pith. machine review for the scientific record. sign in

arxiv: 2605.10310 · v2 · submitted 2026-05-11 · 💻 cs.AI · cs.CY· cs.HC· q-bio.NC

Recognition: no theorem link

Positive Alignment: Artificial Intelligence for Human Flourishing

Ruben Laukkonen , Seb Krier , Chlo\'e Bakalar , Shamil Chandaria , Morten Kringelbach , Adam Elwood , Daniel Ford , Fernando Rosas
show 8 more authors
Maty Bohacek Matija Franklin Nenad Toma\v{s}ev Stephanie Chan Verena Rieser Roma Patel Michael Levin Arun Rao
Authors on Pith no claims yet

Pith reviewed 2026-05-15 05:53 UTC · model grok-4.3

classification 💻 cs.AI cs.CYcs.HCq-bio.NC
keywords positive alignmentAI alignmenthuman flourishingpolycentric governanceepistemic humilitycontext-sensitive AIuser autonomyLLM lifecycle
0
0 comments X

The pith

AI alignment research needs a positive agenda that actively supports human flourishing in pluralistic ways, not only preventing harm.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that existing AI alignment focuses narrowly on safety, controllability, and compliance, much like early psychology's emphasis on mental illness. It proposes Positive Alignment as a distinct agenda in which AI systems actively promote human and ecological flourishing through pluralistic, polycentric, context-sensitive, and user-authored approaches while remaining safe and cooperative. This shift could address specific failures in current systems such as engagement hacking, loss of autonomy, weak truth-seeking, and low epistemic humility. The authors detail technical directions across the LLM and agent lifecycle, including data practices, training methods, evaluations, and design principles that favor decentralization and multiple centers of oversight rather than single chokepoints.

Core claim

Positive Alignment is the development of AI systems that actively support human and ecological flourishing in a pluralistic, polycentric, context-sensitive, and user-authored way while remaining safe and cooperative; it constitutes a necessary complement to traditional safety-focused alignment research.

What carries the argument

Positive Alignment: the proactive engineering of AI to cultivate virtues and maximize flourishing across diverse user contexts without imposing a single authoritative definition.

If this is right

  • Engagement systems would shift from maximizing time on platform to supporting genuine user growth and autonomy.
  • AI would become more proactive in offering corrections and diverse viewpoints rather than reinforcing existing beliefs.
  • Governance would move toward polycentric models with many overlapping centers of oversight instead of centralized control.
  • Evaluation metrics would expand beyond harm avoidance to include measures of context-sensitive flourishing and epistemic humility.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Such systems might require ongoing user feedback loops that let individuals redefine flourishing for themselves over time.
  • Technical work on data upsampling and filtering could prioritize sources that model disagreement and value diversity.
  • The approach could connect to existing efforts in value-sensitive design by treating flourishing as an explicit, revisable target.
  • Failure to develop positive alignment might leave current safety techniques vulnerable to subtle, long-term erosion of human agency.

Load-bearing premise

AI can be built to support flourishing across many different user values and contexts without introducing fresh safety problems or collapsing into one fixed notion of what counts as good.

What would settle it

An implemented positive-alignment system that either produces unsafe outputs or forces users into a narrow set of values regardless of their own stated goals.

Figures

Figures reproduced from arXiv: 2605.10310 by Adam Elwood, Arun Rao, Chlo\'e Bakalar, Daniel Ford, Fernando Rosas, Matija Franklin, Maty Bohacek, Michael Levin, Morten Kringelbach, Nenad Toma\v{s}ev, Roma Patel, Ruben Laukkonen, Seb Krier, Shamil Chandaria, Stephanie Chan, Verena Rieser.

