Replies to This Discussion

Permalink Reply by Smitty on August 20, 2008 at 1:34pm

Gee - this sounds like a question out of the Joint Warfare School....

Permalink Reply by Jon Compton on August 21, 2008 at 11:13am

Hi Eric,

I sat down and read the thesis last night in its entirety. Let me start by saying that the paper is very thorough in the handling of the experiment, and the way it is set up is quite intriguing. I'd like to congratulate your former minion on some fine work.

The following thoughts and critiques are not necessarily a commentary on the paper itself, but on certain methodologies in general.

Hard to pick a place to start, but I'll start with the statistics. I didn't see in the paper what statistical method was used to generate the coefficients and significance values, and I have some concern there. Because of the deterministic results we get from expected value systems, the outputs from each model run should be forming fairly distinct bell curves. Furthermore, although results should be non-linear over various control variables, they should be highly trended in the dependent variable nevertheless. As a result, it comes as no surprise that the statistical results are completely significant and explain almost all the variation. How could they not? The system is closed and has fairly little stochastic process going on. To tease out any real variation, a fair amount of data manipulation would need to occur, and I didn't see any application of that in the paper. That said, I think doing those manipulations would likely strengthen the result owing to the nature of the data, which in turn, still wouldn't change the fact that these are stats being performed on data generated by a closed deterministic system.

The paper goes on to point out that the stochastic processes that were introduced, namely time delays and information accuracy, seemed to have little effect on the outcomes. I think this is due largely to just how closed the system really is. Given that there is only three attack paths, what I lack in information, I can make up for by spreading my forces to minimize Red gain. Because of this, the variation of making bad decisions based upon poor information is quite restrained. Suppose the blue player could only really defend two of the three points of approach (owing to doctrine, landscape, or any other justification you care to use), then you might see a bigger variation in successes and failures, which might add more meaning to that particular bit, and thus have greater impact on outcomes.

Another part I found a bit obvious was the finding that sensor effectiveness topped out at 5 (of the nine possible approaches). Knowing where Red isn't is just as valuable as knowing where he is. If I can cover 4.5 of those 9 points, then, baring the stochastic processes, this will give me maximum effectiveness. I didn't really need a complex model to tell me that, but at least it puts some rigor to it. Nevertheless, the point can be proven easily enough with some basic mathematics. So that particular conclusion sort of falls into the "no kidding" category. Any variation caused by limited info accuracy and time delays is largely washed away by force size. There's no real way to make catastrophic mistakes in this system.

One thing that I've noticed in my various travels and discussions among military academia is that game theory is very popular in analyzing these sorts of questions. I think this is overall a very serious limitation in the way these problems are approached. Game theory is a great way to model decision processes, but it's not well suited to model dynamic environments. I would contend that applying game theoretical methods to something as dynamic as information gathering on the battlefield can only provide very limited insight. It is the structure of the game that determines the outcomes. As a result, game theory can pretty much tell you anything you want it to. As a result, it's not often telling you the truth. It can vet out a likely decision process given a specific set of circumstances, but it can't help you vet out whether the assumptions about the circumstances that were made are of any real use. Thus, the conceptual model of the game in the paper is very useful in the sense that it allows us to see the effectiveness of intelligences vs. force size given a specific environment structure, but it tells us very little about what it might look like in other structures, especially ones more dynamic, and where Red is actually trying to manipulate information space.

All that said, the paper is very correct to say that the problem is very difficult to address. It shows us that some of our intuition about information vs size may be correct, but it certainly doesn't prove it.

If I were going to suggest an avenue of approach, I'd say that the game theoretical work in this paper could profitably be coupled with agent based modeling, where individual agents might use some game theoretical structure to manage their own decisions, but that these decisions would then be tempered by environmental factors. The environment could then be set up to be dynamic, reacting to forces overall, by utilizing a dynamic system of equations. Blue player (or Red) could then interact with this environment at both the agent level and at the environment level. All that, however, is a tall order.

Anyway, hope this is of some use. Please pass on my regards.