Abstract
The paper proposes a synthesis between human scientists and artificial representation learning models as a way of augmenting epistemic warrants of realist theories against various anti-realist attempts. Towards this end, the paper fleshes out unconceived alternatives not as a critique of scientific realism but rather a reinforcement, as it rejects the retrospective interpretations of scientific progress, which brought about the problem of alternatives in the first place. By utilising adversarial machine learning, the synthesis explores possibility spaces of available evidence for unconceived alternatives providing modal knowledge of what is possible therein. As a result, the epistemic warrant of synthesised realist theories should emerge bolstered as the underdetermination by available evidence gets reduced. While shifting the realist commitment away from theoretical artefacts towards modalities of the possibility spaces, the synthesis comes out as a kind of perspectival modelling.
Similar content being viewed by others
Notes
It’s also perhaps a bit unfortunate that contemporary reviews of automated discovery adhere to somewhat dated typologies of artificial intelligence applications within science (f.e. Giza 2017). Reflecting for the most part earlier results entirely omits recent successes delivered by Deep Learning, while underappreciating the influence of artificial representation learning on science in general (cf. ibid.).
As Goodfellow notes, since the players are neural networks, and their parameters acquired by back-propagation of error, heuristically, to secure a non-vanishing gradient it is better to consider the generator as maximising the probability of the discriminator being mistaken (Goodfellow et al. 2014). This slightly changes the nature of the game, since it can no longer be described in terms of a single value function (ibid.). Although this represents a shift from describing the scenario in terms of a minimax game, it doesn’t lessen the game’s relevance towards the theoretical analysis of adversarial artificial representation learning.
The model can get also stranded in an overfitted state, arising from what is usually described as memorisation of the training data (observational evidence), likewise hampering its capability to generalise beyond the evidence. However, as it is underfitting which mostly imperils the current generative adversarial models, and the subject at hand comprises mainly underdetermination, a discussion of overfitting would diverge from the goal of the paper.
We are indebted to John Symons for pointing out this limit of the human–machine learning synthetisation of (realist) scientific theories.
References
Alai M (2004) A.I., scientific discovery and realism. Minds Mach 14:21–42
Alai M (2014) Novel predictions and the no miracle argument. Erkenntnis 79:297–326
Albert A, Strano E, Kaur J, Gonzáles M (2018) Modeling urbanization patterns with generative adversarial networks. arXiv:1801.02710v1 [cs.LG]
Arora S, Ge R, Liang Y, Ma T, Zhang Y (2017) Generalization and equilibrium in generative adversarial nets (GANs). In: Proceedings of international conference on machine learning (ICML). August 6–11 Sydney, Australia, pp 224–232
Arora S, Zhang Y, Risteski A (2018) Do GANs learn the distribution? Some theory and empirics. In: Proceedings of 6th international conference on learning representations (ICLR). April 30–May 03 Vancouver, Canada
Carleo G, Cirac I, Cranmer K, Daudet L, Schuld M, Tishby N, Vogt-Maranto L, Zdeborová L (2019) Machine learning and the physical sciences. Rev Mod Phys 91(4):045002
Chakravartty A (2008) What you don’t know can’t hurt you: realism and the unconceived. Philos Stud 137:149–158
Chakravartty A (2017a) Scientific realism. In: Zalta EN (ed) The stanford encyclopedia of philosophy (Summer 2017 Edition). https://plato.stanford.edu/archives/sum2017/entries/scientific-realism/
Chakravartty A (2017b) Reflections on new thinking about scientific realism. Synth SI New Think About Sci Realism 197:3379–3392