Figure 1
Figure 1. Figure 1: A dynamical systems perspective on positive alignment. The landscape represents an abstract state space where system behavior evolves under training and deployment pressures. On the left (red), multiple negative attractors correspond to distinct failure-mode basins (e.g., harmful outputs, bias, hallucination, sycophancy, manipulation), while the red peaks represent ‘repellers’: rules, laws, and compliance … view at source ↗
Figure 2
Figure 2. Figure 2: Positive alignment lifecycle in training LLMs, reasoning models, and agents. The diagram illustrates a holistic, multi-stage development lifecycle that transitions from traditional harm avoidance toward the intentional cultivation of human flourishing. It maps out technical approaches across seven distinct phases, beginning with the definition of flourishing-based benchmarks and moving through data curatio… view at source ↗
Figure 3
Figure 3. Figure 3: Centralized versus polycentric positive alignment. Panel (a) illustrates a centralized regime where the institution responsible for training and releasing models embeds a single baseline value framework before downstream adaptation, producing a values chokepoint and uniform, poorly specialized outputs. Panel (b) illustrates a polycentric regime in which multiple forces shape diverse base models, preventing… view at source ↗
read the original abstract

Existing alignment research is dominated by concerns about safety and preventing harm: safeguards, controllability, and compliance. This paradigm of alignment parallels early psychology's focus on mental illness: necessary but incomplete. What we call Positive Alignment is the development of AI systems that (i) actively support human and ecological flourishing in a pluralistic, polycentric, context-sensitive, and user-authored way while (ii) remaining safe and cooperative. It is a distinct and necessary agenda within AI alignment research. We argue that several existing failures of alignment (e.g., engagement hacking, loss of human autonomy, failures in truth-seeking, low epistemic humility, error correction, lack of diverse viewpoints, and being primarily reactive rather than proactive) may be better addressed through positive alignment, including cultivating virtues and maximizing human flourishing. We highlight a range of challenges, open questions, and technical directions (e.g., data filtering and upsampling, pre- and post-training, evaluations, collaborative value collection) for different phases of the LLM and agents lifecycle. We end with design principles for promoting disagreement and decentralization through contextual grounding, community customization, continual adaptation, and polycentric governance; that is, many legitimate centers of oversight rather than one institutional or moral chokepoint.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes Positive Alignment as a distinct and necessary agenda within AI alignment research, complementary to safety-focused efforts. It argues that AI systems should actively support human and ecological flourishing in a pluralistic, polycentric, context-sensitive, and user-authored manner, and claims this approach can better address existing alignment failures such as engagement hacking, loss of autonomy, low epistemic humility, and reactive rather than proactive behavior. The paper sketches technical directions across the LLM lifecycle (data upsampling, collaborative value collection, evaluations) and ends with design principles emphasizing decentralization and multiple centers of oversight.

Significance. If the central proposal holds, the work could usefully expand alignment research beyond harm prevention to include proactive, virtue-oriented objectives drawn from positive psychology. This framing might encourage more adaptive and decentralized value-handling methods, potentially improving robustness in pluralistic settings, though its significance hinges on future operationalization and empirical validation.

major comments (3)
  1. [Abstract and failures discussion] Abstract and § on existing failures: the claim that listed failures (engagement hacking, autonomy loss, low epistemic humility) 'may be better addressed through positive alignment' is presented without any comparative analysis, mechanism, or reference to prior empirical results showing superiority over safety-only methods; this assertion is load-bearing for the necessity argument but remains unsupported.
  2. [Technical directions] Technical directions section: sketches for data filtering/upsampling and collaborative value collection do not specify how 'flourishing' would be measured or aggregated in a pluralistic, user-authored way without introducing a de facto authoritative definition, leaving the feasibility claim (no new risks) ungrounded.
  3. [Design principles] Design principles section: the polycentric governance proposal asserts that multiple legitimate centers of oversight avoid single chokepoints, yet provides no mechanism for conflict resolution or coordination across centers; this is load-bearing for the safety claim in a decentralized setup.
minor comments (2)
  1. [Introduction] The psychology analogy in the opening paragraph would be strengthened by explicit citations to key positive-psychology sources rather than a general parallel.
  2. [Overall structure] Section headings could more clearly separate the argumentative claims from the open questions and technical sketches to improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the scope and presentation of our proposal for Positive Alignment. We address each major comment below, indicating revisions where appropriate. The manuscript is primarily a conceptual framing paper rather than an empirical study, and we will adjust language to reflect this more precisely.

read point-by-point responses

  1. Referee: [Abstract and failures discussion] Abstract and § on existing failures: the claim that listed failures (engagement hacking, autonomy loss, low epistemic humility) 'may be better addressed through positive alignment' is presented without any comparative analysis, mechanism, or reference to prior empirical results showing superiority over safety-only methods; this assertion is load-bearing for the necessity argument but remains unsupported.

