Death in Groups

But I mean, isn't it obvious that damage to the truck alone as a result of the attack would imply quite a higher cost than whatever the shotgun was worth? (And yes, I think this is clearly the case even when you consider that the probability of being attacked is quite a bit less than 100%.) I don't think this shows lives being insufficiently valued in the military; I think it just shows the sort of pervasive dysfunction we would expect in any large-scale organization lacking internal mechanisms to ensure accountability and proper response to incentives.

The warrior ethos is the result of an iterated hawk-dove game.

In the hawk-dove game, playing 'hawk' only works if the ratio at which you value winning over surviving is higher than your opponent's.

Gaining the 'warrior ethos' requires driving the amount that you value surviving down to epsilon, so that no one can successfully back you down with a threat to your life.

This is true. I would rate the odds of actually dying 'very low', which my gut wants to ballpark in the <5% region. However, I do have some information that would let us make this a little more precise, so here goes:

I was in-country for a little under 90 days. The first 30 of them were acclimatization and specialized training, which didn't involve being exposed to any real risk of combat. According to the paperwork associated with the incident, we were involved in 19 ambushes while I was there, which includes things like IEDs, but also just people spraying the vehicles with their guns to show us they are not afraid (this is almost zero risk to us). This leaves 19 ambushes over 60 days of operations, going out on average once per day.

So my prior is ~1/3 chance of someone trying to kill us each trip. Now there are two additional factors to consider: one, trips along the same route are not independent; two, the likelihood of dying in an attack is low. As it happens, 'never take the same way twice' is the literal first rule they taught us about movement when I went through training, perhaps because all of the cadre had been deployed previously. The more you drive a route, the more likely the enemy is to be prepared for you to drive the route - so each time you go out, the risk of attack is higher than it was before. When I was trained, they told us it was a given you would be hit the second time around - this is definitely an exaggeration, but I'll say it is half again as likely as the previous trip at a guess, for a multiplier of 3/2. The likelihood of dying in an attack is low because everyone has first-responder training and medical kits, and we have medical evacuation helicopters, so dying usually needs one of three things: dying instantly; being so badly injured you die before medical evacuation; medical error (overlooked internal bleeding, or the like). Further, relatively few attacks are bad enough that someone is injured. I don't actually know how to model these, though I know the Army has statistics for all of them.

Naively I can take a stab and say something like 1/20 attacks where someone is injured lead to one of the death conditions, with the proviso that it is only a guess. I can make a better guess for the probability of injury given an attack - of the 19 ambushes, I think there were 3 events where someone was injured (each corresponding to an IED blowing up a truck, incidentally). So this is what we have currently:

Prior for being attacked on a given trip: 1/3

Additional trips on the same route multiplier: 3/2

Likelihood of someone being injured during an attack: 3/19

Estimated likelihood of death given injury: 1/20

So that looks like:

(1/3)(3/19)(1/20) + (1/3)(3/2)(3/19)(1/20) + (1/3)(3/2)(3/2)(3/19)(1/20) + (1/3)(3/2)(3/2)(3/2)(3/19)(1/20)

This comes out to a little over 0.02 probability of someone dying, for four trips over the same route. Such is the power of history's mightiest civilization.

Of course there's plenty of stuff we left out: dying isn't the only risk, being crippled is much more common; risk of injury is not equal throughout the group, and injuries tend to be clumpy because most of them come from trucks of people getting blown up; I knew none of these calculations at the time.

This doesn't change the underlying intuition. In example, in the rational guesser game it does not matter what fraction of the average-of-guesses is given; the difference between 2/3 and 99/100 is only the number of iterations to 0.

The United States Army is probably the foremost institution of its kind in history. It has high prestige, enormous resources, and a great deal of experience. The feedback loops of violence are tight; the risks to personnel are immediate; the decisions it makes are as close as humanity gets to a stock market for death. And yet, at the point of decision, we will risk our lives for zero gain.

This causes me to view the approach for valuation of lives in X-risk uneasily, and from the opposite direction of sacredness.

The blast rotated the truck 180 degrees on both the axes of the plane, so the crew compartment was facing back the way we came. The rear gunner position was to stand up out of a rear emergency hatch, on one of the seats. We just had a jerry-rigged mechanism to secure them to the truck. They would have no shock distribution from the seats or protection from the seatbelts (which are like the ones they use in Nascar).

As a consequence, anyone in that position might have been torn in half, or splattered against the roof of the truck as it came up to meet them, or if the make-shift harness failed been flung hundreds of meters away. We stopped after units started losing guys when trucks would roll over.

As it happens, that was my job until we changed the procedure.

a) Confidence in the ability and good faith of the small group of people in a position to issue the order went down. In particular, my estimation of their ability to understand risks and trade-offs.

b) No change. This was expected.

c) No change. This was expected. I would also like to flag this as the point that appears to be the largest inferential gulf between most of the audience and I.

d) Confidence in the hardware went up, confidence that my buddies would live through future engagements went up as a result. I've ridden in two prior generations of vehicle, and the odds of there being no fatalities in either approaches zero in my estimation. The progress of the MRAP series of vehicles is on the short list of success stories from those wars. Also, I updated in favor of seatbelts.

It is worth mentioning I had no notion of Bayes at the time of this event.