Dawid R (2013) String theory and the scientific method. Cambridge University Press, Cambridge
de Oliveira L, Paganini M, Nachman B (2017) Learning particle physics by example: location-aware generative adversarial networks for physics synthesis. Comput Softw Big Sci 1:4
Di Sipio R, Giannelli MF, Haghighat SK, Palazzo S (2019) DijetGAN: a generative-adversarial network approach for the simulation of QCD dijet events at the LHC. J High Energy Phys 2019:110
Džeroski S, Langley P, Todorovski L (2007) Computational discovery of scientific knowledge. In: Džeroski S, Todorovski L (eds) Computational discovery of scientific knowledge: introduction, techniques, and applications in environmental and life sciences. Springer, Berlin, pp 1–14
Egg M (2014) Expanding our grasp: causal knowledge and the problem of unconceived alternatives. Br J Philos Sci 67(1):115–141
Erbin H, Krippendorf S (2018) GANs for generating EFT models. arXiv:1809.02612v1 [cs.LG]
Fedus W, Rosca M, Lakshminarayanan B, Dai AM, Mohamed S, Goodfellow I (2018) Many paths to equilibrium: GANs do not need to decrease a divergence at every step. In: Proceedings of 6th international conference on learning representations (ICLR). April 30–May 03 Vancouver, Canada
Gillies D (1996) Artificial intelligence and scientific method. Oxford University Press, Oxford
Giza P (2017) Automated discovery systems and the inductivist controversy. J Exp Theor Artif Intell 29(5):1053–1069
Goodfellow I, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, Ozair S, Courville A, Bengio Y (2014) Generative adversarial networks. In: Proceedings of 27th advances in neural information processing systems (NIPS). December 8–13 Montreal, Canada
Goodfellow I, Bengio Y, Courville A (2016) Deep learning. MIT Press, Cambridge
Hashemi A, Amin N, Datta K, Olivito D, Pierini M (2019) LHC analysis-specific datasets with generative adversarial networks. arXiv:1901.05282v1 [hep-ex]
Holland JH, Holyoak KJ, Nisbett RE, Thagard P (1986) Induction. Processes of inference, learning, and discovery. MIT Press, Cambridge
Hornik K, Stinchcombe M, White H (1989) Multilayer feedforward networks are universal approximators. Neural Netw 2(5):359–366
Hume D (1739/1978) A treatise on human nature. Book I: on human understanding. Oxford University Press, Oxford
Humphreys P (2004) Extending ourselves: computational science, empiricism, and scientific method. Oxford University Press, New York
Humphreys P (2011) Computational science and its effects. In: Carrier M, Nordmann A (eds) Science in the context of application. Springer, Dordrecht, pp 131–142
Humphreys P (2020) Why automated science should be cautiously welcomed. In: Bertolaso M, Sterpetti F (eds) A critical reflection on automated science. Will science remain human?. Springer, Cham, pp 11–26
Kim B, Lee S, Kim J (2020) Inverse design of porous materials using artificial neural networks. Sci Adv 6(1):eaax9324
Kingma DP, Welling M (2014) Auto-encoding variational bayes. In: Proceedings of 2nd international conference on learning representations (ICLR). April 14–16 Banff, Canada
Kukla A (1996) Does every theory have empirically equivalent rivals? Erkenntnis 44(2):137–166
Langley P (2000) The computational support for scientific discovery. Int J Hum–Comput Interact 53:393–410
Laudan L (1990) Demystifying underdetermination. In: Wade Savage C (ed) Scientific theories. University of Minnesota Press, Minneapolis, pp 267–297
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521:436–444
Magnus PD (2010) Inductions, red herrings, and the best explanation for the mixed record of science. Br J Philos Sci 61(4):803–819
Massimi M (2018) Perspectival modeling. Philos Sci 85(3):335–359
Méndez-Lucio O, Baillif B, Clevert DA, Rouquié D, Wichard J (2020) De novo generation of hit-like molecules from gene expression signatures using artificial intelligence. Nat Commun 11:1–10
Mizrahi M (2016) Historical inductions: new cherries, same old cherry-picking. Int Stud Philos Sci 29(2):129–148
