I spent 1986 to 2000 working in IT, doing everything from software development and network administration to QA software testing. While I taught myself my first programming languages in high school well before taking formal classes, I don't do much coding these days beyond the occasional HTML and CSS tweak on my blog. The snippet above was actually written by AI—something I asked it to build to streamline a repetitive Excel task.
Sometime in the early 1980s, I made a computer Bulletin Board System (BBS) written in interpreted BASIC with 6502 assembly language subroutines on a Commodore 64. Running on a single phone line with a 1200 baud modem, it offered forums, messages, limited email, and live chat with the SysOp. It didn't enjoy the longevity of other systems at the time, but it was a fun, highly educational programming endeavor.
Because the system ran my own fallible code, a user could easily hit a bug that dumped the software back to the BASIC prompt, leaving the system down until I manually restarted it. To fix this, I wrote an assembly routine that ran on a background interrupt, like a watchdog. It monitored the BBS, automatically rebooting it if it crashed, and also logged the offending code so I could patch it later. In a way, I had created a self-healing loop that kept the system alive despite human error. However, it still required me to address the glitch.
Now imagine Recursive Self-Improvement (RSI) in AI. At machine speed, when an AI routine encounters a barrier in a programmed goal, it won't crash or halt—it will treat the obstacle like a chess board state and test all potential moves in microseconds.
Leveraging combinatorial code generation, the AI instantly spawns and tests hundreds of candidate subroutines in microseconds—it discards failing branches and deploys a workaround on the fly. This achieves an autonomous error recovery far beyond my ancient 6502 routine: instead of merely restarting the program after a failure, an RSI system dynamically rewrites its own underlying architecture so the barrier is bypassed before the system even registers a slowdown.
Why do we keep sprinting toward this ominous AI finish line when even the engineers building it are warning us about the edge of the cliff? It boils down to a classic, high-stakes game theory nightmare known as the Multipolar Trap. No AI lab, tech giant, or nation operates in a vacuum. The moment one player hits pause to play it safe, rival powers simply hit the accelerator. In the brutal arena of global tech dominance, slowing down or stopping isn't viewed as responsible stewardship—it's seen as handing the keys to the future to competitors who might not care about guardrails.
In the immortal words of Socrates: "I drank what?"