Category: Work in Progress

This project is housed at the Institute of Futures Research and seeks to address some challenges associated with the interpretability, explainability, and comprehensibility of neural networks—often termed ‘black boxes’ due to alleged epistemic opacity. Despite these limitations, I propose that neural network-generated knowledge can be epistemically licensed when they align with the theoretical requirements of scientific theories. Specifically, I focus on scientific theories that can be effectively represented through structures, or formal systems of symbols, statements, and inference rules. My goal is to establish a framework that positions neural networks as a plausible intermediary between terms and empirical statements in the formal apparatus of scientific theory, as satisfied traditionally by theoretical statements. This approach would bridge the gap between the computations and statistical results of neural networks and the epistemic objectives of science, and address concerns associated with the epistemic opacity of these models. By advancing a newly probabilistic account of scientific theories centered on neural networks, I hope to contribute new perspectives to the discourse on the role and interpretation of AI in scientific inquiry and the philosophy of science.

Objectives of this project include:

• Clearly defining and operationalizing in a philosophical context such notions as artificial intelligence, artificial neural network, and the structure of scientific theory
• Given an account of scientific theory, critically examining the available theoretical apparatus and understanding the role of ANNs in the production of science knowledge
• Exploring the senses of epistemic opacity implicated by ANNs, and identifying those most relevant to the project
• Understanding the scope of epistemic concerns surrounding the use of ANNs in the production of science knowledge
• Providing a framework for translating the function and properties of ANNs to the structure, both syntactic and semantic, of theoretical statements in scientific theory
• Demonstrating that ANNs satisfy the requirements of theoretical statements of scientific theory, epistemic and formal
• Providing additional informal motivation towards the epistemic license of ANNs

This approach is not without complications. In particular, the recourse to ANNs in the generation of science knowledge introduces a novel source of uncertainty. If artificial neural networks are the objects of epistemic license in a scientific theory, whatever uncertainty pervades the algorithm reflexively pervades the generated science knowledge, which we would take to be (empirical) hypotheses of the theory. Furthermore, we may decide that such theories, while being adequately prescriptive, are nevertheless inadequately descriptive and transparent. They may be inadequately explanatory, and introduce a novel uncertainty when attempting to reproduce their results.

Within the scope of this research project, I am conducting a review of the literature pertaining to syntactic and semantic accounts of scientific theory, epistemological challenges to neural network methods, the formal specification of artificial neural networks, and contributions of neural network-based research to science knowledge. A nonexhaustive bibliography follows.

I am greatly interested in potential feedback on this project, and suggestions for further reading.

References

Almada, M. (2019). Human Intervention in Automated Decision-Making: Toward the Construction of Contestable Systems. Proceedings of the Seventeenth International Conference on Artificial Intelligence and Law, 2–11. Presented at the Montreal, QC, Canada. doi:10.1145/3322640.3326699

Andoni, A., Panigrahy, R., Valiant, G., & Zhang, L. (22–24 Jun 2014). Learning Polynomials with Neural Networks. In E. P. Xing & T. Jebara (Eds.), Proceedings of the 31st International Conference on Machine Learning (pp. 1908–1916). Retrieved from https://proceedings.mlr.press/v32/andoni14.html

Ananny, M., & Crawford, K. (12 2016). Seeing without knowing: Limitations of the transparency ideal and its application to algorithmic accountability. New Media & Society, 0, 146144481667664. doi:10.1177/1461444816676645

Barredo Arrieta, A., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., … Herrera, F. (2020). Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion, 58, 82–115. doi:10.1016/j.inffus.2019.12.012

Barron, A. (06 1993). Barron, A.E.: Universal approximation bounds for superpositions of a sigmoidal function. IEEE Trans. on Information Theory 39, 930-945. Information Theory, IEEE Transactions On, 39, 930–945. doi:10.1109/18.256500

Bennett, M.T., Maruyama, Y. (2022). The Artificial Scientist: Logicist, Emergentist, and Universalist Approaches to Artificial General Intelligence. In: Goertzel, B., Iklé, M., Potapov, A. (eds) Artificial General Intelligence. AGI 2021. Lecture Notes in Computer Science(), vol 13154. Springer, Cham. https://doi.org/10.1007/978-3-030-93758-4_6

Binns, R., Van Kleek, M., Veale, M., Lyngs, U., Zhao, J., & Shadbolt, N. (2018). “It”s Reducing a Human Being to a Percentage’. Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems. doi:10.1145/3173574.3173951

Boge, F.J. Two Dimensions of Opacity and the Deep Learning Predicament. Minds & Machines 32, 43–75 (2022). https://doi.org/10.1007/s11023-021-09569-4

Breen, P. G., Foley, C. N., Boekholt, T., & Zwart, S. P. (2020). Newton versus the machine: solving the chaotic three-body problem using deep neural networks. Monthly Notices of the Royal Astronomical Society, 494(2), 2465–2470. doi:10.1093/mnras/staa713

Burrell, J. (2016). How the Machine \textquoteleftThinks\textquoteright: Understanding Opacity in Machine Learning Algorithms. Big Data and Society, 3(1). doi:10.1177/2053951715622512

