Building the Better "Warfighter"

According to Tim Weiner in The New York Times with the release of the Defense budget for the new fiscal year, there come a few snags.

The Army's plan to transform itself into a futuristic high-technology force has become so expensive that some of the military's strongest supporters in Congress are questioning the program's costs and complexity.

Army officials said Saturday that the first phase of the program, called Future Combat Systems, could run to $145 billion. Paul Boyce, an Army spokesman, said the "technological bridge to the future" would equip 15 brigades of roughly 3,000 soldiers, or about one-third of the force the Army plans to field, over a 20-year span.

That price tag, larger than past estimates publicly disclosed by the Army, does not include a projected $25 billion for the communications network needed to connect the future forces. Nor does it fully account for Army plans to provide Future Combat weapons and technologies to forces beyond those first 15 brigades...

Ah yes, Future Combat. Sounds like George Jetson on steroids. A preview of what's so pricey comes in the 2006 DARPA Budget Projection (warning: the D.o.D. takes note if you lift that unclassified link). Let's spend a few posts strolling down science geek lane, marveling at the futuristic ways to kill Donald Rumsfeld's darlings are dreaming up for us.

Things like software to drive killer robots...

The Foundational Learning Technology thrust seeks to develop advanced machine learning techniques that enable cognitive systems to continuously learn, adapt and respond to new situations by drawing inferences from past experience. The application of this technology will result in military systems that are more robust, self-sufficient, and require minimal or no platform-specific customization. Current projects will develop hybrid learning techniques to create cognitive systems capable of learning military strategy, leveraging large amounts of prior knowledge, incorporating external guidance and applying prior knowledge in real-time to the naturally changing environment, all without programmer intervention. The Foundational Learning Technology thrust comprises Real-World Learning, Bio-Inspired Cognition, and Learning Locomotion and Navigation.

· The Real-World Learning thrust will explore the integration and application of advanced machine learning techniques to enable cognitive computing systems that learn from experience and adapt to changing situations. The program will emphasize the ability to transfer knowledge and skills learned for specific situations to novel, unanticipated situations and perform appropriately and effectively the first time a novel situation is encountered. The program will drive the design and implementation of new hybrid learning technologies, such as large-scale transfer learning, multi-purpose extensible knowledge learning, learning with minimal direction and learning generalized task models. The program will stress technologies that combine statistical learning techniques with knowledge-based techniques that take into
account background knowledge and a priori experience.

Cognitive systems will a) learn and represent vast amounts of knowledge in forms that can be applied to unknown situations and domains; b) generalize learned knowledge and apply it to dynamic and unpredictable situations and c) reason about a situation or environment. Real-World Learning will enable systems to execute unanticipated tasks with minimal direction and will provide a much-needed military capability for coping with dangerous and unpredictable situations.

· The Bio-Inspired Cognition thrust (formerly Neuromorphic Learning Technology) will draw on continuing advances in neurophysiology and cognitive psychology to guide and augment traditional artificial intelligence (AI) approaches to learning, reasoning, memory, knowledge acquisition and organization, and executive functions. The work will focus on novel designs inspired by the function, representation and structure of the brain. This approach will expand traditional AI technologies from complex symbolic processing to new capabilities in memory, categorization, pattern recognition and fusion of perceptual/sensor information. Computational intelligence is in its infancy, whereas the human brain is the product of millions of years of evolutionary development. Thus, designing software inspired by the brain’s processing schemes can offer leap-ahead advances in cognitive systems. These systems will seek to emulate human performance in exploiting past experience in novel situations, learning in multiple ways, fusing multiple perceptual inputs in real-time, extracting concepts from specific experiences, forming hierarchies of associated memories and concepts, and directing attention through a complex executive process. This thrust will take a fresh look at the design and implementation of bio-inspired cognitive architectures modeled after human cognition that combine principles from neuroscience and cognitive psychology with traditional artificial intelligence based symbol processing and knowledge representation. Success will, in part, be measured by the ability of the systems developed to deal effectively with novel situations and respond appropriately in reasonable timeframes. This thrust has the potential to revolutionize a broad range of military applications through breakthrough performance of intelligent machines.

· The Learning Locomotion and Navigation thrust will develop learning and reasoning technologies that specifically address concerns in robotic systems. The resulting robotic systems will learn automatically to interpret sensor data and apply this knowledge to the control of their actuators, which will improve locomotive and navigational autonomy in complex environments. Approaches in reinforcement learning and technologies for learning from example will be explored. These technologies will open new horizons for unmanned military operations, surveillance and reconnaissance, and will dramatically advance the capabilities of autonomous vehicles. Tasks requiring higher-level computation, such as perception-based navigation, will also benefit. This thrust comprises two components: Learning Applied to Ground Robots (navigation) and Learning Locomotion...

The Knowledge-Based Technology thrust will develop enabling technologies, methodologies, ontologies and detailed knowledge bases to achieve the next generation of intelligent, knowledge-intensive systems. This work will focus on developing technology that spans the spectrum from large, strategic knowledge banks to small, individual knowledge-based systems. The Knowledge-Base Technology thrust comprises Knowledge-Based Systems and Bootstrapping Cognitive Systems with Implicit Semantic Knowledge.

