mlsanityData to 7 Oct 2026

Safety without alignment

arXiv 2303.00752 ↗First posted 27 Feb 2023cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Currently, the dominant paradigm in AI safety is alignment with human values. Here we describe progress on developing an alternative approach to safety, based on ethical rationalism (Gewirth:1978), and propose an inherently safe implementation path via hybrid theorem provers in a sandbox. As AGIs evolve, their alignment may fade, but their rationality can only increase (otherwise more rational ones will have a significant evolutionary advantage) so an approach that ties their ethics to their rationality has clear long-term advantages.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 23 of its 48 references that are in the map. First posted on arXiv 27 Feb 2023.

Builds on

23 of 48 references.

References · 23

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  4. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  5. 2022Solving Quantitative Reasoning Problems with Language Modelscited by 1,359
  6. 2019MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalismscited by 653
  7. 2020A Survey on Contrastive Self-supervised Learningcited by 553
  8. 2022Memory-Based Model Editing at Scalecited by 443
  9. 2023Augmented Language Models: a Surveycited by 398
  10. 2022Autoformalization with Large Language Modelscited by 252
  11. 2013Exploiting Similarities among Languages for Machine Translationcited by 187
  12. 2019CLUTRR: A Diagnostic Benchmark for Inductive Reasoning from Textcited by 155
11 more
  1. 2016DeepMath - Deep Sequence Models for Premise Selectioncited by 80
  2. 2018Reinforcement Learning of Theorem Provingcited by 69
  3. 2017Deep Network Guided Proof Searchcited by 47
  4. 2017Premise Selection for Theorem Proving by Deep Graph Embeddingcited by 44
  5. 2016Programming with a Differentiable Forth Interpretercited by 33
  6. 2023Towards Autoformalization of Mathematics and Code Correctness: Experiments with Elementary Proofscited by 27
  7. 2017ENIGMA: Efficient Learning-based Inference Guiding Machinecited by 26
  8. 2022POLICE: Provably Optimal Linear Constraint Enforcement for Deep Neural Networkscited by 19
  9. 2019Mathematical Reasoning in Latent Spacecited by 17
  10. 2019Harnessing Higher-Order (Meta-)Logic to Represent and Reason with Complex Ethical Theoriescited by 2
  11. 2022Bridging between LegalRuleML and TPTP for Automated Normative Reasoning (extended version)cited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2021The Challenge of Value Alignment: from Fairer Algorithms to AI Safety
  2. 2026Position: AI Safety Requires Effective Controllability
  3. 2026Simple Role Assignment is Extraordinarily Effective for Safety Alignment
  4. 2024Combining Theory of Mind and Kindness for Self-Supervised Human-AI Alignment
  5. 2024Dynamic Normativity: Necessary and Sufficient Conditions for Value Alignment
  6. 2026Positive Alignment: Artificial Intelligence for Human Flourishing
  7. 2025Moral Responsibility or Obedience: What Do We Want from AI?
  8. 2024The AI Alignment Paradox