1

Introduction

Having made steady progress since its founding in the 1950s, the AI field has now accelerated to the forefront of computer science, propelled by super-fast microprocessors, availability of big data, and our supreme appetite for automating human cognitive tasks.  AI has long suffered from hype and overpromising.  We can avoid the hype by grouping AI machines into seven categories by their learning power.

2

Automation

Traditional automation defines machines with fixed functions.  Familiar examples include vending machines, automatic restaurants, containerized shipping, automation, autopilots, and even bureaucracies.  AI is transforming automation by providing machines that learn new functions.  Cognitive tasks transferred to machines often seem unintelligent once automated.  The work that automation is supposed to free up for humans often seems elusive.  Automation is valuable but only for tasks amenable to its tradeoffs: scale and speed versus for irregular substructure.

3

Logic

Expert systems were among the first machines aimed at learning the skills of experts.  They encode rules described by experts into logic formulas in a database so that a machine can follow the same logic and solve the same problems.  However, because human experts take many actions not describable as rules, no software expert system has become an expert.  Systems based on logical deductions were once the core of AI.  With the rise of statistical learning methods such as neural networks and large language models, rule- and logic-based approaches in AI have received much less attention.  Even so, much research focuses on combining rule-based and statistical methods to achieve AI that can generate verifiable software, make scientific discoveries, and explain its reasoning.

4

Neural Networks

Recognizing faces in images is a human function that we do not know how to describe with rules.  With a database of 100 million labeled images, we can teach an artificial neural network to name the faces when shown the images.  Unfortunately, these networks cannot explain what they do and are very sensitive to small, pixel-level changes in images.  Artificial Neural Networks (ANNs) have been commercialized, shifting innovation from model construction to integration into other systems.  The Model Context Protocol (MCP), a new tool, enables agents built around ANNs to coordinate with each other.

5

Reinforcement Learning

Unlike supervised learning, where a neural network is told the correct outputs to produce, in reinforcement learning the neural network learns by trial and error.  The AlphaGo machine learned grandmaster GO in 13 days by playing against itself while reinforcing moves that led to wins.  Most LLMs are fine-tuned using reinforcement learning to align LLM outputs with human concerns.  Reinforcement learning is finding many military applications as well.

6

Human Machine Teaming

When IBM Blue beat him in 1997, Chess grandmaster Garry Kasparov invented a new kind of chess played by human-computer teams.  The teams beat the best machines.  Finding ways to use machines to augment rather than replace human intelligence is a central question in AI.  Human–machine teaming blends human judgment with machine speed, enabling well designed systems to outperform either partner alone. Effective teaming depends on allocating tasks to the partner best suited for them, maintaining shared understanding, and ensuring transparent, trustworthy interaction. Ultimately, intelligent systems succeed not by replacing humans but by integrating their complementary strengths.

7

Aspirational AI

Machines that can carry on intelligent conversations, think, understand, create, care, be self-aware, or be sentient are well beyond our current understanding.  Nonetheless, such speculations have inspired perseverance in the search for intelligent machines.  AI Conservatives worry that AI will dramatically harm people while AI Liberals are confident AI will dramatically benefit people.  The arguments on both sides contain many weaknesses.  Conservatives often claimed that AI could not do things that later it did, and they have not explained how brains could be fundamentally different from machines.  Liberals often overrate the few accomplishments of AI and brush over its failures.  Both sides fail to consider central problems such as failure to include all relevant human knowledge, difficulty of development and testing, lack of transparency, and popularity with totalitarian countries.

8

Vision

Automatically identifying persons and objects in images has been a long quest back to the 1950s.  A breakthrough came around 2012 with Convolutional Neural Networks and powerful parallel computers called GPUs.  Computer vision capabilities exploded.  This kickstarted a revolution in machine learning and artificial intelligence, leading eventually led to the current boom in Large Language Models.

9

Logistics

Moving materials from suppliers to recipients is long standing concern in all sectors, civilian, government, and military.  Military supply networks are in constant flux and are under constant threat.  AI planning tools are finding their way into logistics, where they help plan routes, specify stockpiles and depots, and project where supplies will be need in the near future.  They are becoming indispensable for modern supply networks, which are in constant change and evolution.

10

Robots

Robots have fascinated human beings since well before the electronic computing age.  Remember the Mechanical Turk, a robot invented in 1770 to play chess?  It was quite good.  Eventually it was exposed as a hoax –a human chess player was hidden inside the cabinet.  But even so, it inspired interest in the question of whether a grandmaster chess-playing robots could be built.  It took two hundred years to reach a positive answer to that question.  Modern mobile robots, which combine AI, Mechanical Engineering, and Software Engineering, are even more challenging.

11

Managing AI Projects

The DoD uses AI for many tasks, such as predicting unscheduled maintenance of deployed weapon systems and determining the cost drivers of weapon systems still under development.  However, the management of AI projects is vastly different from traditional project management and the traditional management of software projects.  Because AI systems depend heavily on constantly changing data, life cycle of projects to build them is not based on formal, structured, and sequential phases and milestones.  As the data change, the model must be adapted to ensure responsible and trustworthy AI.

