AI for Defense and Intelligence

Patrick T. Biltgen is a defence-industry engineer who has spent years working on intelligence systems and analytics. His book AI for Defense and Intelligence is a working textbook for officers, analysts, and programme managers who need to understand what artificial intelligence actually does — not the marketing version, but the engineering reality of how these systems are built, trained, and broken.

The argument running through the book is that AI is not a single technology but a toolbox, and that the defence and intelligence communities need fluency with the toolbox before they can specify, buy, or deploy it. Biltgen treats machine learning the way an old field manual would treat a weapons system: here is what it does, here is what it costs, here is when it fails. He resists the framing of AI as either a panacea or a runaway threat, presenting it instead as a set of techniques with measurable strengths and well-understood failure modes.

The first half walks through the fundamentals — supervised and unsupervised learning, reinforcement learning, neural networks and transformers, natural language processing, computer vision, optimisation, and agent-based modelling. Each topic is grounded in problems an analyst would recognise: classifying vessels in satellite imagery, extracting entities from intercepted text, planning routes under constrained fuel and time budgets. The second half maps these techniques onto the intelligence cycle and the warfighting enterprise — sensor fusion across multi-INT collections, automated target recognition, signals analysis in cluttered electromagnetic environments, and the adversarial cases where labels are missing, datasets are poisoned, or an opponent is actively spoofing the model. Biltgen also examines the programme-management side: data pipelines, model evaluation, the test-and-evaluation regime defence acquisition demands, and how vendors such as Palantir, Anduril, and Helsing fit into the broader ecosystem.

Where the book sits in the field is closer to a reference than a thesis. Plenty of titles explain machine learning to a general audience, and plenty more debate the ethics of autonomous weapons. Biltgen’s contribution is the bridge between the two — the translation layer between machine-learning vocabulary and the doctrinal language of the intelligence community, with enough operational detail that a programme officer can read a vendor pitch and tell whether the claims are plausible. It is the kind of book that ends up annotated on the desks of acquisition officers and intelligence-systems engineers, less because it breaks new ground in AI and more because it speaks both languages at once.