Over the past decades researchers have tried to develop protocols to assess the current condition (diagnosis) and the potential for further expansion/distress (prognosis) of aging critical infrastructure. Among promising techniques, particular attention was given to quantitative microscopic procedures such as the Damage Rating Index (DRI). However, those techniques were found to be quite time-consuming and relied on the experience of petrographers able to properly judge their outcomes. It has been found that Machine Learning (ML) techniques might be used for efficient damage assessment in concrete. With the aid of modern software and some custom programming, petrographers could provide some “training” to the computer to diagnose concrete distress mechanisms. This paper discusses the potential use of ML to appraise damage in critical infrastructure. Validation is made with data obtained from laboratory test specimens with different amounts of damage from AAR. Preliminary results illustrate the promising character of ML in diagnosing AAR-affected concrete, which may also be suitable for other types of deterioration.