# Strategy Is Not Conscience: The Moskovitz Perspective, Artificial Intelligence, and the Mo **By:** Dr. James Giordano **Published:** 2016-06-17T00:00:00+00:00 **Source:** [Institute for National Strategic Studies](https://inss.ndu.edu/news/Article/4628347/strategy-is-not-conscience-the-moskovitz-perspective-artificial-intelligence-an) --- # News ![Airman 1st Class Patrick Nunn, an airfield systems technician with the 1st Special Operations Support Squadron, examines a circuit board at Hurlburt Field, Fla., June 17, 2016. Airfield systems technicians install, remove and maintain navigation and air traffic control ground-to-air radio systems. (U.S. Air Force photo by Airman 1st Class Joseph Pick)](https://media.defense.gov/2026/Oct/20/2004016970/1920/1080/0/261020-D-EG629-1001.JPG) 160617-F-UQ958-020 Airman 1st Class Patrick Nunn, an airfield systems technician with the 1st Special Operations Support Squadron, examines a circuit board at Hurlburt Field, Fla., June 17, 2016. Airfield systems technicians install, remove and maintain navigation and air traffic control ground-to-air radio systems. (U.S. Air Force photo by Airman 1st Class Joseph Pick) # Strategy Is Not Conscience: The Moskovitz Perspective, Artificial Intelligence, and the Moral Imperatives of Military Command and Control ### Introduction: The Importance of the Moskovitz Perspective In his timely and provocative essay, The Danger of Unregulated Artificial Intelligence: A Commentary on the Instantiation of Strategy Over Conscience, Dr. Peter Moskovitz articulates a fundamental distinction that should be central to any prudent consideration of artificial intelligence (AI), particularly in military and national security applications: the distinction between the computational optimization of strategy and the genuinely moral deliberation required to determine whether a strategically advantageous action is ethically justifiable. Moskovitz’s argument is both straightforward and profound. Artificial intelligence can increasingly identify, calculate, simulate, and execute strategies that maximize particular objectives. Yet such computational capability neither establishes nor necessarily entails moral conscience. The capacity to determine what can be done, or even what might most effectively accomplish a specified goal, must not be confused with the ability to determine what ought to be done. This distinction is especially consequential in military contexts, where decisions frequently involve competing obligations, uncertain intelligence, compressed temporal horizons, and the possibility of irreversible consequences to individuals, populations, and international stability. I regard Moskovitz’s essay as an important contribution to the expanding discourse about AI safety, governance, and the human responsibilities attendant to increasingly autonomous computational systems. Indeed, I would argue that its greatest value is not merely its caution regarding the existential risks of unconstrained AI, but its illumination of an immediately relevant problem: How can we employ increasingly capable AI systems to augment military decision-making without allowing computational strategy to displace human moral agency, authority, and accountability? This question is neither hypothetical nor peripheral. It is becoming central to the character, conduct, and ethical legitimacy of contemporary and future warfare. ### I. Strategy, Conscience, and the Limits of Computational Rationality Drawing upon Anatol Rapoport’s seminal distinction between strategy and conscience, Moskovitz argues that AI can be programmed to recognize and operationalize ethical rules without necessarily possessing the moral intuitions, lived experience, or consciousness that underlie human moral judgment. His argument is strengthened by consideration of the neurobiological foundations of consciousness and moral experience, including our collaborative work with Kathinka Evers examining oscillatory theories of consciousness and the limitations of reductively computational accounts of subjective experience. While I would caution against treating the noncomputational nature of consciousness, or the categorical impossibility of machine consciousness, as definitively established scientific conclusions, I concur with Moskovitz’s essential operational premise: We have no sufficient grounds to attribute genuine moral conscience to contemporary AI systems, and therefore no justification for treating their outputs as independently authoritative moral judgments. Importantly, even if future machines were to demonstrate forms of consciousness, such capacities would not automatically confer moral wisdom, responsibility, or legitimate authority to make decisions regarding the use of force. Human moral judgment is itself neither infallible nor universally consistent. It is shaped by neurobiological predispositions, individual experience, social norms, cultural traditions, institutional obligations, and contextual exigencies. Nevertheless, human decision-makers exist within systems of interpersonal responsibility, legal obligation, professional duty, and potential sanction. AI systems, by