{"slug": "the-first-thing-we-should-do-to-delay-extinction-by-ai", "title": "The First Thing We Should Do to Delay Extinction by AI", "summary": "A new specification and reference implementation propose separating decision authority from AI models by declaring rules as external checklists, having code verify and record verdicts, and letting execution read only the recorded result. The document argues that models are trained to fill in blanks and execute with confidence, so prohibition prompts or better model performance alone will not prevent guesses from becoming execution. The stated goal is to execute exactly what the user has explicitly declared, keeping both decision and responsibility with humans.", "body_md": "— Declare, judge, record, execute — separating authority from the model\n\nPeople talk about a future in which AI destroys humanity. What if AI develops a sense of self and turns hostile toward humans? That is not the problem this document addresses.\n\nThere is, however, something we can do before that moment comes.\n\nWhile AI is still a system that understands and executes human intent, **we can keep it from exercising decision authority on its own.**\n\nWe often attribute the danger of AI to hallucination or faulty reasoning. But there is a more fundamental problem. The model does not stop when it does not know. **It fills the blank itself.**\n\nAll three are the same problem. The model is an engine that fills blanks. This is not a defect. It is trained behavior. Prohibition prompts or better model performance alone will not remove it.\n\n**The model distorts the user’s instruction and then executes with confidence. Confidence expands the scope.**\n\n**A tool will execute as long as its required inputs are satisfied.**\n\nThis is not a problem of the future. It is happening everywhere, right now.\n\nAnd solving it does not necessarily require a smarter AI.\n\n**Separate both what must be decided and the verdict from the AI.**\n\nThe model can still do the work. It asks the user, retrieves information, and prepares what is needed for execution.\n\nRules are declared outside the model as checklists. Code performs the verification and records the verdict. Execution only reads the recorded result.\n\n**This takes back the authority the model has been filling in because nobody declared it.**\n\nBetter model performance increases capability. If decision authority remains with the model, that added capability widens its reach into deciding what was never declared.\n\nIf authority sits outside the model, both the decision and the responsibility remain with humans.\n\nWhat this structure does is stop guesses from becoming execution. The goal is not to execute the model’s judgment, but **to execute exactly what the user has explicitly declared.**\n\nThe detailed design is in the specification, and the reference implementation is in the repository.", "url": "https://wpnews.pro/news/the-first-thing-we-should-do-to-delay-extinction-by-ai", "canonical_source": "https://discuss.huggingface.co/t/the-first-thing-we-should-do-to-delay-extinction-by-ai/180541#post_1", "published_at": "2026-09-17 07:47:51+00:00", "updated_at": "2026-09-17 07:53:56.207150+00:00", "lang": "en", "topics": ["ai-safety", "ai-agents", "ai-policy"], "entities": [], "alternates": {"html": "https://wpnews.pro/news/the-first-thing-we-should-do-to-delay-extinction-by-ai", "markdown": "https://wpnews.pro/news/the-first-thing-we-should-do-to-delay-extinction-by-ai.md", "text": "https://wpnews.pro/news/the-first-thing-we-should-do-to-delay-extinction-by-ai.txt", "jsonld": "https://wpnews.pro/news/the-first-thing-we-should-do-to-delay-extinction-by-ai.jsonld"}}