Challenges in the Use of Artificial Intelligence-enabled Systems in Modern Armed Forces

Challenges in the Use of Artificial Intelligence-enabled Systems in Modern Armed Forces Presented by: Prof. Dr. Prezelj Date: 12 June 2024 Introduction - Definition of Al: ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings, such as ability to reason, discover meaning, generalize, solve problems or learn from past experience, including the ability to adapt to new circumstances - Two types of Al: AGI and expert systems - First known use in Armed forces: 2016 US Dragon Lady Spy drone during a training exercise - Aims of this presentation - Arguments: • RESPONSIBILITY: a responsible authorization for the use of Al in armed forces and security services requires a thorough knowledge and investigation of the main application challenges in order to prevent various negative scenarios. Specifically, we argue that the main Al-related risks in this area are uncontrolled and unstoppable development of general Al, various ethical and legal risks, and operational and strategic risks. • COMPREHENSIVE: The range of these risks is so wide that it will be difficult to address them comprehensively. The wide range of risks identified and their extremely diverse nature show that regulating the military use of Al will be difficult and complex, requiring all disciplines of law, and that regulatory rules need to be applied at national, regional and global levels. • TIMING: The rapid development of military Al suggests that some risks are likely to be considered and regulated before any malicious military use of Al occurs, while some others will unfortunately only be regulated after the first instances of malicious and illegal use by some armed forces. - Desired goal of society regarding Al use: strike an appropriate balance between the potential benefits and risks - Debate on Al: • Boomers vs. Doomers A Broad spectrum of potential and actual use of Al in AF - Al = analytical enabler, disruptor and force multiplier - Tactical and strategic use - Offensive and defensive use - Dual use - Key element of Al = machine learning: learning from data without being explicitly programmed • Precondition is that system is trained by experts on a large sets of data • Important EXPERIMENTATION: only • experimentally proven and successfully applied Al systems in warfare will ended standard inventory of AF - Use of Artificial intelligence by Armed Forces and Intelligence Services • Intelligence, Surveillance and Multi- Domain Situational Awareness • Targeting by Autonomous Weapon Systems • Maneuverings and Other Actions by Military Autonomous systems • Cyber Warfare and Cyber Security Applications • AI-enabled Logistics • AI-based Training and Exercises Challenges In The Use Of Artificial Intelligence-Enabled Systems In Modern Armed Forces ```python self.target = None import numpy as np def get_position(self): # Mock drone API return self.position class Drone: def set_target(self, target): def_init__(self): self.target = np.array(target) self.position = np.array([0.0, 0.0, def move_towards_target(self): 0.0]) if self.target is None: return print(f"Current Position: direction = self.target - {self.drone.get_position()}") self.position def scan_for_target(self): distance = # Placeholder for target scanning logic np.linalg.norm(direction) potential_targets = [np.array([10, 10, 0]), if distance > 0.1: np.array([20, -5, 0])] self.position += direction / distance * 0.1 return else: potential_targets[np.random.randint(0, self.position = self.target len(potential_targets))] self.target = None def move_and_target(self): # Al Controller point = np.array([5, 5, 0]) class DroneAl: self.navigate_to_point(point) def_init__(self, drone): target = self.scan_for_target() self.drone = drone print(f"Target acquired at: {target}") def navigate_to_point(self, point): self.navigate_to_point(target.tolist()) self.drone.set_target(point) # Example Usage while self.drone.get_position().tolist() != drone = Drone() point: drone_ai = DroneAl(drone) self.drone.move_towards_target() drone_ai.move_and_target() Public Warnings Against The Use Of Artificial Intelligence - Groups of scientists and technologists (open letter 2015): warned against the Al arms race and the potential spread of lethal Al to terrorists and dictators. + called for a ban on offensive autonomous weapons beyond meaningful human control. - groups of employees at technological companies called against the production of weaponized robots and similar warfare technology. - Google consequently published its Al guiding principles, in which it pledged not to design or deploy weapons that cause injury to people, technologies that gather or use information for surveillance