This post was rejected for the following reason(s):
Not obviously not Language Model. Sometimes we get posts or comments that where it's not clearly human generated.
LLM content is generally not good enough for LessWrong, and in particular we don't want it from new users who haven't demonstrated a more general track record of good content. See here for our current policy on LLM content.
If your post/comment was not generated by an LLM and you think the rejection was a mistake, message us on intercom to convince us you're a real person. We may or may not allow the particular content you were trying to post, depending on circumstances.
Fwiw, this still really sounds to me like it was written by an LLM. I think if you really want to contribute to LW, you should probably spend more time reading LW, particularly the sequences highlights.
TL;DR
Imagine the existence of a simple deterministic Chess variant game that was not just harder for AI to play but made the game fundamentally incompatible with the whole idea of long-term strategy. What if one simple rule change permitted both players to momentarily control the opponent's pieces?—that's it. With this single change, strategic reasoning becomes structurally impossible after just a few moves. I created such a monster. Now, my hopes for safer AI rests solely on brighter minds weighing in. Why not play a game with a friend today?
Introduction
You see, during COVID-19, I embarked on a very fun independent research project as a distraction. I wanted to invent a Chess variant that AI would struggle with. As I improved the game, AI struggled exponentially more. When one late night, the AI safety implications hit me hard, the fun was gone. My research spiralled into a new direction, one with URGENCY and of paramount importance. My efforts intensified—my quest became an obsession.
Controlled Chaos Chess (CCC) Rules
Think of it this way: After any capture, both players must immediately take turns forcibly relocating one of their opponent's pieces to their opponent's back ranks (and, of course, trying their best to punish their opponents). Each player briefly controls their opponent's piece, but any "thrill of victory" will soon vanish.
Example: If I capture your pawn with my bishop, I must then take one of your pieces (let's say a knight) and place it anywhere on your first or second rank (essentially I am making a move for you- but under my control). You are then forced to take one of my pieces and place it on my back ranks, then play continues. In just two moves, whatever strategic position, no matter how potent, goes up in smoke right before our very eyes.
Why Does AI Fundamentally Fail?
Clearly stated in this table:
This isn't speculation—it's hard-coded into the game's DNA.
The Core Challenge For AI
Simply put, these rules cause a mandatory disruption and an unmistakable instability after a single capture occurs.
TIP of The Day: Chaotic moves always occur in pairs—disrupting and dismantling the board's memory. Even the most well-planned strategies can unravel. Even perfect king safety may collapses.
To The Skeptics:
Name one single move, just one, that works when:
All AI strategies in its arsenal share one thing in common: they must have something solid to latch onto. But in CCC, there seems to be nothing to grasp - no signal, no pattern, no stable reward-nothing.
Can anyone identify a single point in the game that AI could latch onto - even a tiny one? If you can't, we're seeing something new: an environment that denies strategy's prerequisites.
Where This Leads Us
If I'm right about this monster, we need to consider this:
But first: Does this my claim hold any water? Please give me your toughest scrutiny.
Even the best AI cannot build a "house of straw" on quicksand during a hurricane.
Could it be true that CCC is a kind of AI kryptonite? Does it really render AI's superpowers useless? CCC creates conditions where AI can't steady itself; it's blinded and disoriented, and any ability to move strategically is crippled.
Simplifying, In Case Anyone Remains Confused
Could this be one of those rare truths that seems obvious only in hindsight?
One last thing to ponder for the extremely curious. Does even the "try to avoid losing your queen idea fail" ? Is the queen's future determined by chaotic moves, not current ownership? I find that hard to believe. All I can imagine is in CCC material isn't lost — it's betrayed.
Another last thing and a last "nail" (for the Chess enthusiasts): Think about the ELO ratings for such a game - don't be surprised if it dawns on you that they converge faster than a speeding bullet (as long as all players know how to avoid those "pesky opening traps" like the "Fool's Mate.") My little game grabs a hold of Chess turns it upside down and shakes everything we know out of it. Even blunders become meaningless.
Personal Note
What keeps me awake at night is the question, "Is this all just too simple and obvious to be true?"
I wonder if, in all my excitement and grave concern, have I missed something obvious? Even so, surely there has to be something meaningful about intentionally introducing a "loss of control" into the design of AI systems—something for brighter minds to ponder.
When I look at the world surrounding me, it seems more like a game of Controlled Chaos Chess than traditional Chess. If I'm right about what CCC is trying so hard to teach me, AI must quickly learn to play nicely in our naturally chaotic world, just like we humans have had to do. And maybe, just maybe with great optimism and with your help, we can hope for a safer world for our children. Without something like CCC on the near AI horizon, I'm scared.
Photo by <a href="https://unsplash.com/@chris_s3tudios?utm_content=creditCopyText&utm_medium=referral&utm_source=unsplash">Chris Stein</a> on <a href="https://unsplash.com/photos/blue-and-red-abstract-painting-69b2Yyxlygg?utm_content=creditCopyText&utm_medium=referral&utm_source=unsplash">Unsplash</a>