AI solves military decision-making---being adopted now, effective implementation solves
Matej Tonin 19, current Slovenian Minister of Defense, was an active member of the NATO PA from 2012 to 2020. He served as rapporteur and Chairperson of the Assembly's Sub-Committee on Technology Trends and Security and as Deputy Head of the Slovenian delegation to the NATO PA. "ARTIFICIAL INTELLIGENCE: IMPLICATIONS FOR NATO’S ARMED FORCES" Report from the Sub-Committee on Technology Trends and Security. October 13. https://www.nato-pa.int/download-file?filename=/sites/default/files/2019-10/REPORT%20149%20STCTTS%2019%20E%20rev.%201%20fin-%20ARTIFICIAL%20INTELLIGENCE.pdf //pipk
10. AI is already becoming a reality in armed forces around the world. All modern armed forces are, at a minimum, thinking through AI’s implications, including the ethical and legal implications, and many are introducing concrete AI solutions (see also Section IV). AI holds enormous potential for the defence sector but also presents a set of technical and non-technical challenges. Moreover, the strategic implications of AI remain unclear.
11. Tackling all opportunities, challenges, and uncertainties in depth would go beyond the scope of this report (see also Box 4). Therefore, this section delves into two key areas of opportunity (information and decision support as well as robotic autonomous systems), crucial technical and non-technical challenges, including the debate about lethal autonomous weapons systems, and some potential strategic implications.
A. INFORMATION AND DECISION SUPPORT 12. Humans generally act in an environment of incomplete information. Sometimes, plenty of information is available. In military and strategic affairs, this is decidedly not the case. Political and military leaders must act in the well-known ‘fog of war’. Information and decision support by AI systems is thus of high interest to military and strategic decision makers. Such systems can substantially increase both the pace and the quality of the processing, exploiting, and disseminating of information, as well as of human and machine decision-making.
13. In military terms, AI can considerably boost the speed of analysis and action of humans and machines. AI-enabled information and decision support systems can, for example: - vastly improve the reaction times of defensive systems against fast-acting weapon systems, such as hypersonic missiles, cyberattacks, or directed-energy weapons; - deliver actionable information faster to decision makers, which could potentially deliver a decisive edge on adversaries; - quickly discover cyber intrusions by detecting evasive malicious codes or by scanning for suspicious patterns of behaviour rather than for specific code; and - help identify attempts to manipulate citizens through disinformation campaigns.
14. AI can improve the quality of the decision making of machines, but also, and perhaps most importantly, of humans. AI’s ability to sift through today’s data-rich environment and communicate findings in a compelling manner is crucial in this respect and will become ever-more important. While human resources currently allow for the processing of, at best, 20% of the information produced today, this percentage could go down to a mere 2% (Villani, 2018). To put it in the words of an exasperated British officer, armed forces are already “swimming in sensors, drowning in data, [and] starving for insight” (White, 2019). AI solutions can help by, for example: - providing better visualisation and interpretation of data (Killion, 2018); - automatically extracting objects of interest from data feeds for follow-up actions, for example from surveillance video or satellite imagery (CRS, 2018); - establishing ‘common operating pictures’ from information which emanates from very diverse sources, arrives in very different formats, and is often redundant or incomplete (Killion, 2018); - highlighting abnormalities for follow-up investigations by comparing data points with previously developed normality models; - extracting ‘weak signals’ which do not seem alarming by themselves but may be highly significant if linked with other data (Mercier, 2018); - suggesting a menu of appropriate options and describing each option’s likely effects (Van den Bosch and Bronkhorst, 2018); and - provide insights into adversarial behaviour through foreknowledge (Demchack, 2018).