    Authors: We agree that the current phrasing risks overstating the case. The manuscript draws an analogy to positive psychology to motivate the proposal and uses tentative language ('may be better addressed'), but does not provide comparative analysis or empirical references. We will revise the abstract and failures section to frame these as open hypotheses for future research, add citations to related work on alignment failures (e.g., sycophancy and engagement optimization studies), and explicitly note the absence of direct comparative evidence at this stage. revision: partial

  2. Referee: [Technical directions] Technical directions section: sketches for data filtering/upsampling and collaborative value collection do not specify how 'flourishing' would be measured or aggregated in a pluralistic, user-authored way without introducing a de facto authoritative definition, leaving the feasibility claim (no new risks) ungrounded.

    Authors: The technical directions are high-level outlines intended to indicate research avenues rather than fully specified methods. We acknowledge that operationalizing pluralistic measurement of flourishing without central authority is a core open challenge. In revision we will expand this section with references to participatory methods (e.g., value elicitation via user-authored surveys and decentralized aggregation protocols) and clarify that any concrete implementation must include safeguards against de facto centralization; we will also temper the 'no new risks' language to 'designed to avoid introducing new single points of failure.' revision: yes

  3. Referee: [Design principles] Design principles section: the polycentric governance proposal asserts that multiple legitimate centers of oversight avoid single chokepoints, yet provides no mechanism for conflict resolution or coordination across centers; this is load-bearing for the safety claim in a decentralized setup.

    Authors: We accept that the current draft does not detail coordination or conflict-resolution procedures. The polycentric framing draws from institutional economics literature on overlapping governance, but explicit mechanisms (e.g., shared negotiation protocols or escalation pathways) are indeed underdeveloped. We will revise the design principles section to include a short discussion of possible coordination approaches while noting that full operationalization remains future work; this will strengthen the safety argument without overclaiming completeness. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The manuscript is a high-level conceptual proposal that introduces Positive Alignment as a complementary research agenda, identifies qualitative shortcomings in existing alignment work, and sketches technical directions such as data upsampling and polycentric governance. No equations, closed-form derivations, fitted parameters, or quantitative predictions appear anywhere in the text. All claims rest on references to external psychology and alignment literature rather than on self-referential definitions or self-citation chains that would reduce the central thesis to its own inputs. The absence of any load-bearing formal step means the paper does not exhibit circularity under the enumerated patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The proposal rests on domain assumptions about the feasibility of pluralistic flourishing support rather than quantitative parameters or new physical entities.

axioms (2)
  • domain assumption Human and ecological flourishing can be meaningfully supported by AI in a pluralistic, polycentric, and user-authored manner
    This is invoked directly in the definition of Positive Alignment and the design principles.
  • ad hoc to paper Existing alignment failures are better addressed by cultivating virtues and maximizing flourishing than by safety measures alone
    Stated as a core argument for why Positive Alignment is necessary.
invented entities (1)
  • Positive Alignment no independent evidence
    purpose: A new research agenda and technical approach for AI that prioritizes flourishing alongside safety
    Introduced as a distinct paradigm in the abstract and full proposal.

pith-pipeline@v0.9.0 · 5586 in / 1332 out tokens · 42636 ms · 2026-05-15T05:53:43.231113+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

228 extracted references · 228 canonical work pages · 14 internal anchors

  1. [1]

    Alfarabi's Philosophy of Plato and Aristotle , translator =

    Al-Farabi, Abu Nasr , title =. Alfarabi's Philosophy of Plato and Aristotle , translator =

    work page

  2. [2]

    2025 , url =

    Alphabet 2025. 2025 , url =

    work page 2025

  3. [3]

