I was honored to speak at the Global Nobel Laureates Assembly at the Pope’s summer residence in Borga Laudato Si this past week. Over two and a half days, dozens of Nobel laureates were joined by international experts, diplomats, technologists, and religious leaders from the Vatican and beyond to discuss nuclear risks and their intersection with AI. The Pope’s encyclical Magnifica Humanitas served as the touchstone for the entire event.

I believe the best thing to come from the conference is less the Nobel Laureate declaration and more the discussion it prompts going forward—especially if that discussion extends beyond Rome, to the many communities worried about and affected by these technologies. 

I observed three themes from the conversations at the assembly: 

• the growing dangers of nuclear weapons and the moral imperative to either prevent their use or eliminate them entirely (two related camps of people who have substantial disagreements if similar desired outcomes)
• the threats of AI to humanity in general and war more specifically
• the potential for AI to improve or otherwise support human decisions and in turn help prevent nuclear war

A few attendees have already published their prepared remarks, and I sincerely hope more will do so. Since we were limited to eight minutes, they are succinct distillations of our highest priority issues for these sorts of discussion. My remarks reflect my continued interest in better connecting the policy, technology, and military communities. They also reflect my hope to bring to these conversations the experiences of servicemembers who are endeavoring to serve their country; protect themselves and their shipmates; and fulfill the lawful orders of political leaders with honor, under extraordinary circumstances.

Remarks for the Panel “The Moral Challenges of AI and War: Killer Robots, Autonomous Decision and Escalation, and Machine-Speed Conflict”

Thank you. It is a pleasure to be here, and I approach this panel with humility and admiration—for the laureates and religious and political leaders in this room, and for my co-panelists.

My background is as a former military servicemember. Before the advent of AI and drones, I was trained on and operated what were then among the U.S. military’s most autonomous weapons: the Tomahawk missile and the AEGIS missile-defense system. I was 26. 

Years later, I find myself once again studying the U.S. military’s most widely fielded AI decision-support tool, the Maven Smart System from Georgetown University’s CSET. I wrote about its evolution and adoption in 2024. I have since studied its expansion and visited with the U.S. servicemembers who are using it around the world.

My hope today is to use my experiences as a bridge between these discussions, engineering choices, and the daily actions of military servicemembers. In doing so, I hope to leave you with three points:

• that military processes are flawed and that we should push AI applications within military processes to address those flaws in a way that supports human intelligence 
• that supporting human intelligence also requires thorough preparation of humans—not just ever more capable technologies
• that “machine speed” can be a misnomer, but we must demand time for human judgment

First, let me start with ensuring human intelligence and wrestling with human limitations.

My starting point for this discussion is that I feel acutely the limits of our current military processes and technologies. They are flawed today. Moreover, military service members may be well intentioned, but they are, like all of us, imperfect—sometimes tired, scared, angry—but they nonetheless bear the moral responsibility for their lethal decisions. 

The Pope’s AI encyclical teaches that “humanity flourishes not despite limitations, but often through them,” and I agree. 

But knowing how limited humans are, especially in war, I cannot help but wonder how AI might support human intelligence when humans are at their moment of greatest strain. 

For example, the encyclical warns that “AI does not remove the intrinsic inhumanity of conflict; indeed it can only bring about conflict more quickly and render it more impersonal, …thus reducing victims to data.” I understand this concern. I believe it names a real risk. But I believe it is critical to note that this is a risk that has been realized in the past without the benefit of AI.

Let me propose a different way of understanding AI in military decisions. Militaries today use a highly structured process to weigh courses of action: feint left or right, strike the tanks or the logistics convoy. Despite its structure, the process is limited by time and people. Sometimes the thought process can be described by what American servicemembers wryly call a BOGSAT—which stands for a “Bunch Of Guys Sitting Around a Table”—making calls without much information or much time. Now imagine one AI joins that table. I don’t think it could be just one AI: a single data-fluent voice affecting a tone of objectivity could skew the conversation dangerously. Imagine instead ten or more AI models at the table, each purposefully generating plausible scenarios that challenge the humans to confront their own assumptions and blind spots and to wrestle with patterns in the AI-generated ideas, too. Same technology. Different employment. Different effects on human judgment and war. 

We must be clear-eyed about how AI can support human intelligence and judgement in contrast to the status quo in these moments of great need.

Second, while there are many technical design solutions to consider, none of them will achieve the goal of ensuring meaningful human control or the exercise of appropriate judgement if we do not also demand military readiness to work with AI—most especially human preparedness.

I say this in part because there is a view that a human in an AI decision system is automatically a failsafe. It is not. Putting a human in control, or human in the loop, may allow for accountability in the thin sense. We know who to blame, but we cannot congratulate ourselves for holding someone accountable without first demanding they were prepared. Achieving meaningful human control requires training; certifications; regular exercises; engineers embedded with military units; and technical, political, and military oversight matched to the gravity of the decision. Any state unwilling to pay that price has already delegated the decision to the machine—whatever its written policy says.

This brings me to the issue of machine-speed conflict.

“Machine speed versus human speed” is a convenient shorthand, but it fails to communicate well across communities. It prompts one to envision a process that is either entirely “machine speed” or entirely “human speed.” But military processes are made up of many parts. My recommendation is that we specify which parts of decision-making are specifically concerning.  

To illustrate, here is a hypothetical example: Imagine a soldier tasked to evaluate a target against the law of war: military necessity, proportionality, and distinction. You could conjure an AI that simply answers “yes” or “no” at machine speed to those three questions. I find that highly problematic. Now envision a different AI: one that speeds or even expands the soldier’s search for evidence in a target review, helping them to identify and retrieve relevant information. The evidence-gathering part of the process may accelerate but not necessarily the time required to consider that evidence. 

What militaries do with the time they recover from using AI in decision processes is a choice they will make in the coming years. This forum and others like it have the opportunity to shape that choice by advocating for the preservation of adequate time and people for proper deliberation within AI-enabled processes, understanding that while AI may speed up certain process steps, it may not speed up all steps or speed up all steps equally.  

In closing, I will reiterate my thanks for including me. I believe this interdisciplinary forum has the power to advance practical conversations that shape the nature and culture of AI adoption within militaries, when government to government conversations are challenging given a mutual lack of trust. Moreover, I believe the Vatican and other religious leaders have an especially unique power: When they speak, many people listen, especially military members. My ardent hope is that as we all continue to debate AI for military decision-making, we keep in mind the three themes I raised today:

• How might we use AI to support the human conscience at its moment of greatest strain?
• How might we ensure humans are prepared to operate ethically with these tools? 
• How do we encourage militaries to consider that while some parts of their operations may accelerate, that does not mean that all should accelerate or accelerate equally?

Thank you.