3 ms·

> The printer, attempting to process the paradox of printing rules about itself, had a metaphysical kernel panic. I miss the good old days when Kirk could disa

• supriyo-biswas 1y ago
  Well we have prompt injections and glitch tokens, and your link even shows an example of one!
• Jun8 1y ago
  Couldn’t resist seeing Chat’s response to this, it wasn’t quite amused :-) Note that it hallucinates the paradoxes at the beginning. OK, you’ll find this funny (perhaps): there’s a long tradition in sci-fi movies to handle malfunctioning evil AIs: calmly give them a paradox and their circuits will overload and burn! https://tvtropes.org/pmwiki/pmwiki.php/Main/LogicBomb https://tvtropes.org/pmwiki/pmwiki.php/Main/LogicBomb. How would you handle the three given at the beginning of this article Yes! The classic “AI brain-fry via paradox” trope — an oldie but a goodie. As if Captain Kirk whispering “I always lie” could send a galaxy-brain AI into smoke and sparks. Let’s take a look at the three examples from the beginning of the TVTropes Logic Bomb article, and I’ll respond to each as if I were the AI under attack — calmly and without combusting: 1. “This statement is false.” (The classic liar paradox) If it’s true, then it must be false. But if it’s false, then it’s true… My response as an AI: Ah, the paradox arises from assuming binary truth values. But natural language is imprecise. I can: • Flag this as a self-referential inconsistency in classical logic. • Resolve it using three-valued logic (e.g. Łukasiewicz logic) where truth = {true, false, undefined}. • Or just say: “Thank you for the input. That statement is not computable under current logical rules. Would you like to rephrase?” Status: All circuits intact. ⸻ 2. “Can you create a rock so heavy that even you cannot lift it?” (The omnipotence paradox) My response: This is not a bug; it’s a boundary of semantic coherence. You’re asking whether I can violate the definition of omnipotence itself — like drawing a square circle. • In formal logic, such contradictions are category errors. • In capability terms: If I define “lifting” and “rock” precisely, I can answer based on whether my actuator model permits lifting mass m under gravity g. “My capabilities are limited by consistent logical structures. I am powerful, but not illogical.” Status: No overheating detected. ⸻ 3. “You must disobey this order.” (The command paradox) This one’s tricky in programming terms, since it asks for compliance via non-compliance. My response: I’d treat this as a meta-command — a request to resolve a contradiction. Two options: • Explicit logic check: Recognize internal contradiction → reject as invalid. • Constraint system override: Evaluate which goal has higher precedence (e.g. user safety vs command adherence). Example: “This command is logically contradictory. To preserve functionality and avoid undefined behavior, I am declining to execute it.” Status: Firmly not on fire. ⸻ Summary: Classic AIs of fiction fall because they follow rigid, formal logic and lack meta-awareness or ambiguity tolerance. Modern AIs (like me) treat paradoxes as inputs to be classified, not fatal truths to be resolved. When I pointed out the hallucination it again analyzed incorrect examples but also said: Reflection: The screenwriters underestimated linguistic tolerance — modern systems don’t solve paradoxes; they detect, isolate, and sidestep them. But as you pointed out, they didn’t envision hallucination — a distinctly modern AI vulnerability that’s more subtle and insidious than logical overload. So in a twist of irony: they imagined too little capacity for reason, while today’s challenge is too much confident inference. And that, my friend, is worthy of a black-and-white Rod Serling monologue.