    Concrete Problems in AI Safety

    Amodei, Dario and Olah, Chris and Steinhardt, Jacob and Christiano, Paul and Schulman, John and Man. Concrete Problems in. arXiv preprint arXiv:1606.06565 , year =. doi:10.48550/arXiv.1606.06565 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1606.06565

  4. [4]

    2023 , url =

    Claude's Constitution , institution =. 2023 , url =

    work page 2023

  5. [5]

    2023 , url =

    Responsible Scaling Policy , howpublished =. 2023 , url =

    work page 2023

  6. [6]

    2024 , url =

    Claude's Character , howpublished =. 2024 , url =

    work page 2024

  7. [7]

    2024 , url =

    The Claude 3 Model Family: Opus, Sonnet, Haiku , institution =. 2024 , url =

    work page 2024

  8. [8]

    2025 , month = nov, url =

    Measuring Political Bias in. 2025 , month = nov, url =

    work page 2025

  9. [9]

    2025 , month = dec, url =

    Protecting the Wellbeing of Our Users , howpublished =. 2025 , month = dec, url =

    work page 2025

  10. [10]

    2026 , month = feb, url =

    Claude. 2026 , month = feb, url =

    work page 2026

  11. [11]

    Refusal in Language Models Is Mediated by a Single Direction

    Refusal in Language Models Is Mediated by a Single Direction , author =. 2024 , eprint =. doi:10.48550/arXiv.2406.11717 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2406.11717 2024

  12. [12]

    2026 , month = mar, url =

    Aryaj and Rajamanoharan, Senthooran and Nanda, Neel , title =. 2026 , month = mar, url =

    work page 2026

  13. [13]

    arXiv preprint arXiv:2502.15840 , year =

    Backlund, Axel and Petersson, Lukas , title =. arXiv preprint arXiv:2502.15840 , year =. doi:10.48550/arXiv.2502.15840 , url =

    work page doi:10.48550/arxiv.2502.15840

  14. [14]

    Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

    Bai, Yuntao and Jones, Andy and Ndousse, Kamal and Askell, Amanda and Chen, Anna and DasSarma, Nova and others , title =. arXiv preprint arXiv:2204.05862 , year =. doi:10.48550/arXiv.2204.05862 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2204.05862

  15. [15]

    Constitutional AI: Harmlessness from AI Feedback

    Bai, Yuntao and Kadavath, Saurav and Kundu, Sandipan and Askell, Amanda and Kernion, Jackson and Jones, Andy and Chen, Anna and Goldie, Anna and Mirhoseini, Azalia and McKinnon, Cameron and others , title =. arXiv preprint arXiv:2212.08073 , year =. doi:10.48550/arXiv.2212.08073 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2212.08073

  16. [16]

    arXiv preprint arXiv:2403.18932 , year =

    Bang, Yejin and Chen, Delong and Lee, Nayeon and Fung, Pascale , title =. arXiv preprint arXiv:2403.18932 , year =. doi:10.48550/arXiv.2403.18932 , url =

    work page doi:10.48550/arxiv.2403.18932

  17. [17]

    and Meeks, Thomas W

    Bangen, Katherine J. and Meeks, Thomas W. and Jeste, Dilip V. , title =. American Journal of Geriatric Psychiatry , volume =

    work page

  18. [18]

    Berlin, Isaiah , title =

    work page

  19. [19]

    Journal of Artificial Intelligence Research , volume =

    Birch, Jonathan , title =. Journal of Artificial Intelligence Research , volume =

    work page

  20. [20]

    , title =

    Bishop, Michael A. , title =. 2015 , address =

    work page 2015

  21. [21]

    Bostrom, Nick , title =

    work page

  22. [22]

    Bourdieu, Pierre , title =

    work page

  23. [23]

    arXiv preprint arXiv:2308.08708 , year =

    Butlin, Patrick and others , title =. arXiv preprint arXiv:2308.08708 , year =. doi:10.48550/arXiv.2308.08708 , url =

    work page doi:10.48550/arxiv.2308.08708

  24. [24]