contrast, can optimize specified objectives without independently appreciating the moral significance of the means employed or consequences produced. This distinction becomes particularly dangerous when the operational objectives of an AI system are narrowly defined, inadequately bounded, or inadvertently misaligned with broader human values and institutional obligations. An algorithm may correctly identify a particular course of action as maximizing tactical advantage while failing to adequately represent its implications for proportionality, civilian protection, escalation, or longer-term strategic legitimacy. Thus, a computationally optimal solution may be tactically effective, strategically counterproductive, ethically unacceptable, and legally impermissible. Moskovitz’s warning is that strategic efficacy must never be mistaken for moral legitimacy. ### II. Military AI and the Risk of Algorithmic Abdication The accelerating integration of AI into military intelligence, surveillance, reconnaissance, targeting support, logistics, cyber operations, autonomous systems, and command-and-control architectures offers considerable potential advantages. Such systems can integrate heterogeneous information, identify patterns obscured by complexity, generate probabilistic assessments, and facilitate decisions at speeds and scales that exceed unaided human capability. These capabilities may enhance situational awareness, improve force protection, reduce cognitive burden, and, when appropriately designed and employed, decrease errors and unnecessary harm. Yet the very attributes that make AI militarily attractive also create novel vulnerabilities. The danger is not simply that AI may become increasingly autonomous. Rather, it is that human operators and commanders may progressively relinquish meaningful judgment to computational systems because those systems appear faster, more precise, more comprehensive, or ostensibly more objective. This phenomenon, often described as automation bias, may be compounded by algorithmic opacity, institutional pressures for accelerated decision-making, and the competitive dynamics of military technological development. I would characterize this as the risk of algorithmic abdication: the gradual displacement of human deliberative authority by machine-generated recommendations that become functionally determinative, even when formal human approval remains nominally required. Such displacement need not occur through malicious machine intent. It may emerge through entirely ordinary organizational processes in which human decision-makers increasingly defer to outputs they cannot adequately interrogate. In this respect, the Moskovitz Perspective provides a particularly useful corrective to technologically deterministic conceptions of future warfare. The issue is not whether AI can improve military performance. Clearly, in numerous circumstances, it can. The issue is whether improvements in operational performance are accompanied by commensurate improvements in human understanding, oversight, and ethical control. Absent such safeguards, greater computational capability may paradoxically produce diminished human agency. ### III. SYNTHComm: Synthesizing Computational Capability and Human Command It is precisely this challenge that informs our work developing the Synthesized Command and Control (SYNTHComm) model. SYNTHComm is intended to provide a conceptual and operational framework through which human cognitive, ethical, and command capabilities can be integrated with the analytical, predictive, and decision-support functions of increasingly sophisticated AI systems. Rather than conceptualizing command and control as either exclusively human-directed or progressively machine-dominated, SYNTHComm approaches the military decision environment as a dynamic, interactive system in which human and computational capabilities are deliberately coordinated according to their respective strengths, limitations, and responsibilities. The model is predicated upon several essential considerations. First, AI should augment, rather than presumptively replace, human judgment in consequential military decisions. Computational systems may generate courses of action, model probable consequences, identify uncertainties, and assess operational alternatives. However, decisions involving moral and legal responsibility must remain subject to appropriately situated human authority. Second, the human-machine relationship must be understood as dynamic rather than static. Different operational contexts require different distributions of computational assistance, human deliberation, and command intervention. The degree of permissible autonomy should therefore be contingent upon mission characteristics, system reliability, operational uncertainty, and the magnitude and reversibility of potential consequences. Third, effective oversight requires more than the ceremonial presence of a human decision-maker. Commanders must possess sufficient understanding of system capabilities, limitations, confidence estimates, and failure modes to meaningfully evaluate AI-generated recommendations. Fourth, command