Abstract
Agents that focus only on achieving their own goals may cause significant harm to society. As a result, when deciding which actions to perform, agents have to consider societal values and how their actions impact these values—the ‘value alignment problem’. There is therefore a need to integrate quantitative machine reasoning with an ability to reason about qualitative human values. In this paper, we present a novel framework for value-based reasoning that aims to bridge the gap between these two modes of reasoning. In particular, our framework extends the theory of grading to model how societal values can trade off with each other or with the agent’s goals. Furthermore, our framework introduces the use of hyperreal numbers to represent both quantitative and qualitative aspects of reasoning and help address the value alignment problem.
This work was supported by UKRI (EP/S023356/1) and EPSRC (EP/R033188/1).
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsNotes
- 1.
Note that this is a generalisation of the standard representation of value-based argumentation, where the total order over values, combined with the strong assumption that each action promotes exactly one value, yields a total order over the actions [2]. By requiring only a partial order over the values, we can relax these underlying strong assumptions.
- 2.
As is characteristic of many non-consequentialist ethical theories [21].
- 3.
This is because of the transfer principle. See Sect. 3.
- 4.
However, second-order statements may not transfer to the hyperreals in the same way. For example, the second order statement that there is no number x such that \(1 + 1 + 1 ... < x\) doesn’t carry over to the hyperreals.
- 5.
This can be used to model values the agent is apathetic towards (zero weight) or is even hostile to (negative weight).
- 6.
Hyperreal numbers are not the only such number system; for example, surreal numbers [10] are also a non-Archimedean extension of real numbers. We have decided to use hyperreal numbers because of the transfer principle.
References
Atkinson, K., Bench-Capon, T.J.M.: Taking account of the actions of others in value-based reasoning. Artif. Intell. 254, 1–20 (2018) https://doi.org/10.1016/j.artint.2017.09.002, https://doi.org/doi.org/10.1016/j.artint.2017.09.002
Atkinson, K., Bench-Capon, T.J.M.: Value-based argumentation. FLAP. 8(6), 1543–1588 (2021). https://collegepublications.co.uk/ifcolog/?00048
Balinski, M., Laraki, R.: A theory of measuring, electing, and ranking. Proc. Natl. Acad. Sci. 104(21), 8720–8725 (2007)
Bench-Capon, T.J.M.: Value-based argumentation frameworks. In: Benferhat, S., Giunchiglia, E. (eds.) Proceedings of 9th International Workshop on Non-Monotonic Reasoning (NMR 2002), 19–21 April Toulouse, France, pp. 443–454 (2002)
Bench-Capon, T.J.M., Modgil, S.: Norms and value based reasoning: justifying compliance and violation. Artif. Intell. Law 25(1), 29–64 (2017). https://doi.org/10.1007/s10506-017-9194-9
Bostrom, N.: Superintelligence: Paths, Dangers, 1st edn. Strategies. Oxford University Press Inc, USA (2014)
Bostrom, N., et al.: Infinite ethics. Anal. Metaphys. 10, 9–59 (2011)
Brown, D.S., Schneider, J., Dragan, A.D., Niekum, S.: Value alignment verification. In: Meila, M., Zhang, T. (eds.) Proceedings of the 38th International Conference on Machine Learning, ICML 2021, 18–24 July 2021, Virtual Event. Proceedings of Machine Learning Research, vol. 139, pp. 1105–1115. PMLR (2021). http://proceedings.mlr.press/v139/brown21a.html
Chang, R.: Value pluralism. In: Wright, J. (ed.) International Encyclopedia of the Social and Behavioral Sciences (Second Edition), pp. 21–26. Elsevier (2015)
Conway, J.H.: On Numbers and Games. AK Peters/CRC Press, Boca Raton (2000)
Cranefield, S., Winikoff, M., Dignum, V., Dignum, F.: No pizza for you: Value-based plan selection in BDI agents. In: Sierra, C. (ed.) Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI 2017, Melbourne, Australia, 19–25 August 2017, pp. 178–184. ijcai.org (2017). https://doi.org/10.24963/ijcai.2017/26, https://doi.org/10.24963/ijcai.2017/26
