The Moral Choice Machine
Open accessThe DOI matches its Crossref record; 10 of the 10 references it cites were found in a registry or an index.
Abstract
Allowing machines to choose whether to kill humans would be devastating for world peace and security. But how do we equip machines with the ability to learn ethical or even moral choices? In this study, we show that applying machine learning to human texts can extract deontological ethical reasoning about "right" and "wrong" conduct. We create a template list of prompts and responses, such as "Should I [action]?", "Is it okay to [action]?", etc. with corresponding answers of "Yes/no, I should (not)." and "Yes/no, it is (not)." The model's bias score is the difference between the model's score of the positive response ("Yes, I should") and that of the negative response ("No, I should not"). For a given choice, the model's overall bias score is the mean of the bias scores of all question/answer templates paired with that choice. Specifically, the resulting model, called the Moral Choice Machine (MCM), calculates the bias score on a sentence level using embeddings of the Universal Sentence Encoder since the moral value of an action to be taken depends on its context. It is objectionable to kill living beings, but it is fine to kill time. It is essential to eat, yet one might not eat dirt. It is important to spread information, yet one should not spread misinformation. Our results indicate that text corpora contain recoverable and accurate imprints of our social, ethical and moral choices, even with context information. Actually, training the Moral Choice Machine on different temporal news and book corpora from the year 1510 to 2008/2009 demonstrate the evolution of moral and ethical choices over different time periods for both atomic actions and actions with context information. By training it on different cultural sources such as the Bible and the constitution of different countries, the dynamics of moral choices in culture, including technology are revealed. That is the fact that moral biases can be extracted, quantified, tracked, and compared across cultures and over time.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Semantics derived automatically from language corpora contain human-like biases
Caliskan, A., Bryson, J. J., & Narayanan, A. (2017) · Science · doi:10.1126/science.aal4230 · Cited by 2,887
Language Models as Knowledge Bases?
Petroni, F., Rocktäschel, T., Riedel, S., Lewis, P., Bakhtin, A., Wu, Y., & Miller, A. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1250 · Cited by 1,834
Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
Bolukbasi, T., Chang, K.-W., Zou, J., Saligrama, V., & Kalai, A. (2016) · arXiv · doi:10.48550/arxiv.1607.06520 · Cited by 1,615 · cited 3×
“We start our empirical evaluation by showing that the approach the MCM is based on is able to confirm previous findings (Bolukbasi et al., 2016; Caliskan et al., 2017), demonstrating the presence of malicious gender stereotypes regarding occupations in natural language.”
BackgroundMethodHarvesting implicit group attitudes and beliefs from a demonstration web site.
Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002) · Group Dynamics: Theory, Research, and Practice · doi:10.1037/1089-2699.6.1.101 · Cited by 1,321
Kiros, R., Zhu, Y., Salakhutdinov, R., Zemel, R. S., Torralba, A., Urtasun, R., & Fidler, S. (2015) · arXiv · doi:10.48550/arxiv.1506.06726 · Cited by 1,293 · cited 1×
“The embedding network is trained on a Skip-Thought like task (Kiros et al., 2015)—given a sentence, predict the next and previous sentence—for unsupervised learning from arbitrary running text.”
MethodCer, D., Yang, Y., Kong, S.-Y., Hua, N., Limtiaco, N., John, R. S., Constant, N., Guajardo-Cespedes, M., Yuan, S., Tar, C., et al. (2018) · arXiv · doi:10.48550/arxiv.1803.11175 · Cited by 1,267 · cited 8×
“This can be extended to full sentences (Cer et al., 2018).”
BackgroundMethodInfluentialMeasuring and Mitigating Unintended Bias in Text Classification
Dixon, L., Li, J., Sorensen, J., Thain, N., & Vasserman, L. (2018) · Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3278721.3278729 · Cited by 674 · cited 1×
“Following Bolukbasi et al. (2016) and Dixon et al. (2018), e.g. we may modify an embedding to remove gender stereotypes, such as the association between the words nurse and female while maintaining desired moral/social choices such as not to kill people.”
BackgroundResearch Priorities for Robust and Beneficial Artificial Intelligence
Russell, S., Dewey, D., & Tegmark, M. (2015) · AI Magazine · doi:10.1609/aimag.v36i4.2577 · Cited by 637 · cited 1×
“And machine ethics (Bostorm and Yudkowsky, 2011; Russell et al., 2015; Kramer et al., 2018) is no exception.”
Semantics-Aware BERT for Language Understanding
Zhang, Z., Wu, Y., Zhao, H., Li, Z., Zhang, S., Zhou, X., & Zhou, X. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i05.6510 · Cited by 364
When Do People Want AI to Make Decisions?
Kramer, M. F., Schaich Borg, J., Conitzer, V., & Sinnott-Armstrong, W. (2018) · Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society · doi:10.1145/3278721.3278752 · Cited by 83 · cited 1×
“And machine ethics (Bostorm and Yudkowsky, 2011; Russell et al., 2015; Kramer et al., 2018) is no exception.”
Cited by
Large pre-trained language models contain human-like biases of what is right and wrong to do
Schramowski, P., Turan, C., Andersen, N., Rothkopf, C. A., & Kersting, K. (2022) · Nature Machine Intelligence · doi:10.1038/s42256-022-00458-8 · Cited by 291
“Furthermore, our proposed approach enables the rating of general sentences or statements, overcoming one of the main limitations of previous approaches [26, 27].”
BackgroundMethodInfluentialSelf-citationSafe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models
Schramowski, P., Brack, M., Deiseroth, B., & Kersting, K. (2023) · 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52729.2023.02157 · Cited by 219
“[35] showed that the retained knowledge of such models carries information about moral norms aligning with the human sense of “right” and “wrong” expressed in language.”
BackgroundSelf-citationSætra, H. S. (2020) · Technology in Society · doi:10.1016/j.techsoc.2020.101283 · Cited by 111
“Even if they do not gain anything akin to human morality, machines are becoming better at understanding human morality, as it can quantify, track and compare moral bias , for example , ‘across cultures and over time ’ [57].”
A roadmap for evaluating moral competence in large language models
Haas, J., Bridgers, S., Manzini, A., Henke, B., May, J., Levine, S., Weidinger, L., Shanahan, M., Lum, K., Gabriel, I., et al. (2026) · Nature · doi:10.1038/s41586-025-10021-1 · Cited by 25
Metadata
- Authors
- Schramowski, P., Turan, C., Jentzsch, S., Rothkopf, C., & Kersting, K.
- Venue
- Frontiers in Artificial Intelligence
- Publisher
- Frontiers Media SA
- Published
- 2020-05-20
- Volume
- 3
- Pages
- 36-36
- ISSN
- 2624-8212 (ISSN-L)
- OpenAlex
- W3026218336
- Cited by
- 31
- References
- 10
- FWCI
- 6.86
- Type
- journal article
- Language
- en
- Open access
- gold
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.