Categories

Tags

 “The Navy must get to work now to both build more ships, and to think forward—innovate—as we go. To remain competitive, we must start today and we must improve faster.” –CNO white paper on future fleet, May 2017

In World War Two, the United States faced a threat to its existence. To defeat that threat, we used every asset available to prevail and secure out nation’s future. The defense establishment’s most decisive effort was a multipronged project, engaging the brightest minds in science, technology and academia in a coordinated, fast-paced program to develop a new technology that drew the war to a rapid close. This effort, the Manhattan Project, created an atomic bomb, and all the infrastructure to manufacture it, between 1942 and 1945. This success assured a massive strategic advantage for the United State that lasted for decades. But it also created an air of technological invincibility that has infected out military organizations and defense industry. That sense of invincibility has become increasingly less justified in the first half of the 21st century.

The challenge the world throws at our fleet and operators continues to increase geometrically in diversity and danger. Shipboard combat systems and their operators can be overwhelmed with data but lack information and knowledge for offensive and defensive action. Existing systems, some of whose architectures date to the 1960s, have evolved in response, but are running short of the ability to grow. The defense industrial and technical base does not have the knowledge and techniques needed to fix this situation, but that ability is present outside the defense community. The time is right for a “step function” improvement in the ability to compute, evolve, and fight in an effort like the Manhattan Project. A broad-based, coordinated effort, drawing on new technologies and knowledge resident outside the defense community is needed to rapidly advance new solutions into service while also supporting long term developments.

The innovation engine that exists in the commercial technology universe of Silicon Valley and the rest of America’s high-tech innovation industry can provide an answer. The promise of artificial intelligence (AI), machine learning, and big data, combined with an aggressive rapid prototyping approach and teams of new experts, indicate a path. By engaging the non-defense technology community, learning its ways, and daring to create technical change, we can move ahead of the problem.

Toward An Artificial Intelligence Combat System

U.S. Navy ships routinely are forward deployed to dangerous places where they may be exposed to attack. Asymmetric threats and the emergence of near-peer competitors, especially China and Russia[1], require forward deployed U.S. Navy ships and fleets to respond to a broad and expanding set of offensive and defensive challenges—from weaponized jet-skis to robotic drone swarms, to supersonic surface-to-surface, air-to-surface cruise and ballistic missiles. Of particular concern are multiple, simultaneous attacks that can overwhelm both current combat systems and operators. While modern networked systems and information technologies have been adopted, on-board, remote, and space-based sensors can inundate at-sea decision makers and warfighters with a virtual Niagara Falls of precise, real-time data. This mass of communications pipelines and networks, disaggregated databases, and diverse information sources requires combat system operators to collate, fuse, and turn this welter of data into actionable information quickly and without fail. In peacetime steaming, this task is challenging; in the stress of combat conditions, operator overload can be deadly, as clearly shown by the 1988 incident in which USS Vincennes (CG-49) shot down a commercial airliner[2].

While military and naval technology has improved since then, state-of-the-art commercial technology has moved faster. Today, advances in machine learning, AI, and multisensor integration could remove much of the cognitive overload and contribute to a combat system that is versatile, adaptive, and intelligent. The Navy should move now to embrace these technologies now before the threat becomes unmanageable.

Artificial Intelligence

As has been the case for several decades, commercial innovation is driving much of the technology base. Breakthroughs like IBM’s Jeopardy-winning Watson and Apple’s personal assistant Siri, both debuting in 2011, are two well-known examples of AI applications. Since then, there has been a seven-fold increase in investment in artificial intelligence research and development[3]. Promising new fields of study within AI, like machine learning, are leveraging advances in computer hardware technology and massive amounts of ready data to train computers in ways previously not possible. AI is being built into a wide variety of new commercial applications for sophisticated diagnostic and predictive medical purposes, financial asset management, and self-driving vehicles. Analytical software and advanced computing to process massive amounts of big data are becoming commonplace.

The phrase “artificial intelligence” was coined in 1956 and attributed to Dartmouth professor John McCarthy to describe machine behavior that appears intelligent. Clearly, the technology to make a machine intelligent did not exist in that time. Today, increasingly sophisticated computer algorithms are taking on challenges once thought impossible, like speech processing, facial recognition, real-time financial market predictions, and industrial process optimization. Simple, rules-based expert systems software that mimicked rudimentary intelligence in limited domains are giving way to computer systems that can learn from running repeated trials on large data sets to optimize outcomes.

