Artificial Intelligence and War Crimes

<Largo>

“Hm… Why does this AI behave like this?”

In my room at the girls’ dorm, I was groaning in front of my computer shogi software.

“Shogi is that Japanese game with rules kind of like chess, right?”

I thought I was just talking to myself, but Lala cut in.

“The basic rules are mostly the same. The game ends when the king is taken, just like in chess. The biggest difference is probably that in shogi, you can use the pieces you capture as your own. There’s even a rumor that, when Japan lost the war and was occupied by America, this rule almost got shogi banned because it was seen as abuse of prisoners.”

“Huh. So, what’s up with the shogi AI?”

“Once checkmate—or ‘tsumi’ in shogi—becomes inevitable, the moment the king is cornered, the AI starts using every piece it has, including its ‘prisoners’, to launch pointless suicide attacks against my king. Even though the AI’s loss is already certain and all it’s doing is stalling, it keeps going until its own territory is reduced to scorched earth.”

I meant it as a complaint, but Lala’s eyes sparkled.

“That’s probably because the AI is built using an objective function.”

“Objective function?”

Lala used a term I hadn’t heard before.

“The computer calculates all the possible next moves it can take, and assigns each one a score. The AI chooses the move that earns the highest score.”

“It’s just like karma points.”

This suddenly connected with the numerical metrics I’d been grappling with recently. Lala kept explaining.

“There are basically two types of actions that increase the score: moves that corner the enemy’s king, and moves that buy time to prevent your own king from being cornered.”

With that explanation, the mystery of the AI was solved.

“I get it! Once your own king is doomed, you can no longer score points by winning. The only way to keep earning points is to stall for time until defeat. And according to the rules, the only way left to stall after checkmate is to launch suicide attacks at the enemy king.”

“Exactly.”

Lala really is good at explaining things.

“But, you know, it’s okay because this is just a game. If it were a real war, it’d be terrifying.”

“There you go again, linking everything to karma points.”

Lala looked exasperated, as if to say, “This again?” But I didn’t stop.

“If, after defeat, it’s not even about the survival of your nation or people, but just about preserving a dictator’s pride, and even when defeat is certain, soldiers’ lives are sacrificed one after another just to extend the king’s life…”

I have actually heard news stories about wars like that. I bet, when human societies make bizarre decisions like a shogi AI, even if things aren’t that extremely quantified, the rules that govern organizations get simplified like a game and decisions end up being based on just a few factors. Like when maintaining a corrupt power structure becomes the nation’s top priority, to the point where a revolt would almost be justified.

“Right. Operations on a national scale, like wars, are driven by countless numbers. Without numbers, you couldn’t make plans at all. But in situations like that, numbers easily swap out the ends for the means.”

“But numbers aren’t the goal—they’re just a tool.”

“A tool, huh…”

All kinds of numbers I’d encountered in my life floated through my mind, only to fade away again.

“Ah!”

“What’s wrong?”

“I forgot my hair salon appointment. I can still make it if I hurry!”

“Time is a number that builds trust, you know!”

“I know!”

I dashed out of the room in a hurry.