©WebNovelPlus
Video Game Tycoon in Tokyo-Chapter 923: The AI with Crazy Iterations
Chapter 923 - The AI with Crazy Iterations
"Huh... this is kind of interesting. Those little guys are actually trying to dodge now. Are we sure they're not being controlled by a human?"
"Since the president already said everything's controlled by AI, then it must be true. There's no way he'd lie about something like this."
Takayuki couldn't hear exactly what people were saying below, but he could guess.
He continued, "This is after one hour of AI training. The red and blue agents have started developing different strategies."
"The blue ones first tried hiding behind fixed barriers. But since the barriers had gaps, the red ones could slip in and catch them. After enough training, the blue ones learned to block those gaps using movable cubes, effectively stopping the red ones. At this point, the red team was at a loss."
As Takayuki explained, the audience became more focused.
It was fascinating.
All this behavior was generated through AI self-training.
So... does this mean AI has started to think?
Interesting. Very interesting.
Some people even forgot they were here to learn about how AI would be applied to game development.
"Let's fast forward another hour."
On the screen, Benedict fast-forwarded the footage again. He had spent an entire day running this training on the lab's best computers, recording the progress hourly for this demonstration.
Takayuki continued:"After another hour, the red team figured out a new strategy. They realized they could climb triangular structures and jump over the barriers to reach the blue team."
The screen showed two red agents dragging triangles to the barrier, using the slope to climb and jump in to catch the blues.
"Then, the red team had the upper hand. But in the next hour..."
The blue team adapted again. They dragged the triangles into the barrier before the game even began, sealing off access and leaving the reds helpless.
Without the climbing tools, the red agents couldn't reach them anymore.
Th𝗲 most uptodate novels are published on ƒгeewёbnovel.com.
But then... another hour later, the reds found yet another new trick.
And so it went, with each side constantly discovering strange and creative ways to win. The strategies kept evolving in ways that surprised everyone.
Eventually, the blue team even started trying wild ideas—like running far away at the start of the match. The farther they got, the harder it was to be caught.
The red team, meanwhile, began exploiting subtle bugs in the simulation to win.
Everyone watching was stunned.
"Wait... computers can do that? These strategies look like things a human would take forever to figure out."
"No need to guess. That's exactly what AI is capable of."
"I just had a thought... what if we gave Sekiro's bosses this kind of AI learning ability? Wouldn't that make them unbeatable?"
At that, everyone nearby turned to stare at the guy.
He shifted uncomfortably. "W-why are you all looking at me like that?"
"Are you even human? If we gave bosses that kind of AI, no one would ever beat the game. Would it be the player playing the game, or the game playing the player?"
Sekiro's difficulty felt intense, but it was carefully designed. Each enemy's challenge curve was handcrafted to keep the player feeling like victory was always just within reach—if they could avoid major mistakes.
The Tenth Development Department even had dedicated elite enemy and boss design teams for this.
If Sekiro's bosses had real AI? Almost no one would ever finish it.
Difficulty in games is meant to create satisfaction—not despair. If it's too high, it just drives players away.
Games like Demon's Souls were partly a product of their time, but also just a bit too hard in their early design. Later, FromSoftware learned how to better shape difficulty curves, leading to more balanced experiences like Dark Souls III and the legendary Elden Ring.
"I was just saying... I didn't mean it seriously," the guy muttered, shrinking back.
Then another person spoke up. "Actually, now that you mention it, Monster Hunter could probably benefit from AI like this. The monsters are already powerful—it would make sense to push them further, right?"
Now that guy got the same deadpan stares.
Yep, another madman. Everyone silently agreed that people like him should never be allowed near difficulty balancing.
The demo concluded, and everyone was amazed at how fast the AI had learned.
From 10 a.m. in the U.S. when the program was finalized, Benedict's lab ran nonstop training until around 7 p.m.—roughly 9 hours.
In that time, the four AIs had evolved through several iterations. It was stunning.
To Benedict, though, this was child's play.
His real goal was to train an invincible StarCraft-playing AI—one that no human could ever beat.
This little simulation didn't even fully tax their lab's computers.
But even so, he was blown away by Takayuki's creativity.