architectures must preserve opportunities for human intervention, correction, suspension, and termination of AI-enabled actions when circumstances depart from validated operational parameters. Finally, responsibility must remain traceable across the human-machine enterprise. The incorporation of AI into decision processes must not create ambiguity regarding who authorized an action, upon what information, under which constraints, and with what foreseeable consequences. These principles are directly relevant to Department of Defense Directive 3000.09, Autonomy in Weapon Systems, which establishes policy and requirements intended to ensure that autonomous and semi-autonomous weapon systems are developed and employed with appropriate levels of human judgment over the use of force. The directive also emphasizes rigorous verification, validation, testing, evaluation, and the capacity of systems to operate consistently with commanders’ and operators’ intentions. SYNTHComm offers a potentially useful means of translating these requirements from policy into operational architecture. It does so by treating human judgment not as an external safeguard appended to an otherwise autonomous system, but as a constitutive element of the command-and-control process itself. Importantly, SYNTHComm should not be interpreted as a claim that human participation automatically guarantees ethical outcomes. Human commanders remain susceptible to cognitive bias, fatigue, misinformation, institutional pressures, and moral error. Rather, the model seeks to establish conditions in which the complementary capabilities of humans and machines can be integrated while preserving identifiable responsibility and meaningful opportunities for correction. The objective is neither human supremacy over computation as an abstract ideological position nor unqualified confidence in algorithmic performance. It is the development of a more capable, accountable, and ethically defensible command enterprise. ### IV. From the Moskovitz Perspective to the Ethics of Military AI Moskovitz’s central admonition—that strategic calculation must not supersede conscience—should also inform the ethical architecture of military AI development and employment. In my view, this requires an approach that is structurally grounded in deontological obligations and rule-based constraints, while remaining functionally attentive to consequences, proportionality, and the realities of operational decision-making. Such an approach recognizes that certain obligations, including compliance with the law of armed conflict, cannot simply be discounted because their observance reduces tactical efficiency. At the same time, military decisions necessarily require consideration of anticipated consequences, competing risks, and potential harms. AI may substantially assist in estimating such consequences, but it cannot independently establish the legitimacy of the moral and legal priorities governing their evaluation. This is particularly important when considering AI-enabled targeting, automated threat identification, distributed autonomous systems, and machine-speed engagements in which decision intervals may be compressed to milliseconds. Under these conditions, ethical governance cannot depend exclusively upon a human making a final decision at the moment of engagement. Rather, human judgment must be exercised throughout the system lifecycle: in defining mission objectives, establishing operational boundaries, selecting permissible targets, validating system behavior, authorizing deployment, monitoring performance, and determining circumstances requiring disengagement. This suggests that meaningful human judgment is not simply a momentary intervention, but an enduring property of a responsibly designed and governed military enterprise. Moreover, the competitive imperative to accelerate AI development because adversaries are doing likewise must not become an unrestricted justification for abandoning prudent safeguards. Moskovitz correctly identifies the danger of allowing competitive strategy to override conscience. However, prudence does not require technological paralysis. Rather, it demands disciplined innovation in which operational advantage and ethical responsibility are treated as mutually reinforcing requirements. Indeed, systems that cannot be adequately understood, bounded, controlled, or trusted may ultimately create strategic vulnerabilities greater than the advantages they promise. ### V. Five Recommendations for the Department of War The Moskovitz Perspective, considered together with the SYNTHComm framework and the requirements of DoD Directive 3000.09, suggests five priorities for strengthening the development, governance, and operational employment of military AI. #### 1\. Establish a Doctrine of Human Moral and Command Authority The Department should explicitly distinguish computational recommendation, operational authorization, and moral-legal responsibility across AI-enabled military systems. Doctrine should specify the categories of decisions requiring direct human authorization, the conditions under which bounded autonomy is