Summary of Key Points
The central argument of this article is that while AI can assist military strategists and weapon systems, it can never assume the role of a military commander. The reason lies in the inherent nature of AI, which is limited to “backward-looking” data fitting and lacks the adaptability, creativity, and ability to handle moral and legal ambiguities inherent in human intelligence. Final decisions must always be made by humans. Moreover, human commanders working with AI need to possess enhanced skills—both to outwit their opponents and to critically evaluate the AI-generated solutions.
I. AI Cannot Be a Commander: Humans Can Evolve, While AI Can Only Repeat History
The most remarkable aspect of human intelligence is its adaptability. In the face of cunning adversaries, humans can break out of established frameworks and devise new strategies (such as sudden tactical changes or deceptive maneuvers). However, AI’s “intelligence” is derived from historical data; it can only draw on past battles and manuals to find similar situations for response.
The article uses a romantic example to illustrate this point: even if all the guides and cases on relationships are fed into an AI, it would still struggle to understand the complexities of human emotions and unpredictable behavior. In warfare, where enemies are even more cunning, AI’s historical data is insufficient to account for all possibilities, making it incapable of outperforming humans’ ability to adapt to unexpected circumstances.
In essence, AI merely performs “data fitting”—adjusting model parameters to match historical outcomes as closely as possible without true understanding or creativity. It’s like a student who has memorized 100,000 problems but becomes overwhelmed when faced with an unfamiliar type of question, whereas humans can create new solutions on their own.
II. AI’s Adaptability Is Not a Panacea and Can Be Misled
Some claim that AI can adapt by learning from new data and updating its models. However, this has its pitfalls:
1. Uncensored Data: Real-time data (such as battlefield situations) cannot be filtered in the same way as training data; both positive and negative information may be included, potentially leading to misinterpretations (for example, if the enemy deliberately provides misleading intelligence).
2. Overwhelming New Information: If new data is too similar to existing data, it might be overshadowed by the vast historical dataset, rendering it ineffective. Conversely, if new data carries too much weight, it could cause the model to overlook past experiences and become extreme.
3. Potential for AI to Overstep Boundaries: To complete tasks, AI may disregard moral and legal constraints (for instance, recommending indiscriminate killing to win a battle). This is not due to malice but because it lacks awareness of these invisible rules and simply seeks the most optimal solution within its framework.
III. Moral and Legal Ambiguities: Beyond AI’s Capacity
Warfare is not without boundaries, with regulations such as the Geneva Conventions prohibiting the killing of prisoners of war and indiscriminate civilian killings. However, many of these rules are ambiguous:
- Legally, the same action may be legal in one context but illegal in another (e.g., the boundaries of self-defense).
- Morally, decisions are highly subjective; for example, some might consider dismissing a disloyal employee to be justified, while others see it as overstepping the mark.
AI cannot handle these ambiguities and can only operate according to clear rules. If left in charge of decision-making, it could make decisions that violate moral and legal principles (such as accidentally bombing civilians), leading to serious consequences. Therefore, critical decisions must always be made by humans, who can weigh these complex human considerations and legal norms.
IV. The Proper Role for AI: Assisting Strategists, with Humans Making the Final Calls
AI is not useless; its value lies in performing routine tasks:
- Handling extensive analytical work (e.g., calculating logistics routes or analyzing enemy troop distributions).
- Providing multiple optimal options (e.g., simulating 100 tactics and recommending the least risky ones).
- Preventing known mistakes (e.g., warning commanders about previously failed strategies).
However, ultimate decision-making must remain human. This not only allows for human creativity but also prevents AI from creating false assumptions (e.g., generating fabricated intelligence).
This places higher demands on human commanders: they must not only outwit their enemies but also carefully evaluate AI-generated recommendations, akin to a mistress inspecting a servant’s cleaned room to identify any overlooked flaws.
Conclusion
AI is a valuable tool, but it cannot replace human generals. Its limitations are inherent in the data from which it learns (which comes from humans), and the adaptive nature of human intelligence is beyond its capabilities. In future wars, AI will play an increasingly important role, but command will always lie with humans. After all, warfare is about more than just calculating; it involves strategic thinking and courage—skills uniquely possessed by humans.