Three recent innovations that have made machine learning possible are neural network algorithms, fast parallel microprocessors, and extremely large data sets[4]. The following is a highly simplistic description of these three key elements of an artificial intelligence system.

Neural network algorithms are based, abstractly, on the way neural pathways in the brain work. With each thought, memory, or idea thousands of neurons are electrically activated in the brain, connecting them into a neural pathway. These pathways, once established, are used for cognition and recall. Similarly, in a machine’s neural network software nodes in algorithms act as neurons that are activated when sufficiently weighted inputs are available. Outputs from each node can become inputs to other nodes, creating an artificial neural pathway. Input weights represent decision criteria and adjusting these weights throughout the neural pathways helps improve and optimize the machine’s outputs over time.

However, conventional computer hardware can quickly become overloaded by the calculations necessary to accommodate even a relatively simple neural network. The number and complexity of calculations can be increased by processing them in parallel, using more sophisticated hardware. Interestingly, this hardware exists today because of the video gaming industry. Modern, fast-paced games, like Grand Theft Auto and Halo, display life-like video where characters and objects react in realistic, physics-based environments. This requires substantial parallel computing power of specialized graphics processor units (GPU). The technology in these gaming GPUs is ideally suited to the massively parallel requirements of neural networks.

Big data is the final enabler of a modern artificial intelligence system. While simple neural networks with relatively few nodes can be programmed with user-adjusted weighted inputs, the job becomes impossible for a human as the number of nodes increases in complex networks. The machines must adjust and adapt for themselves, using very large data sets to practice and learn how to optimize the variety of input weights to achieve the desired outcomes. For example, using the database of voice commands from millions of users, Siri is continuously optimizing voice recognition across a wide variety of accents and dialects. Likewise, Amazon, Facebook, and other corporations tune their AI systems with data collected from users to offer up better products and advertisements.

The Manhattan Project used recent advances in nuclear fission and associated technologies to quickly build a war-winning weapon system. Given these three enabling technologies—neural networks, parallel processors, and large data sets—and an increased interest and investment in AI research and development, the field is becoming more sophisticated at a rapid pace. Artificial intelligence research currently is making progress in several areas of importance, including intelligent interaction, deep learning, automated planning and control, optimized decision making, and perception through images, sounds, and other sensory inputs[5]. Given the successful examples of existing AI systems coupled with robust research and development, the field is ripe for applications that promise to revolutionize Navy combat systems.

AI Applied to Navy Combat Systems

Navy combat systems challenges have long been characterized through the detect-to-engage sequence. Warships rely on their onboard sensors and data from off-board networked sensors to locate, identify, and track other ships, aircraft, and submarines. In a dynamic at-sea environment, sensor settings must be adjusted as environmental conditions change, or to avoid mutual interference or hostile jamming.

Once detected, quickly determining the identity of a sensor contact involves correlation with other sensors, like electronic warfare systems, or operators’ analyses of the track’s course, speed, altitude, and maneuvers. Based on the totality and synthesis of all the information collected and a trained combat systems team’s best judgment, the track can be identified as a friendly, neutral, or hostile ship, aircraft, or submarine.

If a contact is judged to have hostile intent, its information is passed to an appropriate weapon system to acquire and track the target with sufficient fidelity to enable an appropriate weapon to be launched. Multiple contacts in close proximity may hinder an engagement by creating track ambiguity and uncertainty that could result in firing weapons on a nearby friendly or neutral contact.

Once an engagement is underway, operators must select an appropriate weapon, fire it, track the intercept, assess success, and decide whether a reengagement is needed. If a fast-moving target is closing the ship’s position, the operator may need to decide whether to pass the reengagement to a different weapon system. For example, at longer range, standard missiles might be launched, reengaging as necessary with shorter range evolved Seasparrow and then with rolling airframe missile (RAM) or the close-in weapon system (CIWS). A laser or other beam weapon may play a part in the future. Meanwhile, operators may decide also to employ other defense systems such as decoys, chaff, or flares. Engaging simultaneous attacking targets like swarms of small craft, aircraft, and drones or missiles can quickly overload current systems and operators.

