John says.