12

Cyber security

Current approaches to artificial intelligence (AI) pose significant cybersecurity challenges because models are being built and deployed on software, hardware, and networks that were never designed for security or adequately secured after deployment.  Adversarial attacks can target the infrastructure as well as the AI itself.  LLMs provide examples of such vulnerabilities.  For example, sensitive data provided to a seemingly private LLM interface can be surreptitiously leaked into the Cloud.  New methods for securing AI infrastructure are essential for safe and trustworthy AI systems.

13

Intelligence

AI has become imperative for military intelligence.  The imperative pressures come in three dimensions: information overload (quantity), uncertainty of information pedigree (quality), and pace of AI-enabled warfare (Tempo). Opportunities for using AI to improve intelligence include faster data processing, decision support, and autonomous surveillance. Challenges to AI for military intelligence include adversarial AI, atrophy of human Intelligence information, and increasing difficulties in ascertaining what is true.

14

Targeting

Selection of military targets has always been a complex process.  AI is accelerating every aspect, from finding and tracking to engagement and assessment.  AI brings new risks, limitations, and challenges to the critical and enduring role of human judgment. This process is a clear example of how data and algorithms are reshaping modern warfare.

15

Ethics

AI has raised numerous ethical dilemmas.  They can be approached by distinguishing between malum in se (intrinsically wrong) and malum prohibitum (wrong because prohibited).  Are AI’s moral problems inherent or contextual?  Thought experiments like the “racist soldier” and “sociopath killer” shed light on whether reasons and intentions matter morally in AI decision-making, particularly for lethal autonomous weapons. Key issues are Sparrow’s “responsibility gap” argument and the “opacity problem” where AI decisions become unintelligible to humans. The deepest concern raised is the “lowering the threshold” problem: that AI makes warfare so easy and low-cost it will tempt nations to fight unjust wars more frequently, with no clear ethical solutions.

16

Strategy and Power

AI is reshaping military power and geopolitical competition by accelerating decision-making, enabling autonomous systems, and transforming cyber and information operations. The United States and China are pursuing starkly different AI pathways: China’s centralized, party-driven model treats AI as a tool for political control and military dominance, while the U.S. relies on a private sector that advances AI primarily for commercial gain. These divergent political cultures yield sharply different approaches to innovation incentives and civil-military cooperation. Global power struggles have shifted toward economic-political issue including semiconductors, export controls, rare-earth dependencies, and competing governance blocs.  AI’s speed, autonomy, and attribution challenges complicate deterrence and raise escalation risks in future conflicts.  Projects like Maven illustrate the resulting friction between U.S. defense needs and tech-sector ethics.

17

Wargaming

Wargaming is an old military practice that seeks to understand how parties immersed in military operations will respond to fast changing and unpredictable field conditions.  AI is reshaping military wargaming by accelerating design, modeling adversaries, and sharpening analysis while highlighting the risks of imaginary precision and the enduring role of human game makers and players.

18

Al in DOW

Motivated by a concern to deal realistically with newly emerging great powers competition, the US Department of War (formerly known as DOD) has issued various strategy documents for AI research, implementation, and industry cooperation.  AI is not new to the DOW.  Its long history of AI research and adoption of AI-enhanced capabilities traces back to the 1960s. This talk describes how AI is being leveraged in Great Power Competition, clarifies what is meant by AI-enhanced systems for the DOW, summarizes U.S. national strategy and policy on AI for defense, provides AI innovation examples within the DOW, points out how AI is being applied throughout the defense enterprise, and enumerates some of the dilemmas encountered by the DOW in adopting AI.

19

New Face of War

In 1979 John Keegan published a book, The Face of Battle, in which he analyzed the practical mechanics of battle and how they affect outcomes as much or more than strategy.  AI is radically transforming the space of possibilities available to commanders and warfighters.  It emphasizes small, networked, distributed, swarming, autonomous agents over large platforms.  AI also adds a new dimension, where the agents have their own decision-making authority.

20

Trust, Truth, Hallucinations

Geopolitical conflict today engages in actions that gain ground for the aggressor but fall short of starting a war.  The decision systems to plan and execute actions depend increasingly on AI.  AI systems can make serious mistakes if attacked by an adversary, if trained on data containing unseen biases, or if dependent on fragile AI technology.  Trustworthy AI is crucial not only for success but to avoid mistakes that precipitate war.

21

Hype, futures

This course has sought to cut through AI hype by exposing the base principles of AI.  Although we currently have no intelligent machines, we do have seven varieties of machines that can rapidly learn to do complex human tasks.  These machines have produced significant advances and vulnerabilities in important domains such as vision, robotics, natural language processing, and cyber security.  The quest to make these machines reliable and secure has unearthed a host of hype-fueled dilemmas that implementers must face.