Artificial intelligence has seen groundbreaking developments in recent years that have yielded commercial and scientific applications affecting daily life around the world. In national defense, AI technology will advance — and even transform — warfighting. The Commandant of the Marine Corps emphasized this in his 2019 planning Guidance: Autonomous systems and artificial intelligence are rapidly changing the character of war. We have already seen these changes on today’s battlefields, but we are only at the leading edge of revolutionary changes. Our potential peer adversaries are investing heavily to gain dominance in these fields. We must aggressively research, innovate, and adapt to maximize the potential these offer while mitigating their inherent vulnerabilities and risks.
In this era of great power competition, it is imperative that the U.S. Marine Corps pursue AI rapidly and strategically. Marine Corps leadership must choose areas of exploration and investment wisely while simultaneously moving forward with needed enablers. Yet the technology, requirements and implementation associated with AI are complex. Proceeding without sufficient understanding risks wasting crucial time and resources. In the following pages, we provide a short overview of AI technology to give Marines Corps decision-makers a high-level understanding of key concepts.
WHAT IS AI?
The ultimate aim of AI is to solve problems by replacing human cognition with a combination of hardware and software. It is a continually evolving area of research focused on the development of new concepts, approaches and toolkits in pursuit of this goal. In the 1960s, early AI techniques enabled some of the first applications of natural language processing, symbolic reasoning and chess playing. In the 1990s, more advanced AI techniques showed their capabilities in IBM’s Deep Blue, the system that defeated chess champion Garry Kasparov in 1997.
Today, AI refers to modern methods such as machine learning. Machine learning encompasses a wide variety of techniques to detect patterns in data and to learn and make predictions from them. Machine learning is the source of most major AI advancements in recent years. Machine learning algorithms can be classified by how they obtain information about the data they receive. For example, in supervised learning, AI agents are given data that have been labeled with the type of information the agents must learn to discern. By contrast, in reinforcement learning, AI agents gather their own data and learn from it through trial and error. Deep learning is an important subcategory of machine learning that uses multiple computational layers, each of which is modeled on biological neural networks. The impact of this new generation of fast-learning algorithms has been amplified by substantial increases in computing power, the declining cost of digital storage, and the availability of large datasets on which the algorithms can “train.”
Today’s AI technology enables decision-making at levels of complexity and speeds that often greatly surpass human capabilities. Modern techniques have enabled groundbreaking advances such as:
- Machines that can defeat elite players at games that are computationally complex — notably Go, poker, and the videogame StarCraft II.
- Self-driving cars.
- Sophisticated image processing, such as facial recognition.
But AI is not a panacea. AI technology is ill-suited for some environments and problem types. It struggles with unpredictable environments and unstructured tasks that require common sense or an understanding of context. And even in situations where AI performs adequately, it may not offer meaningful improvements over legacy methods. AI today is generally well-suited for applications that require, for example, detection, pattern recognition, optimization and natural language processing.Download full report
- Document Number: DMM-2021-U-029844-Final
- Publication Date: 5/1/2021