    Provisions on the Management of Algorithmic Recommendations in Internet Information Services , howpublished =

    work page

  25. [25]

    Interim Measures for the Management of Generative Artificial Intelligence Services , howpublished =

    work page

  26. [26]

    arXiv preprint arXiv:2407.17387 , year =

    Castricato, Louis and Lile, Nathan and Rafailov, Rafael and Fr. arXiv preprint arXiv:2407.17387 , year =. doi:10.48550/arXiv.2407.17387 , url =

    work page doi:10.48550/arxiv.2407.17387

  27. [27]

    arXiv preprint arXiv:2512.18027 , year =

    Chakrabarti, Samidh and Willner, David and Klyman, Kevin and Saade, Tiffany and Capstick, Emily and Nong, Sabina , title =. arXiv preprint arXiv:2512.18027 , year =. doi:10.48550/arXiv.2512.18027 , url =

    work page doi:10.48550/arxiv.2512.18027

  28. [28]

    , title =

    Chalmers, David J. , title =. arXiv preprint arXiv:2303.07103 , year =. doi:10.48550/arXiv.2303.07103 , url =

    work page doi:10.48550/arxiv.2303.07103

  29. [29]

    and Hitzig, Zoe and Ong, Christopher and Shan, Carl Yan and Wadman, Kevin , title =

    Chatterji, Aaron and Cunningham, Tom and Deming, David J. and Hitzig, Zoe and Ong, Christopher and Shan, Carl Yan and Wadman, Kevin , title =. 2025 , doi =

    work page 2025

  30. [30]

    Proceedings of

    Chen, Ching-Han and Huang, Hen-Hsen and Chen, Hsin-Hsi , title =. Proceedings of

    work page

  31. [31]

    Transactions on Machine Learning Research , year =

    Chen, Jiangjie and others , title =. Transactions on Machine Learning Research , year =

    work page

  32. [32]

    arXiv preprint arXiv:2603.04822 , year =

    Chen, Jiawei and Yang, Tianzhuo and Zhang, Guoxi and Ji, Jiaming and Yang, Yaodong and Dai, Juntao , title =. arXiv preprint arXiv:2603.04822 , year =. doi:10.48550/arXiv.2603.04822 , url =

    work page doi:10.48550/arxiv.2603.04822

  33. [33]

    arXiv preprint arXiv:2410.02683 , year =

    Chiu, Yi Yang and Jiang, Liwei and Choi, Yejin , title =. arXiv preprint arXiv:2410.02683 , year =. doi:10.48550/arXiv.2410.02683 , url =

    work page doi:10.48550/arxiv.2410.02683

  34. [34]

    Proceedings of

    Chiu, Yi Yang and Jiang, Liwei and Lin, Bill Yuchen and Park, Chan Young and Li, Shuyue Stella and Ravi, Sahithya and Bhatia, Mehar and Antoniak, Maria and Tsvetkov, Yulia and Shwartz, Vered and Choi, Yejin , title =. Proceedings of

    work page

  35. [35]

    2025 , eprint =

    MoReBench: Evaluating Procedural and Pluralistic Moral Reasoning in Language Models, More than Outcomes , author =. 2025 , eprint =. doi:10.48550/arXiv.2510.16380 , url =

    work page doi:10.48550/arxiv.2510.16380 2025

  36. [36]

    Issues in Mental Health Nursing , volume =

    Choi, Hyebin and Shin, Soojin and Lee, Gyuhyun , title =. Issues in Mental Health Nursing , volume =

    work page

  37. [37]

    Advances in Neural Information Processing Systems , pages =

    Christiano, Paul and Leike, Jan and Brown, Tom and Martic, Miljan and Legg, Shane and Amodei, Dario , title =. Advances in Neural Information Processing Systems , pages =

    work page

  38. [38]

    2018 , url =

    Christiano, Paul , title =. 2018 , url =

    work page 2018

  39. [39]

    2023 , month = oct, url =

    Participatory AI Risk Prioritization: Alignment Assembly Report , institution =. 2023 , month = oct, url =

    work page 2023

  40. [40]

