As the U.S. military ramps up AI use, what role should the technology play in war?

MARK BRODIE: AI has been changing many areas of public life — including in the military.

The Washington Post reports the U.S. used the most advanced AI at the start of the war in Iran than it has ever used before.

It’s a subject my next guest has been thinking about, too. Chuck Blanchard is senior counsel at the law firm Arnold and Porter and is a former general counsel of both the Army and Air Force; he also has served in the Arizona Senate.

He wrote about this recently in a Substack titled “Ban Killer Robots?: AI and the Military.”

Blanchard joined The Show to talk more about this.

Full conversation

MARK BRODIE: Chuck, what got you thinking about this at this time?

CHUCK BLANCHARD: Well, I was heavily involved in the issue when I was general counsel of the Air Force because there was a lot of discussion at the time in 2013, 2014 on what the department should be doing about autonomous weapon systems.

But at the time, I think the feeling was this was all theoretical because we just didn’t have the technological capability to make AI really be something we would be comfortable with in the battlefield. And then I watched AI just explode in the last few years.

I’ve seen how my law firm and other law firms are using it. I know clients are using it effectively. I’ve seen the capabilities. And so it made me think we need to go back to rethink exactly how should we approach autonomous weapon systems.

MARK BRODIE: Could you, back in 2013, let’s say, have envisioned what AI has become and the way that it could be used in weapons?

CHUCK BLANCHARD: AI was already beginning to be used by DOD. Certainly it was used in intelligence analysis. So the issue was, is the AI so capable that we would trust an autonomous system to make attack decisions without a human in the loop? And at the time, the answer was no, that we don’t trust the technology. We don’t think it could comply with international law.

MARK BRODIE: Do you draw a distinction between using AI, for example, to make decisions about who or what to attack and using AI to actually do the attacking?

CHUCK BLANCHARD: Yes, and I think most of the community on both sides believe that there is an important distinction. We already depend heavily on machines to help identify targets, to say, “This is the enemy. This is not the enemy.” And the best example is a pilot in an F-35 never sees an enemy aircraft; they are too far away. What they do see are radar signals that are interpreted by a computer based on a library of signatures to identify that’s the enemy, that’s not the enemy, that’s a commercial aircraft.

And so we already depend on that kind of issue. But we also depend on a pilot making a decision if it doesn’t look right, if what the computer is telling them doesn’t look right, the human being can say, “No, I’m not going to fire.”

And it was interesting at a conference I was at in England in 2014 that everyone seemed to agree that the critical issue was whether or not there was a human ultimately making the targeting decision.

The use of computers or technology or AI to help assist the human being is not really the moral or legal issue. It’s whether or not we completely eliminate the role of a human being in the process.

MARK BRODIE: Well, but as you write, it seems like humans could at least in theory eliminate themselves, right? Like they could just trust the AI over their own judgment, and so even though there is a human in the loop, their role is limited to nonexistent.

CHUCK BLANCHARD: Exactly. And so in response to that, you have to sort of train your pilots or your warfighters that they got to be careful about over-relying on the technology. That’s a reality.

And probably for a Patriot missile system that’s on automatic mode, it’s very likely that human beings are going to defer to the machine because the times are just so short for a missile coming in that they really can’t be an effective human interaction. But at least in that case, what you’re firing on is a piece of metal, a missile; you’re not firing on another human being.

But I think it does become more problematic if you’re talking about air defense systems that attack aircraft where there is a human being there, and then this becomes a concern that you train your folks to show to use their judgment and not just check their judgment at the door because a machine’s telling them what ought to be done.

MARK BRODIE: Do these questions become different in some way when you’re talking about an air war or a sea war as opposed to actual human soldiers on a battlefield?

CHUCK BLANCHARD: Yeah, I think actually the land warfare where you have a lot of civilians mixed in and you have cities raises significant compliance issues with international law. The law of war requires lots of really tough judgments that become very difficult in a land war with lots of civilians.

So I think we may be getting close to where we can comply with international law in the air domain and naval domain against air targets and against naval targets. I think we’re still much further away when we’re talking about ground war — which of course still raises the more philosophical issue: Should we still do it even if we’re capable of complying with international laws? Is there something still morally wrong with allowing a machine to make life-and-death decisions?

MARK BRODIE: I’m curious where you come down on that question.

CHUCK BLANCHARD: I think there should be a high threshold for the use of autonomous weapons. And the threshold’s not only we have to be doubly sure we can comply, but we also have to make doubly sure that there’s a reason to use AI. It’s not just a lazy reliance on machines, but there’s a real, compelling military reason to do so.

And if those both are met, however, I personally think that as long as human beings have some decision-making process, such as, “We’re going to attack in this geographical area, legitimate military targets,” I think this is not that far away from where we already are from a legal point of view.

MARK BRODIE: Yeah, I mean, it seems like there’s a difference — and we’ve talked about this and you alluded to it — there’s a difference between, for example, an autonomous drone being sent in to Ukraine or Iran or wherever, and whether, as you sometimes see in sci-fi movies, like a walking humanoid robot with a machine gun that can just go and shoot anything in its sight.

CHUCK BLANCHARD: Right. Although ironically, if you look at the most common violations of the law of war, it tends to be by human beings in land war because emotions take over, fear takes over, rage takes over. And so you could imagine a future AI machine that is not going to get angry, is not going to be fearful, is going to be more willing to take risk to itself in order to avoid harm to civilians.

You can imagine a sophisticated AI fighting machine — I don’t think we have them, we’re not there yet, but you can imagine that would occur. And then the difficult decision is, if we know that these machines are going to be more protective of civilian lives than a human being, should we use them for the very reason of protecting civilian lives, or is the notion of a scary-looking robot on the battlefield making life-and-death decisions, is that really going too far?

MARK BRODIE: All right, that is Chuck Blanchard, senior counsel at Arnold & Porter. Chuck, great to talk to you. Thank you.

CHUCK BLANCHARD: Thank you.