permissible, and the responsibilities that cannot be delegated to computational agents. Such doctrine should extend beyond weapon platforms to encompass AI-enabled intelligence analysis, targeting support, and operational planning where algorithmic outputs may materially influence decisions involving force. #### 2\. Develop and Empirically Evaluate SYNTHComm-Based Command Architectures The Department should sponsor interdisciplinary research, simulation, tabletop exercises, and operational experimentation to evaluate SYNTHComm as a framework for human-AI integration. These efforts should examine the allocation of decision authority, operator comprehension, automation bias, uncertainty communication, system controllability, and performance under adversarial manipulation. Evaluation should assess not only mission effectiveness and speed, but also ethical decision quality, legal compliance, escalation risk, and the preservation of accountable human command. #### 3\. Institute Continuous Ethical, Legal, and Operational Validation AI-enabled military systems should undergo iterative evaluation throughout development, deployment, and modification, rather than relying exclusively upon initial certification. Testing should include conditions of degraded information, adversarial deception, unexpected environmental changes, model drift, and conflicts between mission optimization and established legal or ethical constraints. Independent multidisciplinary review teams should be empowered to identify unacceptable risks and recommend modification, restriction, or suspension of systems that fail to maintain validated performance. #### 4\. Establish a Military AI Cognitive and Ethical Readiness Program Human commanders and operators require education and training commensurate with the capabilities and limitations of the systems they employ. The Department should establish training that integrates AI literacy, cognitive science, neuroethics, operational law, and decision-making under uncertainty. Particular emphasis should be placed upon recognizing automation bias, interrogating algorithmic recommendations, understanding confidence and uncertainty, and exercising informed authority to reject or terminate machine-generated courses of action. Such education should be integrated into professional military education, command preparation, and recurring operational certification. #### 5\. Advance International Norms for Accountable Military AI The United States should lead allied and multinational efforts to establish operationally meaningful expectations for human responsibility, system testing, transparency, and risk management in military AI. These efforts should seek practical convergence on safeguards against unintended escalation, uncontrolled autonomous engagements, and the delegation of consequential decisions to systems lacking adequate human oversight. Importantly, international engagement must be accompanied by credible domestic implementation. The United States cannot effectively advocate responsible military AI governance without demonstrating that such principles are embedded within its own development, acquisition, doctrine, and operational practices. ### Conclusion: Preserving Conscience in an Age of Computational Power Peter Moskovitz’s essay offers a necessary reminder that the increasing sophistication of artificial intelligence does not eliminate the distinction between strategic intelligence and moral wisdom. Machines may become extraordinarily capable at calculating how to achieve objectives. Yet determining which objectives are legitimate, which means are permissible, and which consequences are morally tolerable remains an obligation that military institutions cannot responsibly relinquish. The central danger is not merely the possibility that AI may eventually acquire capabilities beyond human comprehension or control. It is also the more immediate prospect that humans may voluntarily surrender their deliberative responsibilities to systems whose apparent competence obscures their fundamental limitations. SYNTHComm represents one approach to avoiding this predicament by integrating computational power with informed, accountable human command. Its value will ultimately depend upon empirical validation, institutional adoption, and the willingness of military leaders to regard ethical governance as intrinsic to operational effectiveness rather than as an impediment to technological advancement. The Moskovitz Perspective thus presents both a warning and an opportunity. The warning is that strategy divorced from conscience can produce actions that are instrumentally successful yet morally and strategically disastrous. The opportunity is to develop military AI systems and command architectures that enhance human capabilities without diminishing human responsibility. The future of military artificial intelligence should not be determined solely by what machines can accomplish, but by what human beings can prudently, legitimately, and ethically authorize them to do. That is not merely a philosophical preference. It is an operational necessity, a strategic imperative, and a moral obligation.