Can AI Recreate the ‘Soul’ of Games? The Cutting Edge of AI Software Archaeology on a Classic from 1986

What Happened? News Overview

• Developers tasked Claude with recreating the browser version of the classic 1986 BBC Micro game ‘Thrust’. However, the initial output was far from the mark, resembling a pile of ‘slop’ with physics and control feeling completely off.
• Switching gears from using AI for ‘code generation’ to using it as an ‘analysis tool’, the project shifted to a form of ‘software archaeology’, allowing the AI to decode the original 6502 assembly language, leading to dramatic results.
• The AI successfully extracted hidden features from the old source code, including the mechanics of Q7.8 fixed-point arithmetic and turret shutdowns due to generator destruction.

Why Does This Matter? Key Takeaways

• The Challenge of ‘Soul’ Recreation: Even when gravity constants and thrust values matched the original code, the AI struggled to replicate the play experience (weightiness and inertia) due to timing differences inherent to the hardware of that era.
• Assembly Analysis by AI: This project demonstrated that modern AI (Claude) can comprehend the low-level code of the 1980s, which is often challenging for humans, and can automatically generate detailed specifications for subsystems.
• AI as Archaeology: It showcases how AI can act as a ‘bridge of knowledge’ when reviving past legacies (legacy code) with modern technology.

🦈 Shark’s Eye (Curator’s Perspective)

Simply tossing a “make this” command at an AI will only yield superficial imitations—this fact fires up the competitive spirit of engineers! Notably, the AI functioned as an ‘archaeologist’ interpreting 6502 assembly. The fact that it accurately explained intricate techniques like Q7.8 fixed-point arithmetic—a craftsmanship born from the resource-constrained 8-bit era—and distilled it into modern specifications is nothing short of miraculous! Ultimately, the ‘feel’ of a game isn’t just about constants; it lies in the ‘dense design’ that includes the execution timing of the hardware. Interrogating AI to extract this information will become an unbeatable weapon in future legacy system upgrades!

What’s Next?

The practice of ‘software archaeology’ through AI is set to become mainstream, accelerating the analysis of lost classic game source codes and black-boxed core systems. However, the final touch of ‘sensibility’ and ‘fine-tuning’ will continue to rely on the intuition of human players and interaction with AI.

A Shark’s Take

Feeding old code to an AI is like jumping into a time machine! The thrill of uncovering the craftsmanship hidden within is absolutely exhilarating! 🦈🔥

Glossary

• 6502 Assembly: A low-level language for the CPU “MOS 6502”, widely used in PCs and game consoles during the ’70s and ’80s. It is close to machine code and requires advanced knowledge to read.

• Q7.8 Fixed-Point Arithmetic: A computational method for handling decimals in older PCs without floating-point units, cleverly allocating 7 bits for the integer part and 8 bits for the fractional part.

• Software Archaeology: A method of analyzing undocumented or old source code to uncover its mechanics and design philosophy.

• Source: AI Can’t Recreate the Thrust Game (But It Can Help You Understand It)