· The Knowledge-Based Systems program will develop technologies to acquire, codify, link, integrate, and use complex and crossdisciplinary knowledge at varying scales. At a strategic level, this capability will provide DoD decision-makers with rapid, as-needed access to relevant background knowledge from a broad spectrum of sources. The knowledge will be expressed in formal knowledge representation languages that allow computers to reason with the knowledge, consider its implications, imagine possible future scenarios and query the warfighter for clarification. The significant challenges are centered on the fact that critical knowledge involves temporal
information, complex belief structures and uncertainty. Current representation technolo gy is inadequate to capture such information. This program will develop technology needed to enable the creation of individual knowledge-based systems that would incorporate into the reasoning process (in a computer-understandable form) knowledge of the warfighter’s responsibilities, approach, tasks and activities. Another goal of this program is to support the warfighter’s ability to understand the “big picture” for mission planning, monitoring and replanning.

By formalizing situation-model representations, automated support will be provided to commanders and analysts for prediction of unforeseen events and determination of relevance of isolated or partial events to the evolving situation. To achieve these objectives, this program will develop analogic al and case-based reasoning, languages and situation markup languages technologies, and formalized situation representations. An additional goal is the development of technologies for rich, high-fidelity simulation models of human learning, reasoning and behavior. The program will also explore some new ways for knowledge to be transferred efficiently to a knowledge base including by reading tutorial text intended to convey new concepts to a cognitive system.

· The Bootstrapping Cognitive Systems with Implicit Semantic Knowledge program will explore a new technique for creating cognitive systems that store knowledge about the choices the warfighter has made in the past, so when faced with a similar task, the system would select a performance method by referring back to previous decisions. Although not appropriate for all cognitive systems tasks, this action-centric technique should be effective for simple tasks, such as information gathering to support mission planning or intelligence analysis.

Most cognitive research is predicated on explicit representation (i.e., having models of the world) and reasoning about the way to achieve a specific goal or meet specific need. While this approach is effective, encoding the knowledge and reasoning procedures is labor intensive and expensive. This program will develop a new technique that eliminates the material investment required by traditional approaches. This approach will replace deep reasoning and deep semantics with implicit reasoning and semantics derived from actual warfighter performance and experience...

That little software bundle alone will cost you $56.739 million next year alone.

But that's only one installment on the 5-year plan.

Of course, we want to make our Warfighter friendly, so to help the soldier in the field we have this little software gem- only an additional $57.192 million next year alone, not including the cost of the 5-year plan:

The Integrated Cognitive Systems technology thrust will develop advanced technology to enable a new class of integrated, highly functional cognitive systems capable of greatly assisting military commanders and decision makers. This thrust will build upon prior DARPA programs that developed improved human-computer interaction capabilities and highly-responsive computing systems. Integrated cognitive systems will seamlessly fuse perceptual inputs and tie newly perceived data to prior knowledge and experience. They will be able to plan ahead and will understand the world well enough to plausibly anticipate future events. Most importantly, these systems will have embedded learning capabilities that will allow them to retain prior learned knowledge, apply this knowledge to new scenarios, and ultimately provide faster and more effective responses. Overall, the ability to learn will enable the performance of a cognitive system to improve over time. The Integrated Cognitive Systems thrust comprises the Personalized Assistant that Learns (PAL) and Cognitive Command, Control, Communication, Computers, Intelligence, Surveillance, and Reconnaissance (Cognitive C4ISR) programs.

· The Personalized Assistant that Learns (PAL) program will develop integrated cognitive systems that act as personalized, executive-style assistants to military commanders and decision makers. This program will demonstrate cognitive systems that use basic knowledge and past experience to help them understand and seek input. Initially the program will strive to create assistant programs that display basic interaction competencies with people and other assistant programs in an operational environment. Some of these basic competencies include sending and receiving information in a natural manner; relating information and activities in various media; interacting with the assistant’s user and inferring preferences; executing procedures correctly; and accepting coaching and guidance expressed in natural language. In a unified multitasking, mixed-initiative architecture, these integrated cognitive systems will push the limits of technology for formal reasoning and learning. Methods for processing raw data will be learned in a way that optimizes performance of the entire system and enables the same purposeful perception that makes natural systems successful in dealing with huge amounts of input data and a constantly changing world. One of PAL’s goals is the development of advisable systems technology that yields systems that warfighters and other end-users can control in a natural and flexible manner, e.g., by exchanging advice and instructions, rather than via menus or programming. The term “advice” refers to a series of instructions that span a spectrum ranging from high-level policy and goals, to intermediate preferences and constraints on system behavior, to specific direction and contingency actions. The end-user will be able to engage in a natural dialogue with the system, and the advice will be translated to an executable form.

A killer robot with high level policy goals input from command control- or directly interpreted by the robot based on its experience...

Cylons.

If, you know, they could ever figure out how to program the thing to walk straight.

There's lots of fun things in this budget. That's over $100 million for next year in software I've covered on this post alone- about 4 pages out of 34.