Deep Learning: Measure Twice, Cut Once

Many in the industrial and defense communities are expecting artificial intelligence technologies to solve a host of problems. Others are concerned that adversaries investing heavily in these technologies will produce highly autonomous and adaptive weapons that will overmatch any known defenses. AI technologies are now enjoying their success because of two enabling developments. The availability of large amounts of appropriately labeled training data and the continued growth in sheer computing power permit the decades-old neural network technologies to reach surprising performance levels. These success stories beg answers to questions on the limits of performance. This paper describes artificial intelligence in its historical context of boom and bust cycles and discusses limitations of the technologies.