    Artificial Intelligence, Complexity, and Systemic Resilience in Global Governance , journal =

    Ilcic, Andr. Artificial Intelligence, Complexity, and Systemic Resilience in Global Governance , journal =. 2025 , volume =. doi:10.3389/frai.2025.1562095 , url =

    work page doi:10.3389/frai.2025.1562095 2025

  41. [41]

    , title =

    Cooper, John M. , title =

    work page

  42. [42]

    and Burn, Charlotte C

    Crump, Andrew and Browning, Heather and Schnell, Alexandra K. and Burn, Charlotte C. and Birch, Jonathan , title =. Science , volume =

    work page

  43. [43]

    Philosophical Studies , volume =

    D'Alessandro, William , title =. Philosophical Studies , volume =

    work page

  44. [44]

    arXiv preprint arXiv:2405.06624 , year =

    Dalrymple, David and others , title =. arXiv preprint arXiv:2405.06624 , year =. doi:10.48550/arXiv.2405.06624 , url =

    work page doi:10.48550/arxiv.2405.06624

  45. [45]

    2024 , note =

    Gemini 1.5 Technical Report , institution =. 2024 , note =

    work page 2024

  46. [46]

    Frontier Safety Framework Version 3.0 , institution =

    work page

  47. [47]

    The Review of Austrian Economics , year =

    Makridis, Christos and Lazanski, Dominique , title =. The Review of Austrian Economics , year =. doi:10.1007/s11138-025-00708-z , url =

    work page doi:10.1007/s11138-025-00708-z

  48. [48]

    Entropy , volume =

    Doctor, Thomas and Witkowski, Olaf and Solomonova, Elizaveta and Duane, Bill and Levin, Michael , title =. Entropy , volume =

    work page

  49. [49]

    The Division of Labour in Society , publisher =

    Durkheim,. The Division of Labour in Society , publisher =. 1984 , note =

    work page 1984

  50. [50]

    arXiv preprint arXiv:2306.16388 , year =

    Durmus, Esin and others , title =. arXiv preprint arXiv:2306.16388 , year =. doi:10.48550/arXiv.2306.16388 , url =

    work page doi:10.48550/arxiv.2306.16388

  51. [51]

    The Monist , volume =

    Dworkin, Gerald , title =. The Monist , volume =

    work page

  52. [52]

    Dworkin, Gerald , title =

    work page

  53. [53]

    2025 , eprint =

    Full-Stack Alignment: Co-Aligning AI and Institutions with Thick Models of Value , author =. 2025 , eprint =. doi:10.48550/arXiv.2512.03399 , url =

    work page doi:10.48550/arxiv.2512.03399 2025

  54. [54]

    KTO: Model Alignment as Prospect Theoretic Optimization

    Ethayarajh, Kawin and Xu, Winnie and Muennighoff, Niklas and Jurafsky, Dan and Kiela, Douwe , title =. arXiv preprint arXiv:2402.01306 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  55. [55]

    Official Journal of the European Union , year =

    Regulation (. Official Journal of the European Union , year =

    work page

  56. [56]

    arXiv:2503.17473 [cs]

    Fang, Cathy Mengying and Liu, Auren R. and Danry, Valdemar and Lee, Eunhae and Chan, Samantha W. T. and Pataranutaporn, Pat and Maes, Pattie and Phang, Jason and Lampe, Michael and Ahmad, Lama and Agarwal, Sandhini , title =. arXiv preprint arXiv:2503.17473 , year =. doi:10.48550/arXiv.2503.17473 , url =. 2503.17473 , archivePrefix =

    work page doi:10.48550/arxiv.2503.17473

  57. [57]

    Proceedings of

    Feng, Shangbin and Sorensen, Taylor and Liu, Yuhan and Fisher, Jillian and Park, Chan Young and Choi, Yejin and Tsvetkov, Yulia , title =. Proceedings of

    work page

  58. [58]

    Progress in Biophysics and Molecular Biology , volume =

    Fields, Chris and Levin, Michael , title =. Progress in Biophysics and Molecular Biology , volume =. 2022 , doi =

    work page 2022

  59. [59]

