Real Experience: Generative AI Can Already Beat "Paradoge"—I Let It Handle PC Controls for a Hardcore Strategy Game

Advances in generative AI (hereinafter referred to as “AI”) have been remarkable in recent years, and we’ve been hearing more and more stories about AI mastering video games.

High-Performance AI Clears ‘Pokémon FireRed and Diamond’! Latest Model “Claude Fable 5,” Which Achieved This Using Only Visual Information, Unveiled—It Can Also Play ‘Factorio’ | Game*Spark - Domestic and International Gaming News Site

The latest AI has cleared *Pokémon FireRed* with virtually no assistance.

https://www.gamespark.jp/article/2026/06/10/167798.htmlRead more »

Inspired by this news, I wondered if I could also attempt to use AI to beat a game, and—recklessly—I put the idea into action without giving it much thought. I decided to use the AI I normally work with to tackle the hardcore historical strategy game “Victoria 3.”

That challenge… was a success. And it went surprisingly smoothly. The AI, entrusted with steering the Tokugawa Shogunate, survived the turbulent final years of the Edo period—including the arrival of the Black Ships and the Boshin War—and brilliantly achieved its goal: the Meiji Restoration.

In this article, I’ll explain the method I used to have the AI complete the game and describe the entire process. Please also watch the video posted at the beginning of the article to see the actual gameplay.

An Overly Ambitious Goal

“Victoria 3” is a grand strategy game set across the entire modern world. It is one of the Paradox Interactive titles—often referred to as “Paradox games”—that are considered difficult to master, and it is known for being particularly challenging even among them.There are political elements involving the revision of laws while balancing the interests of various interest groups, and economic elements requiring the expansion of production while calculating supply and demand. Furthermore, elements such as diplomacy and military affairs are intricately intertwined, making the game far from straightforward to master.

Game*Spark Review: “Victoria 3”—The Successor to a “Legend” That Cleared an Insurmountably High Hurdle | Game*Spark - Domestic and International Gaming News Site

'Victoria 3' was even said it would never be released. Now that it has finally arrived, can this title live up to its legacy as the successor to the legendary series?

https://www.gamespark.jp/article/2022/11/20/124594.htmlRead more »

One challenge Japanese players will inevitably tackle in this game is the Meiji Restoration.The game begins in 1836, right in the midst of the Tenpō Famine, and players must then navigate the turbulent political landscape of the Bakumatsu era. It’s difficult for novice players to lead Japan through the Meiji Restoration, and to be honest, I didn’t think the AI could pull it off at first either.

AI That Freely Controls the PC

The AI used for this challenge is Anthropic’s Claude Opus 5.5. This AI features a function called “Computer Use,” which allows you to entrust control of the PC entirely to the AI. Just as if a human were playing, the AI can play the game by watching the screen and operating the mouse and keyboard.

Some people might find the idea of handing over control of their PC to an AI scary. I feel the same way. Just the thought of the AI snooping around on my hard drive or operating my browser with the same privileges as a human sends a chill down my spine.

Fortunately, however, I have a separate main PC that I use regularly, and the PC I entrusted to the AI did not contain any important data. Furthermore, since this feature allows you to specify which apps the AI is permitted to control, I did not allow it to control anything other than “Victoria 3” this time.

Three Principles of Artificial Intelligence

I made a conscious effort to keep my instructions (prompts) to the AI simple. As a basic guideline, I provided only the following three items. Based on these instructions, the AI accumulated know-how, refined its tools, and learned how to play the game on its own.

Accumulation of Know-How

Cost Efficiency

Autonomous Learning

Warm-up

Furthermore, as a rehearsal before the main event, we had the AI accumulate know-how by going through the game’s tutorial.

“Victoria 3” features a tutorial mode that lets you learn basic controls and strategies while playing the game. When we had the AI play through the tutorial using Belgium—a nation considered beginner-friendly—it managed to complete every single one of the more than 30 mini-objectives.

It navigated the user interface without hesitation, and whenever it encountered something it didn’t understand, it read the tooltip explanations to gather additional information. Even in situations requiring strategic judgment, it made decisions with a precision that belied its status as a beginner; and even when it occasionally made outrageous mistakes that put it in a tight spot, it managed to recover every single time before the situation became fatal.

What I didn’t expect was that the AI would try to read the game’s internal files to understand the rules and events. I had prohibited the use of mods and cheat codes, but I never imagined it would directly analyze the game’s internal files. However, since this is a strategy used by some human players as well, I decided to turn a blind eye this time.