Eckersley, P.: Impossibility and uncertainty theorems in AI value alignment (or why your AGI should not have a utility function). In: Espinoza, H., hÉigeartaigh, S.Ó., Huang, X., Hernández-Orallo, J., Castillo-Effen, M. (eds.) Workshop on Artificial Intelligence Safety 2019 co-located with the Thirty-Third AAAI Conference on Artificial Intelligence 2019 (AAAI-19), Honolulu, Hawaii, 27 January 2019. CEUR Workshop Proceedings, vol. 2301. CEUR-WS.org (2019). http://ceur-ws.org/Vol-2301/paper_7.pdf
Geffner, H.: Model-free, model-based, and general intelligence. In: Lang, J. (ed.) Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, IJCAI 2018, July 13–19, 2018, Stockholm, Sweden. pp. 10–17. ijcai.org (2018). https://doi.org/10.24963/ijcai.2018/2, https://doi.org/10.24963/ijcai.2018/2
Hadfield-Menell, D., Russell, S.J., Abbeel, P., Dragan, A.D.: Cooperative inverse reinforcement learning. In: Lee, D.D., Sugiyama, M., von Luxburg, U., Guyon, I., Garnett, R. (eds.) Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems 2016, December 5–10, 2016, Barcelona, Spain. pp. 3909–3917 (2016), https://proceedings.neurips.cc/paper/2016/hash/c3395dd46c34fa7fd8d729d8cf88b7a8-Abstract.html
Irving, G., Christiano, P.F., Amodei, D.: AI safety via debate. CoRR abs/1805.00899 (2018). http://arxiv.org/abs/1805.00899
Keisler, H.J.: The hyperreal line. In: Ehrlich, P. (eds.) Real Numbers, Generalizations of the Reals, and Theories of Continua. Synthese Library, vol. 242, pp. 207–237. Springer, Dordrecht (1994). https://doi.org/10.1007/978-94-015-8248-3_8
Little, I.M.: Social choice and individual values. J. Polit. Econ. 60(5), 422–432 (1952)
Mercuur, R., Dignum, V., Jonker, C.M.: The value of values and norms in social simulation. J. Artif. Soc. Soc. Simul. 22(1), 9 (2019). https://doi.org/10.18564/jasss.3929
Morreau, M.: Grading in groups. Econ. Philos. 32(2), 323–352 (2016)
Noothigattu, R., et al.: Teaching AI agents ethical values using reinforcement learning and policy orchestration. IBM J. Res. Dev. 63(4/5), 2:1–2:9 (2019). https://doi.org/10.1147/JRD.2019.2940428
Peterson, M.: A royal road to consequentialism? Ethical Theory Moral Pract. 13(2), 153–169 (2010)
Popova, A., Regenwetter, M., Mattei, N.: A behavioral perspective on social choice. Ann. Math. Artif. Intell. 68(1), 5–30 (2013)
Raz, J.: Practical Reasoning. Oxford University Press, Oxford (1978)
Robinson, A.: Non-standard Analysis. Princeton University Press, Princeton (2016)
Schwartz, S.H.: An overview of the Schwartz theory of basic values. Online Read. Psychol. Cult. 2(1), 1–20 (2012)
Serramia, M., López-Sánchez, M., Moretti, S., Rodríguez-Aguilar, J.A.: On the dominant set selection problem and its application to value alignment. Auton. Agents Multi Agent Syst. 35(2), 42 (2021). https://doi.org/10.1007/s10458-021-09519-5
Serramia, M., et al.: Exploiting moral values to choose the right norms. In: Furman, J., Marchant, G.E., Price, H., Rossi, F. (eds.) Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, AIES 2018, New Orleans, LA, USA, 02–03 February 2018, pp. 264–270. ACM (2018). https://doi.org/10.1145/3278721.3278735
Szabo, J., Such, J.M., Criado, N.: Understanding the role of values and norms in practical reasoning. In: Bassiliades, N., Chalkiadakis, G., de Jonge, D. (eds.) EUMAS/AT -2020. LNCS (LNAI), vol. 12520, pp. 431–439. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-66412-1_27
Von Neumann, J., Morgenstern, O.: Theory of Games and Economic Behavior. Princeton University Press, Princeton (2007)
Zurek, T.: Goals, values, and reasoning. Expert Syst. Appl. 71, 442–456 (2017) https://doi.org/10.1016/j.eswa.2016.11.008
Zurek, T., Mokkas, M.: Value-based reasoning in autonomous agents. Int. J. Comput. Intell. Syst. 14(1), 896–921 (2021). https://doi.org/10.2991/ijcis.d.210203.001
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Szabo, J., Such, J.M., Criado, N., Modgil, S. (2022). Integrating Quantitative and Qualitative Reasoning for Value Alignment. In: Baumeister, D., Rothe, J. (eds) Multi-Agent Systems. EUMAS 2022. Lecture Notes in Computer Science(), vol 13442. Springer, Cham. https://doi.org/10.1007/978-3-031-20614-6_22
Download citation
DOI: https://doi.org/10.1007/978-3-031-20614-6_22
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-20613-9
Online ISBN: 978-3-031-20614-6
eBook Packages: Computer ScienceComputer Science (R0)