I came across this article that said AI was almost if not as good as humans in team player games like DOTA. Now DOTA has landscapes, multiple players, different abilities etc. So that immediately set off a question in my head:

Why don't we take all the detailed satellite images we have of earth, and set off AI's to simulate war scenarios and learn best strategies from them? Naval Warfare in the English Channel for example. Let A be UK and B be an assortment of Germany, France, Bulgaria etc. The Himalayas for another example, Nepal v. China or India v. China or whatever.

Surely we can model the necessary complexity of artillery, air-force, navies etc. into our models?

So why isn't this being done? Is it not feasible? Or is it already being done by under super tight wraps?

PS: I do not know what the right community is for this kind of question, world building, machine learning, security, skeptics, etc. Feel free to move it to the right place but I suspect this is already being (secretly) done, hence I chose skeptics as the place to ask.

  • 3
    Questions here require a notable claim. I did do a search on this and I found Project Maven, but I don't think there's much room to doubt its existence. – Laurel Aug 3 '18 at 1:38
  • Okay. Thanks! I don't think the idea is old enough to be notable as of yet, and if it were under the wraps, then the media would conceivably have push back against reporting on it as well... but to know it is indeed being put into practice makes one feel a little less crazy. – ITA Aug 3 '18 at 1:43
  • Short version: because, from an AI perspective, an entire DOTA map is roughly as complicated as your front yard. – Mark Aug 3 '18 at 2:14
  • 2
    @Mark much simpler than a front yard, actually - front yard isn't a digital model with tons of developer-set markers specially for the AI analyzing it to make correct decisions. – Danila Smirnov Aug 3 '18 at 3:23
  • 3
    @ITA If you reformulate your question a bit (for example, "Do military forces use AIs in planning, and if not, why?") you could ask it on Artificial Intelligence SE. – Danila Smirnov Aug 3 '18 at 3:26

We are

The military do use AI and various other similar tech. The example that comes to my mind is this episode of Top Gear in which they race a self driving military truck. I think what you're asking is about a more total warfare

Surely we can model the necessary complexity of artillery, air-force, navies etc. into our models?

No we can't model systems that complex, there are simply too many variables. WAY too many varibles. Take the example of Goggle AlphaZero chess engine, one of the strongest examples of AI at the moment. It works (Simplification Obviously) by taking the rules of chess and then playing against itself thousands of times to find out what's best

Now what would the 'Rules' of war be? With Chess you have a closed system, thats not the same in real life. Think about the mathematics of working out where a bullet will land, you need to think about wind, the mass of the bullet, cross section and muzzle velocity of the bullet, local gravity (its not the same everywhere), temperature, barometric pressure, curvature of the Earth, Coriolis just to name a few. That might give you an estimation of a single bullet from a single gun. Obviously once you've fired one bullet that all gets thrown out the window from recoil and heating of the barrel etc. and that's before you even thinking of bringing in the human element !

The concept of modelling combat with mathematics is not new its been around since WW2, look up the lanchester's Laws if you're interested. It's still worked on today Here's a paper on it. I think the introduction to the paper sums up the situation quite well, that maths(and therefore AI) can give insights but it cannot be seen as an exact science.

Despite all the research and despite all the resources spent, combat modelling can hardly be considered a fully scientific undertaking. The reason for this statement is that combat models can hardly be exposed to empirical testing. Certainly it is not physicaly impossible to carry out adequate empirical testing on combat models, but in practice it will usually be considered unacceptable both for economic and moral reasons. Some scattered pieces of empirical evidence may be obtained from studies of war history, but this will hardly be sufficient to establish combat modelling as scientific. Nevertheless combat modelling is generally accepted as a practical tool for military decision makers. Combat modelling may be useful for military decision makers to support summarizing, processing and reaching conclusions from relevant and available information

Not the answer you're looking for? Browse other questions tagged .