AI for Military Support

Interesting empirical research: “Black Box Warfare: Human Judgment and Military Decision-Making in the Age of AI.”

Abstract: How is AI transforming decision-making in modern conflict? This study provides a unique empirical window into that question by deploying a high-fidelity replica of an AI decision-support system (DSS) used in military targeting. After reconstructing the interface and functionality of the real-world system, we tested its impact on combat decisions in two experiments involving 2,015 Israeli military personnel. Contrary to widespread fears of automation bias, we find strong evidence of algorithmic aversion, especially in scenarios involving high collateral damage. Yet we also show that integrating “explainable AI” features reduces algorithmic aversion and promotes more thoughtful evaluations of algorithmic recommendations. These findings challenge prevailing assumptions, revealing that trust in military AI is dynamic, varying with individual predispositions, perceived operational stakes, and the informational features of the interface. By grounding normative concerns in empirical evidence, our study offers critical insight into the integration of AI in warfare and underscores the enduring importance of human agency in high-stakes military decision-making.

Tags: academic papers, AI, military, war

Posted on August 11, 2026 at 7:18 AM • 4 Comments

Comments

KC • August 11, 2026 11:07 AM

re: Explainable AI

According to this study, humans generally trusted AI more in targeting decisions when they had greater transparency about inputs.

Studies 1 and 2 compare AI v. human analysts, where study 2 has explanatory data.

See:

Figure 1: Strike scenarios (p 9)
Figure 6: Strike scenarios with three points of source data (p 17)

The results indicate that the preference for human analysts decreased with more AI transparency.

Interesting, to note also, were the human factors that increased strike approval rates (Table 1, 2): vengefulness towards adversary, gender differences (male higher approval), conservative political orientation. In regards to decision certainty: technological literacy

Clive Robinson • August 11, 2026 11:52 AM

@ KC,

You note from your reading,

“trusted AI more in targeting decisions when they had greater transparency about inputs”

And,

“In regards to decision certainty: technological literacy”

This would be what I expect with the analysts I’ve worked with in the past. It is in effect a “professional approach” to what is a difficult and stressful activity.

However there is now an issue that I think people should actually take into account.

Back when I used to be involved with such activities the entire information chain was “human” from the boots on the ground all the way to the keys of the mechanical typewriter (yup they were still in use back then).

There were ways information could be double or triple checked and given a reliable rating for accuracy.

These days AI is happily polluting all along the information supply chain, and you can nolonger tell the quality of “input data” any longer because you can not “look back” along the chain and tell if Intel is AI or just other electronic gathering.

Thus we have entered,

“The lies of ommission phase”

Yes you might be able to see “input” but how do you judge if you are “seeing it all” or if “some is being withheld”.

AI can arbitrarily withhold information from an analyst and the analyst would not be able to tell.

In the past with humans all along the chain sufficient would leak through for an analyst to “pick up on” and thus be more cautious.

AI can easily “False Flag” it’s own sides analysts based on what some other entity had prioritized…

Testing for and preventative training to reduce/avoid this issue is something that needs to be done, but in all honesty I don’t think it will be “under current management”.

Rontea • August 11, 2026 12:23 PM

Fascinating study. I am happy to see that we are not automating tragedy. This is the kind of work that cuts through assumptions and forces us to look at how people actually interact with AI on the battlefield. The results highlight that soldiers aren’t simply rubber-stamping algorithmic outputs – if anything, high-stakes scenarios amplify their skepticism. That aversion shifting once explainability is introduced is a critical takeaway. It’s a reminder that human judgment isn’t going away; it’s being reshaped by interfaces, trust calibration, and the context of decision-making. In any operational environment, transparency features that can bolster confidence without over-simplifying the underlying complexity are going to be essential for responsible deployment.

Untitled • August 11, 2026 2:47 PM

I wonder whether the personnel being studied reacted to situations differently than they might have done if they weren’t under scrutiny. I note this statement in the paper:

Lavender alone identified more than 37,000 targets in the early weeks of the [Gaza] campaign. While human operators formally remained “in the loop,” approving or rejecting each target before a strike was ordered, reports suggested that trust in the system grew so strong that many operators approved its suggestions in as little as 20 seconds. [my emphasis]

Yet the study finds “strong evidence of algorithmic aversion”.

AI for Military Support

Comments

Leave a comment Cancel reply