Abstract This article is the first to explore the limitations of artificial intelligence in strategic decision-making and in preventing strategic surprise. Using well-known cases, it examines repeated difficulties involved in the collection, analysis, and dissemination of intelligence. Given collection challenges, corrupted data, legal limitations, classification barriers, and spoofing by adversarial AI systems, the article shows that adequate data to avoid surprise are often unavailable. Nor will AI eliminate the dangers of misapprehending information or the politicization of intelligence. The article then considers the theoretical limitations on the ability of computers of any level of power to predict the future. The future is random; any system in a given state can produce multiple future states. Consequently, even in a closed laboratory system, the ability to predict is subject to error and degrades rapidly over time. Unavoidable observational error, imprecision, and incompleteness create further difficulty. In strategic decision-making, these difficulties are compounded. AI will enhance strategic planning and improve short-term political forecasting within a margin of error, but it cannot be relied on for judgment, and it will not eliminate strategic surprises.
Joel Brenner (Thu,) studied this question.