Mizrahi M (2017) The history of science as a graveyard of theories: a philosophers’ myth? Int Stud Philos Sci 30(3):263–278
Mustafa M, Bard D, Bhimji W, Al-Rfou R, Lukić Z (2017) Creating virtual universes using generative adversarial networks. arXiv:1706.02390v1 [astro-ph.IM]
Paganini M, de Oliveira L, Nachman B (2018) Accelerating science with generative adversarial networks: an application to 3D particle showers in multilayer calorimeters. Phys Rev Lett 120:042003-1–042003-6
Psillos S (1999) Scientific realism: how science tracks truth. Routledge, London
Putnam H (1975) Mathematics, matter and method, philosophical papers, vol I. Cambridge University Press, Cambridge
Quine WVO (1951) Two dogmas of empiricism. Philos Rev 60(1):20–43
Quine WVO (1970) On the reasons for indeterminacy of translation. J Philos 67(6):178–183
Quine WVO (1975) On empirically equivalent systems of the world. Erkenntnis 9:313–328
Radovic A, Williams M, Rousseau D, Kagan M, Bonacorsi D, Himmel A, Aurisano A, Terao K, Wongjirad T (2018) Machine learning at the energy and intensity frontiers of particle physics. Nature 560:41–48
Rescher N (2006) Epistemetrics. Cambridge University Press, Cambridge
Rodríguez AC, Kacprzak T, Lucchi A, Amara A, Sgier R, Fluri J, Hofmann T, Réfrégier A (2018) Fast cosmic web simulations with generative adversarial networks. Comput Astrophys Cosmol 5:4
Saatsi J (2019) Historical inductions, old and new. Synthese 196(10):3979–3993
Stanford PK (2006) Exceeding our grasp: science, history, and the problem of unconceived alternatives. Oxford University Press, New York
Stanford PK (2017) Underdetermination of scientific theory. In: Zalta EN (ed) The stanford encyclopedia of philosophy (Winter 2017 edition). https://plato.stanford.edu/archives/win2017/entries/scientific-underdetermination/
Stolle R, Bradley E (2007) Communicable knowledge in automated system identification. In: Džeroski S, Todorovski L (eds) Computational discovery of scientific knowledge: introduction, techniques, and applications in environmental and life sciences. Springer, Berlin, pp 17–43
Symons J, Horner J (2014) Software intensive science. Philos Technol 27(3):461–477
Symons J, Horner J (2017) Software error as a limit to inquiry for finite agents: challenges for the post-human scientist. In: Powers TM (ed) Philosophy and computing. Springer, Cham
Symons J, Horner J (2019) Why there is no general solution to the problem of software verification. Found Sci. https://doi.org/10.1007/s10699-019-09611-w
Thagard P (1988) Computational philosophy of science. MIT Press, Cambridge
van Fraassen BC (1980) The scientific image. Clarendon Press, Oxford
Wolpert D (1996) The lack of a priori distinctions between learning algorithms. Neural Comput 8:1341–1390
Wray KB (2018) Resisting scientific realism. Cambridge University Press, Cambridge
Zamudio-Fernandez J, Okan A, Villaescusa-Navarro F, Bilaloglu S, Cengiz AD, He S, Levasseur LP, Ho S (2019) HIGAN: cosmic neutral hydrogen with generative adversarial networks. In: Machine learning and the physical sciences workshop at the 33rd conference on neural information processing systems (NeurIPS)
Zhang C, Bengio S, Hardt M, Recht B, Vinyals O (2017) Understanding deep learning requires rethinking generalization. In: Proceedings of 5th international conference on learning representations (ICLR). April 24–26 Toulon, France
Acknowledgements
We are grateful to our reviewers for the helpful feedback as well as to the editors at Axiomathes. Especially, we would like to thank John Symons for his comments and support. This research was supported by Charles University Research Centre of Excellence UNCE/HUM/037 ‘The Human–Machine Nexus and International Order’.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Spelda, P., Stritecky, V. What Can Artificial Intelligence Do for Scientific Realism?. Axiomathes 31, 85–104 (2021). https://doi.org/10.1007/s10516-020-09480-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10516-020-09480-0