A modern combat system based on AI could reduce or alleviate many of the challenges encountered in the detect-to-engage process. Large data sets are amassed from combat systems trials, fleet exercises, and missile tests that could be used to train an AI combat system. Machine perceptions gained from analysis of sensor data could be used to build an extensive knowledge base of target performance. Analogous to facial recognition, characteristics of the sensor data could be used to recognize and report contact types and identifications with a high level of confidence.

Deep learning from sensor technical data also could be used in signal processors to help optimize settings in dynamic environmental conditions, leading to more effective and thorough search and detection. From deep learning about the most minute details of each platform’s performance in friendly, neutral, and enemy orders of battle, the AI combat system could provide high confidence, quality combat identification and precise recommendations regarding weapons employment (both the weapon and the specific, best positioned shooter) against a hostile track.

Artificial intelligence enables continuous, simultaneous planning of multiple surface combatant missions, a key attribute for a future combat system to seamlessly transition from one mission to the next. Automated planning and control systems could help battle groups optimize ship and aircraft stationing in any given scenario to protect high-value assets, recommend optimal weapons employment to maximize probability of kill while minimizing inventory expenditure, and help reduce operator cognitive load in complex inter-domain and multi-domain operations.

Operator workload likewise could be reduced through intelligent interaction with the combat system using natural language processing. Instead of entering cryptic commands via keyboard to touch screen, operators might instead be able to issue simpler verbal commands like, “Siri, optimize the radar for land clutter in the northwest sector and alert me with the likely ID of any contacts closing from that sector.” Today, this command would take coordinated and thoughtful actions from several operators, require significant time and “button smashing,” and still may not result in optimized system performance.

Together these AI systems could revolutionize the performance and effectiveness of Navy combat systems. Disaggregated data from multiple sensors and platforms could be intelligently fused and translated into actionable information. AI could optimize ship and battle group sensor and weapons employment. Operators will benefit from AI that could significantly reduce cognitive load and provide intelligent recommendations and tactical advice in a natural language interface. Given the evolving and emerging threats to our fleet, the time is now to act to bring these new technologies to bear.

A Way Ahead

Today, no one in the defense industry or DoD yet knows how to build a system to do the things we have described, but there is a path to success. Part of the Defense Department’s “third offset” strategy was establishment of an organization to help innovation efforts just like this one. The Defense Innovation Unit-Experimental (DIUx) works in close collaboration with innovators in Silicon Valley and across the country will be needed to build a new AI combat system. [6] DIUx is working with commercial entrepreneurs and industry innovators who can apply their solutions to military problems, do rapid prototyping, and move forward quickly to testing and production.

In addition to DIUx’s connections and access to the Silicon Valley innovation community, the organization has also developed a new acquisition process that is built to move fast and attract new offerors who bring new ideas and solutions for defense problems. The process, called the Commercial Solutions Opening (CSO), pioneers new competition and contracting methods, and works in weeks or months rather than years or decades.

To bring this radical new technology to fruition, the Navy should launch a “Digital Manhattan Project” to create a working combat system prototype. DIUx and its CSO process are two enablers needed to launch this effort. This would require substantial collaboration between DIUx, Navy operators, the Naval Sea Systems Command’s Program Executive Office for Integrated Warfare Systems (PEO IWS), Space and Naval Warfare Systems Command System Center Pacific, and the Navy ballistic missile defense program office to define goals and envision how best to use the products and knowledge produced (see sidebar below: Creating a Digital Manhattan Project). DIUx would work with innovators and academia to identify innovative AI technologies, approaches, and critical non-defense community partners to assist in adapting commercial products for Navy combat system application. A CSO project can be developed and initiated in less than 90 days.

Using the CSO process, a rapid prototyping test bed could be established to test these new architectures and innovations (perhaps at the DIUx site in Mountain View, California, or at the nearby Naval Postgraduate School). The test bed should be used to systematically try new ideas, computing techniques and technologies, to figure out quickly how to apply these new assets in systems, and develop functional prototypes to solve real problems— to build, to test, and to learn. What we learn should be immediately fed into the Navy’s combat system development processes. Successful prototypes should be installed aboard ship as upgrades to current combat systems to provide additional intelligent capability as quickly as possible. Early experimental installations also would allow the continued collection of live data to teach and optimize the AI system, leading to rapid exploitation of progress.

