Cognitive Lethal Autonomous Weapons Systems (CLAWS)
With the debut of lethal autonomous weapons systems (LAWS) in combat, opponents of LAWS have called on States to fast-track the creation of international law that either bans the use of these weapons or mandates meaningful human control over them. If LAWS are used more broadly in future combat, then the latter would ensure a check on the autonomous technology’s limitations, such as rigidness (inability to subjectively analyze situations and modify behavior to changing circumstances), non-explainability (inability to understand the machine’s decision-making processes), and potential biases. In addition to mitigating technology-based limitations, opponents argue that meaningful human control would also preserve the possibility for compassionate behavior and emotion in combat.
Neuromorphic Computing
While human emotion cannot be experienced by artificial intelligence (AI) and is therefore out of the scope of current technological remedy, an emerging AI may soon address LAWS’s operational limitations.[1] Specifically, neuromorphic computing, the next generation of AI, may allow LAWS to function in a manner more comparable to a human combatant. Where previous generations of AI draw inspiration from biology, neuromorphic computing actually mimics the functions of a brain, which allows the technology to support dynamic learning in the context of complex and unstructured data.
This shift to biologically accurate operations is made possible by moving away from the traditional computational architectures found in deep neural networks, generally known as the von Neumann architecture, and towards a neural network that operates similar to the brain. The latter functions through spikes of encoded information, and in simple terms, the brain-like function of these spiking neural networks (SNNs) operate in a manner analogous to a drum. Stated further:
Drums can respond with different and complex vibration states when they are stimulated, and they can be also understood on computational terms: input (hits), rules (physical laws, physical constraints such as material, tension, etc.), and outputs (vibration, sounds, normal modes). Indeed, the brain has many more similarities with a dynamical system as a drum than with digital computers, which are based on discrete states…. In abstract terms, drums are also “computing” and processing information, but this information processing is a dynamical reaction from external/internal stimuli more than a formal calculation process.[2]
Thus, by mimicking the processes of the brain “the various computational elements are mixed together and the system is dynamic, based on a ‘learning’ process by which the various elements of the system change and readjust depending on the type of stimuli they receive.”[3]
Advantages
This mode of computing is advantageous (with respect to conventional AI) for three main reasons. First, the spikes—i.e., the discrete events that take place at points in time—allow for faster propagation of information. This can also lead to the possibility of pseudo-simultaneous information processing when combined with an event-based sensor. Second, computing via spikes leads to increases in computational efficiency and decreases in power consumption. In comparison, traditional computations where information is repeatedly shuttled between functional units such as memory (MU), control processing (CPU), arithmetic/logic (ALU), and data paths can lead to the von Neumann bottleneck. Third, SNN’s energy efficiency allows for available processing power to increase as well. In other words, scaling up the neural capacity increases the ability to solve larger, more complex problems.
In sum, AI technology that is modeled to mimic the brain is better suited than conventional AI to adapt to context-specific situations. More specifically, emulating the principles of neural information processing enables the technology to function in a cognitive paradigm. This means that the technology can quickly plan, anticipate, and respond to complex and unstructured data.
By incorporating brain-like capabilities into technology such as LAWS these cognitive LAWS (CLAWS) could, in turn, employ more human-like discretion in targeting decisions. This is especially salient with regard to the law of armed conflict (LOAC) targeting principles of distinction, proportionality, and military necessity, which all require context-specific judgements. For example, the principle of distinction in Article 48 of Additional Protocol I requires that military operations be directed only at military objectives. Such distinctions are especially imperative in today’s urban battlefields where combatants and civilians exist side-by-side. In these circumstances the challenge lies in amending behavior based on the complex possibilities presented. This requires the decision maker to carefully observe conditions and adjust behavior in relation to unfolding information, a task that entails “on-site” learning and flexibility.
In the case of LAWS, which cannot “learn” dynamically, rigid decision-making capabilities may be problematic for a fully autonomous operation in unpredictable conditions. However, with CLAWS’s on-chip learning they could “interpret the features extracted from images, perceive and analyze multi-faceted situations during an attack, and adapt behavior based on the information gathered.” [4] This brain-like capability highlights CLAWS’s potential for making the context-specific decisions required by the targeting principles of the LOAC for fully autonomous operations.
Additionally, CLAWS could also be capable of probabilistic computing. This is important because the system’s rationale and decision-making processes could be accessible for review, which would allow for analysis regarding the system’s reliability and bias. Thus, CLAWS have the potential to “conduct complex decision making by managing, planning, anticipating, and adapting to unstructured battlefield environments, all with amplified efficiency and in an environment of reduced bias and increased transparency.”[5]
Limitations
While the future use of CLAWS seems promising, the neuromorphic technology necessary for CLAWS’ success is still in development. Furthermore, novel issues arising from the use of biologically realistic processors are still being addressed, such as an instability that is characteristic of a sleep-deprived state. Nevertheless, based on the mounting success of neuromorphic computing, it is likely that this technology will eventually be incorporated into specialized products, including weapons of future combat. With the prospective introduction of CLAWS, opponents’ technology-based objections to the use of LAWS may be overcome. However, their concerns regarding the preservation of emotion in combat would remain unresolved because CLAWS would be incapable of feeling emotion.
While the lack of emotion may seem trivial, respect and honor have been integral to combat and central to the warrior’s code of conduct. The United States Department of Defense Law of War Manual notes that “the principle of honor draws from warriors’ codes from a variety of cultures and time periods,” and that “[h]onor demands a certain mutual respect between opposing military forces.” Therefore, if respect is a fundamental aspect of war regulations, then ensuring its role in combat is critical. Solutions that account for this element of warfare will be imperative for the successful use of CLAWS.
Conclusion
Ultimately, the future of combat demands innovative proposals, and weapons systems such as CLAWS can provide pertinent solutions to increasingly technical and automatized warfare. By incorporating cognitive processes that mimic biology, not only can CLAWS effect a human-like discretion in targeting decisions, but their technological capabilities can also contribute to more exacting results. This amalgamation of intelligent and functional dexterity may ultimately yield an end result of unmatched performance in future warfare.
***
Carolyn Sharp is a law student at Brigham Young University. Carolyn focuses her research on the impacts of advanced technology on international law and the law of armed conflict.
***
Footnotes
[1] Carolyn Sharp, Status of the Operator: Biologically Inspired Computing as Both a Weapon and an Effector of Laws of War Compliance, 28 Rich. J.L. & Tech., no. 1 (2021).
[2] Camilo Miguel Signorelli, Can Computers Become Conscious and Overcome Humans?, Frontiers In Robotics and AI 1, 2 (2018).
[3] Neuromorphic Computing: From Materials to Systems Architecture, Report of a Roundtable Convened to Consider Neuromorphic Computing Basic Research Needs, 7 (U.S. Dept. of Energy 2015).
[4] Carolyn Sharp, Status of the Operator: Biologically Inspired Computing as Both a Weapon and an Effector of Laws of War Compliance, 28 Rich. J.L. & Tech., no. 1 (2021).
[5] Id.