violating internationally accepted norms, and not to develop technologies that violate generally accepted international law and human rights, despite its continued cooperation with the US government. - International Campaign for Robot Arms Control and the Campaign to Stop Killer Robots were launched in 2009 and 2013 to mobilize nation states, the public and industry. - Several faith and interfaith declarations against autonomous weapons have been adopted, including one by the Catholic Church stating that it is fundamentally immoral to use a weapon that we cannot fully control. - a discussion was initiated on a possible new formal protocol to the UN Convention on Certain Conventional Weapons to improve the regulation of fully autonomous weapons, but the US, Russia and the UK objected. - Some observers labelled Al as a new weapon of mass destruction, as they saw similarities with the development phase of the atomic bomb. - the Austrian and Swedish governments, the Belgian Parliament and the European Parliament, also called for a ban on autonomy - weapons. 2024: Al systems have capacity for DECEPTION - CONCLUSION = premature introduction of Al systems and technology in military practice is problematic. - Many of the risks associated with the use of this technology stem from this problem. • Problem = new int.tregulations do not limit development and use in the field of Defense and Intelligence Implementation of Al In Armed Forces Will Be Difficult - Slower than expected, but still faster than any other previous technology - INSTITUTIONAL ABSORPTION BARRIERS: • Implementation of Al will require organizational and doctrinal changes, changes of command processes - HUMAN ABSORPTION BARRIERS: personal values and characteristics, religious orientations affect absorption and attitude towards Al: • Research results show: respondents from East Asia are less fearful of Al than Europeans, Muslims and Buddhists feel more fear of Al - Lack of talent identified: Al computer experts ready to work in AF - Lack of big training data to train Al models - CONCLUSION: implementation of Al in AF will be slower Uncontrolled and unstoppable development of general Al = worst case - Exceeding human performance: present AGI level = ant or cockroach = weak or narrow Al • first threat will come from those who will first use real AGI for their own narrow illegal interest, not from Al itself - Self-directed, self-replicating and self-improving beyond human control - Pursuing objectives that are not consistent with human interests: -> existential threat to human security • scenario of liberation of the first Al system from human control: likely because research on this can not be really limited • scenario of conflict between human society and Al systems, dominance of Al over human society • creation of a new Al civilization: movie scenarios: the Matrix the Skynet Ethical And Legal Risks - Limited Al capacity to understand the law of armed conflict, international humanitarian law and other legal basis - Accountability gap between the operators and Al systems. - Limited ability to make moral judgements. - Tendency to violate human rights and privacy (threat to privacy and human rights). Operational Risks - The issue of overconfidence in Al systems and the problem of surprising and incomprehensible decisions. - Problematic validity of Al-based recommendations or decisions. - Al outcomes and decisions based on narrow training experience. - The risk of accidental use and uncontrolled conflict escalation. - Vulnerabilities of Al systems. Strategic Risks - Lower use and violence thresholds. - The risk of easy proliferation to other malicious states, criminal and terrorist individual or collective actors. - Risky and difficult to control dual-use potential of Al technology. - The risk of global Al arms race and competition. - Al capability-related distrust among countries. - The risk of system mispositioning of Al-based decision-making. - The risk of increased police and intelligence state through the use of Al. Conclusion And Identification Of Regulative Needs - Regulation of Al will be as usual: one step lagging behind practice - Al systems will be first tested in practice and then become a part of standard use - We need to build national/regional/global Al governance system: • Comprehensive - all nations, • Include also companies and programmers, defense industry • Built around key risks: unpredictability od Al systems, easy • Proliferation of Al algorithms, dual use nature, low trust among nations AI CSBM measures - Application of Al in AF: according to DOTMPFLI principle = technical innovation requests also doctrinal, organizational, structural... changes = comprehensive sociotechnical change = RMA V