Company Of Heroes: Opposing Fronts- A Skirmish Match Between Human And Artificial Intelligence Reveals Weaknesses In AI’s Armour

There is an interesting and informative compilation on Current Methods to Create Human-Level Artificial Intelligence in Computer Simulations and Wargames (Google search). The opening page mentions that computers have difficulty thinking out of the box.

However, we know what we want the AI to do: confronted with the situation that faced Grant before Vicksburg in 1863 we want the AI to ignore conventional military wisdom and send its troops below the objective and conduct a lightening campaign that ends with the envelopment and surrender of the enemy forces. Likewise, in a simulation of the Korean peninsula in September, 1950, we want the AI to conceive of a daring amphibious invasion 180 miles behind enemy lines as MacArthur did.
These battles illustrate the human ingenuity, as a bold commander can take calculated risks and perform unsuspected and illogical manoeuvres to meet the objectives. According to the article, artificial intelligence (AI) is still lacking, and we are not only talking about gamers’ machines which have limited power and thus cheating AI is the order of day.

What about the military systems where genuine AI has to assist commanders in making consequential decisions? Of course hardware is not a constraint here and smart decisions by machines can be critical.

Company of Heroes: Opposing Forces is a real time strategy game set in World War II, and allows for realistic playing with tactics and strategy.

closeup, where all the action was happeningThe objective of the Victory Point match is to hold either two or all three of the victory points (two shown in this figure, the third was captured by enemy and I made no efforts for that) while a ticker counts off- more for the weaker force.

There are mobile headquarters (British) and I shift mine adjacent to a corner Victory Pont and establish a heavy defence including machine-gun nests, hidden mortars, anti-tank guns and slit trenches, plus a medic station covering all these men.

The south portion of my block consisted of soft targets like HQs, artillery etc. I then recapture the central Victory Point after a small fight, and establish a slit trench. By the way slit trenches are like God-mode. It’s very difficult to kill soldiers stationed inside them.

Now this is what the artificial intelligence throws at me: Waves and waves of soldiers and tanks along path 3. Some occasional personnel will try to capture middle Victory Point, but they are covered by a mortar and some rifle squads etc separate from the main offensive block. Thus I own two out of three and timer is lessening for my enemy and I only have to play the waiting game.

Also, during this time, AI sends a squad to capture sectors and resources and it captures a strategic point at location 6. Thus it comes to know of my soft targets (like headquarters, artillery etc).

Also, the resources are limited and I have many, many weapons laid bare on northern front. So? That means my south is open to attack with little defense.

A (human) commander would have utilised this scenario and given me a tough time by routing his armour and men (marked by ?) along the dotted path and first destroying my HQs, then artillery etc before swarming my other forces.

That would have forced me to take evasive actions and thus leave the Victory Points up for grabs. which is the objective of the match. But from AI, that attack never came.

The complete sketch of the scenario:
sketch of the situation of me vs computer, where little ingenuity was shown by the artificial intelligence
(larger image)

Originally published on my personal blog here