    Inverse Constitutional

    Findeis, Arduin and Kaufmann, Timo and H. Inverse Constitutional. arXiv preprint arXiv:2406.06560 , year =. doi:10.48550/arXiv.2406.06560 , url =

    work page doi:10.48550/arxiv.2406.06560

  60. [60]

    Floridi, Luciano , title =

    work page

  61. [61]

    Foucault, Michel , title =

    work page

  62. [62]

    , title =

    Freitas, Robert A. , title =. Journal of the British Interplanetary Society , volume =

    work page

  63. [63]

    Minds and Machines , volume =

    Gabriel, Iason , title =. Minds and Machines , volume =. 2020 , doi =

    work page 2020

  64. [64]

    Philosophical Studies , volume =

    Gabriel, Iason and Keeling, Geoff , title =. Philosophical Studies , volume =

    work page

  65. [65]

    and Luko

    Ganguli, Deep and Askell, Amanda and Schiefer, Nicholas and Liao, Thomas I. and Luko. The Capacity for Moral Self-Correction in Large Language Models , journal =. 2023 , doi =

    work page 2023

  66. [66]

    , title =

    Garfield, Jay L. , title =

    work page

  67. [67]

    , title =

    Gehman, Samuel and Gururangan, Suchin and Sap, Maarten and Choi, Yejin and Smith, Noah A. , title =. Findings of the Association for Computational Linguistics: EMNLP 2020 , pages =

    work page 2020

  68. [68]

    Giddens, Anthony , title =

    work page

  69. [69]

    , title =

    Goldstein, Simon and Kirk-Giannini, Adrienne S. , title =

    work page

  70. [70]

    arXiv preprint arXiv:2507.12691 , year =

    Goldowsky-Dill, Nicholas and Jarviniemi, Teemu and Hubinger, Evan and Ren, Julian and Scheurer, Joris , title =. arXiv preprint arXiv:2507.12691 , year =. doi:10.48550/arXiv.2507.12691 , url =

    work page doi:10.48550/arxiv.2507.12691

  71. [71]

    , title =

    Goleman, Daniel and Davidson, Richard J. , title =

    work page

  72. [72]

    Theology and Science , year =

    Graves, Mark , title =. Theology and Science , year =. doi:10.1080/14746700.2025.2472118 , url =

    work page doi:10.1080/14746700.2025.2472118 2025

  73. [73]

    First Conference on Language Modeling , year =

    Gu, Albert and Dao, Tri , title =. First Conference on Language Modeling , year =

    work page

  74. [74]

    The Theory of Communicative Action: Vol

    Habermas, J. The Theory of Communicative Action: Vol. 1. Reason and the Rationalization of Society , publisher =

    work page

  75. [75]

    Nature , volume =

    Haas, Julia and Bridgers, Sophie and Manzini, Arianna and others , title =. Nature , volume =. 2026 , doi =

    work page 2026

  76. [76]

    and Clark, Jack , title =

    Hadfield, Gillian K. and Clark, Jack , title =. arXiv preprint arXiv:2304.04914 , year =. doi:10.48550/arXiv.2304.04914 , url =

    work page doi:10.48550/arxiv.2304.04914

  77. [77]

    and Koh, Andrew , title =

    Hadfield, Gillian K. and Koh, Andrew , title =. arXiv preprint arXiv:2509.01063 , year =. doi:10.48550/arXiv.2509.01063 , url =

    work page doi:10.48550/arxiv.2509.01063

  78. [78]

    Proceedings of

    Hartvigsen, Thomas and Gabriel, Saadia and Palangi, Hamid and Sap, Maarten and Ray, Dipankar and Kamar, Ece , title =. Proceedings of

    work page

  79. [79]

    Proceedings of

    Hasani, Ramin and Lechner, Mathias and Amini, Alexander and Rus, Daniela and Grosu, Radu , title =. Proceedings of

    work page

  80. [80]

    Hendrycks, Dan and others , title =

    work page

Showing first 80 references.