This journey into the future will be exciting and transformational. The technology is sufficiently mature to begin experimentation and prototyping. The operational need is clear and becoming more compelling every day. We have the opportunity to get ahead of emerging and evolving threats, but in some areas, we have to scrap the conventional approach and quickly do more of the nerdy things we need to do to get the job done right. Organizations are in place to begin this work—the Navy and its extensive combat systems experience together with DIUx and its Silicon Valley innovators. What is needed is a sense of urgency and firm resolve to bring the Navy’s combat systems into the 21st century and beyond.

Like the Manhattan Project in World War II, we are faced with an existential threat. Also like the Manhattan Project, we have available a diverse, creative, and innovative community to draw on for solutions.

Time is the enemy. We must think big, move fast, and bet like we are going to win. Let’s move out now.

“We must shake off any vestiges of comfort or complacency that our previous advantages may have afforded us, and move out to build a larger, more distributed, and more capable battle fleet that can execute our mission . . . Time is of the essence.” -CNO whitepaper on future fleet, May 2017

 

Creating a Digital Manhattan Project

An effective prototype effort will pioneer a path to develop a set of new design paradigms for future combat systems. Such a system should have attributes such as:

  • Versatile and adaptive performance
    • Rapidly adaptable to new roles, missions, sensors, weapons, and networks
    • Flexible and assistive in new strategic/tactical situations
    • Native support to inter-domain and multi-domain operations
    • Cyber hard/Cyber immune
    • Platform agnostic
    • Born integrated
  • Artificial Intelligence architecture
    • All-source sensor and data fusion – capable of handling big data sets
    • Smart sensor coordination and contact identification support
    • Intelligent engagement scheduling and weapons selection
    • Unmanned systems employment
      • Flight planning, execution and control
      • Optimal tactical positioning for sensor/weapons coverage
      • Management of flight schedules
    • Natural language and smart messaging to reduce operator cognitive overload
    • Advanced displays and operator interfaces

 

Endnotes:

[1] Dave Majumdar, “The U.S. Military’s Greatest Fear: Russia and China are Catching up Fast,” The National Interest, 17 May 2016, nationalinterest.org/blog/the-buzz/the-us-militarys-greatest-fear-russia-china-are-catching-16242.

 

[2] Luke Swartz, “Overwhelmed by Technology: How did user interface failures on board the USS Vincennes lead to 290 dead?” xenon.stanford.edu/~lswartz/vincennes.pdf.

 

[3] David Kelnar, “The fourth industrial revolution: a primer on Artificial Intelligence (AI),” MMC Ventures, 2 Dec 2016, medium.com/mmc-writes/the-fourth-industrial-revolution-a-primer-on-artificial-intelligence-ai-ff5e7fffcae1.

 

[4] Kevin Kelly, “The Three Breakthroughs that have Finally Unleashed AI on the World,” Wired, 27 Oct 2014, wired.com/2014/10/future-of-artificial-intelligence/.

 

[5] Alex Castrounis, “Artificial Intelligence, Deep Learning, and Neural Networks Explained,” Innoarchitect, 1 Sep 2016, innoarchitech.com/artificial-intelligence-deep-learning-neural-networks-explained/.

 

[6] Cheryl Pellerin, “ Deputy Secretary: Third Offset Strategy Bolsters America’s Military Deterrence,” DoD News, 31Oct 2016, defense.gov/News/Article/Article/991434/deputy-secretary-third-offset-strategy-bolsters-americas-military-deterrence/. Fred Kaplan, “The Pentagon’s Innovation Experiment,” MIT Technology Review, 19 Dec 2016, technologyreview.com/s/603084/the-pentagons-innovation-experiment/.

 

Blog Update

Announcement

Categories

Tags

The Naval Institute Blog is on hold at the moment. Our plan is to move it to the Proceedings site and rename it “Proceedings Blog” in 2024. More information to follow soon!

Back To Top