Research in the Intelligence Community in the Age of Artificial
Total Page:16
File Type:pdf, Size:1020Kb
Research Forum Using artificial intelligence in the IDF Computer Service Directorate. Photo: Bamahane website Research in the Intelligence Community in the Age of Artificial Intelligence Shmuel Even and David Siman-Tov By their very nature, intelligence communities are ideal customers for new information technologies. This article addresses two questions: How do new information technologies, primarily artificial intelligence, contribute to the intelligence community’s research activities? What challenges are posed by the assimilation of artificial intelligence in this field? The integration of artificial intelligence in strategic research may provide intelligence agencies with enhanced capabilities in helping leaders understand an emerging reality, detect changes of course early, and manage risks and opportunities. However, the path to achievement of these capabilities is replete with challenges. Keywords: intelligence research, intelligence assessment, war alert, situation assessment, decision making, strategy, artificial intelligence, machine learning Shmuel Even and David Siman-Tov | Research in the Intelligence Community in the Age of Artificial Intellligenc 37 Introduction 2. An artificial system developed in computer Artificial intelligence (AI) is a major evolutionary software, physical hardware, or other context step in the broad field of information that solves tasks requiring human-like technologies. The elements enabling AI perception, cognition, planning, learning, technology applications are the augmentation communication, or physical action. of computerization capabilities, growth in the 3. An artificial system designed to think or volume of information, better algorithms, and act like a human, including cognitive growth in investments (Grimland, 2018). AI-based architectures and neural networks. applications are increasingly integrated in daily 4. A set of techniques, including “machine life, and their impact can be expected to expand learning,” that is designed to approximate and intensify in many spheres (“How to Ensure a cognitive task. Artificial Intelligence Benefits Society,” 2020). The 5. An artificial system designed to act rationally, same is true with regard to AI-based applications including an intelligent software agent or in the security and intelligence systems. By their embodied robot that achieves goals using very nature, intelligence communities are ideal perception, planning, reasoning, learning, candidates for employing artificial intelligence, communicating, decision making, and since the majority of their activities involve acting. collecting enormous amounts of information (US Congress, 2019) from a variety of sources, processing data, conducting research, and formulating scenario- These definitions are partially based on the based assessments and predictions. similarity between the mode of operation and This article discusses information output of an AI system and human modes technologies in the age of artificial intelligence, of thinking. In this article, the achievement in the context of intelligence research in required of the machine will be examined intelligence communities, particularly strategic primarily according to the degree of its research. It will attempt to clarify how IT “intellectual capacity” to benefit humans capabilities (“the technology” or “the machine”) in achieving their goals, and in this context, can contribute to intelligence research benefit research entities in the intelligence activity and to the formulation of intelligence community, while exploiting the relative assessments, and what challenges are posed advantages of machines that are capable of by the assimilation of artificial intelligence in performing specific tasks at levels of accuracy, intelligence research. speed, volume, and complexity that far exceed the capabilities of the human mind. The Artificial Intelligence Concept The brief description below of the nature There is no fully accepted definition of the term of artificial intelligence is based on the article “artificial intelligence.” The following section is “A DARPA Perspective on Artificial Intelligence” based primarily on official United States national (Defense Advanced Research Projects Agency, security texts. According to the definition in the an agency of the United States Department of John S. McCain National Defense Authorization Defense responsible for the development of Act for Fiscal Year 2019, artificial intelligence is: advanced technologies) by John Launchbury, 1. Any artificial system that performs the Director of the Information Innovation tasks under varying and unpredictable Office at DARPA (Launchbury, 2017). His circumstances without significant human article presents four characteristics of artificial oversight, or that can learn from experience intelligence capabilities: and improve performance when exposed 1. The ability to grasp the environment outside to data sets. the machine (perceiving): the machine’s 38 Strategic Assessment | Volume 23 | No. 4 | October 2020 ability to detect, analyze, and respond to the sense that the researcher understands the information that it or systems connected to it logical cause-and-effect connections that are collect from outside of the computer, such as operating in the machine, and the source from a system that detects and analyzes vehicular which the machine took or received each item traffic data, images in an environment, and of data. However, this clarity is limited to a so on. specific deterministic process that the person 2. The ability to learn (learning): the system’s dictates through an algorithm and through ability to learn from examples and apply the the information that it feeds into the machine. knowledge to new circumstances. The capabilities at this stage (which have been 3. The ability to abstract (abstracting): for implemented in recent decades) are still of example, the ability to take knowledge considerable value today. For example, in a discovered at a particular level and apply DARPA project to augment cyber security, the it at a higher level. This ability enables the first-wave systems succeeded in helping detect creation of new meanings, but it requires a cyber security vulnerabilities. contextual capability. The second wave is now at the center of the discourse and action in particular areas of artificial intelligence. The second wave According to Hallman, the main challenge is includes applications of voice recognition, “taking the vast volumes of digital intelligence facial recognition, photo sorting, and more. that the CIA receives from around the world and The second-wave systems are highly capable transforming them into a digital, dynamic and of perceiving the environment outside the credible picture of the future.” Hallman adds that “intelligence, in this context, becomes almost a computer (using sensors and a link to big data). super power.” These systems are characterized by statistical learning and include the use of artificial neural networks that are characterized by deep 4. The clarity of causality (reasoning): for learning. Using this technology, an AI system example, to what extent can a human user knows how to identify a phenomenon based of a machine understand the correlation on characteristics that it learned independently between the raw data and the machine’s from examples (such as from a series of imaging conclusions. This is a critical ability for of a disease) and not solely according to planning and decision making. characteristics that the researcher input into the According to DARPA, the development of machine. DARPA uses second-wave technology, artificial intelligence may be differentiated by inter alia, to analyze the spread of cyberattacks three waves: and to gain autonomous tools. Despite all of The first wave is the programmed these advantages, second-wave systems engage ability to process information. Experts take in specific tasks and have minimal capability knowledge that they have of a particular of presenting reasoning. They do not have the subject, characterize it according to rules ability to give an explanation (explainability).1 that are computer-compatible, and in turn, Another limitation is that second-wave systems the machine processes the data according need an enormous amount of data in order to to algorithms that they wrote and generates learn,2 and they are not immune from errors. output according to a defined pattern. Examples The third wave is still at its initial research of this are logistics software, chess software, and development stages. It is supposed and tax computing software. At this stage, to overcome some of the limitations of the the capabilities of artificial intelligence are earlier waves and create additional capabilities. characterized by high “clarity of causality,” in Sources in DARPA believe that the third wave Shmuel Even and David Siman-Tov | Research in the Intelligence Community in the Age of Artificial Intellligenc 39 will be designed using contextual models that Thus, first-wave artificial intelligence is will enable, inter alia, the design of systems still relevant; the second wave provides high that know how to learn from a limited number capabilities in particular areas; and the third of examples, how to provide explanations for wave reflects expectations of advances in the their results, and how to create new meanings coming decade. It certainly will not be the last from data (the ability to abstract). wave. Table 1. Characteristics of AI technologies Situation Perceiving Learning