Carabantes, M. (06 2020). Black-box artificial intelligence: an epistemological and critical analysis. AI & SOCIETY, 35. doi:10.1007/s00146-019-00888-w

Carnap, R., & Gardner, M. (1966). Philosophical Foundations of Physics: An Introduction to the Philosophy of Science. Retrieved from https://books.google.com/books?id=rP3YAAAAIAAJ

Carvalho, D., Pereira, E., & Cardoso, J. (07 2019). Machine Learning Interpretability: A Survey on Methods and Metrics. Electronics, 8, 832. doi:10.3390/electronics8080832

Cichy, R., & Kaiser, D. (02 2019). Deep Neural Networks as Scientific Models. Trends in Cognitive Sciences, 23. doi:10.1016/j.tics.2019.01.009

Doran, D., Schulz, S., & Besold, T. R. (2017). What Does Explainable AI Really Mean? A New Conceptualization of Perspectives. ArXiv, abs/1710.00794.

Dressel, J., & Farid, H. (01 2018). The accuracy, fairness, and limits of predicting recidivism. Science Advances, 4, eaao5580. doi:10.1126/sciadv.aao5580

Dorffner, G., Wiklicky, H., & Prem, E. (1994). Formal neural network specification and its implications on standardization. Computer Standards & Interfaces, 16(3), 205–219. doi:10.1016/0920-5489(94)90012-4

Durán, J., & Formanek, N. (12 2018). Grounds for Trust: Essential Epistemic Opacity and Computational Reliabilism. Minds and Machines. doi:10.1007/s11023-018-9481-6

Durán, J. M., & Jongsma, K. R. (2021). Who is afraid of black box algorithms? On the epistemological and ethical basis of trust in medical AI. Journal of Medical Ethics, 47(5), 329–335. doi:10.1136/medethics-2020-106820

Fiesler, E. (1994). Neural network classification and formalization. Computer Standards & Interfaces, 16(3), 231–239. doi:10.1016/0920-5489(94)90014-0

Gilpin, L., Bau, D., Yuan, B., Bajwa, A., Specter, M., & Kagal, L. (10 2018). Explaining Explanations: An Overview of Interpretability of Machine Learning. 80–89. doi:10.1109/DSAA.2018.00018

Girard-Satabin, J., Charpiat, G., Chihani, Z., & Schoenauer, M. (2019). CAMUS: A Framework to Build Formal Specifications for Deep Perception Systems Using Simulators. CoRR, abs/1911.10735. Retrieved from http://arxiv.org/abs/1911.10735

Hollanek, T. (11 2020). AI transparency: a matter of reconciling design with critique. AI & SOCIETY, 1–9. doi:10.1007/s00146-020-01110-y

Humphreys, P. (2009). The Philosophical Novelty of Computer Simulation Methods. Synthese, 169(3), 615–626. doi:10.1007/s11229-008-9435-2

Ismailov, V. (2020). A three layer neural network can represent any discontinuous multivariate function. arXiv [Cs.LG]. Retrieved from http://arxiv.org/abs/2012.03016

Jumper, J., Evans, R., Pritzel, A. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). https://doi.org/10.1038/s41586-021-03819-2

Karniadakis, G.E., Kevrekidis, I.G., Lu, L. et al. Physics-informed machine learning. Nat Rev Phys 3, 422–440 (2021). https://doi.org/10.1038/s42254-021-00314-5

Kim, J. (1988). What Is ‘Naturalized Epistemology?’ Philosophical Perspectives, 2, 381–405. Retrieved from http://www.jstor.org/stable/2214082

Knuuttila, T. (2011). Modelling and representing: An artefactual approach to model-based representation. Studies in History and Philosophy of Science Part A, 42(2), 262–271. doi:10.1016/j.shpsa.2010.11.034

Marquis, P., Papini, O., & Prade, H. (2020). A Guided Tour of Artificial Intelligence Research: Volume III: Interfaces and Applications of Artificial Intelligence. Retrieved from https://books.google.com/books?id=z07iDwAAQBAJ

Mahbooba, B., Timilsina, M., Sahal, R., & Serrano, M. (01 2021). Explainable Artificial Intelligence (XAI) to Enhance Trust Management in Intrusion Detection Systems Using Decision Tree Model. Complexity, 2021, 11. doi:10.1155/2021/6634811

Martínez-Ordaz, M. del R. (2023). Scientific understanding through big data: From ignorance to insights to understanding. Possibility Studies & Society, 0(0). https://doi.org/10.1177/27538699231176523

Montavon, G., Samek, W., & Müller, K.-R. (2018). Methods for interpreting and understanding deep neural networks. Digital Signal Processing, 73, 1–15. doi:10.1016/j.dsp.2017.10.011

Pappas, G. S. (2012). Justification and Knowledge: New Studies in Epistemology. Retrieved from https://books.google.com/books?id=6FlgBgAAQBAJ

Ramsey, W. (1997). Do Connectionist Representations Earn Their Explanatory Keep? Mind and Language, 12(1), 34–66. doi:10.1111/j.1468-0017.1997.tb00061.x

Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). ‘Why Should I Trust You?’: Explaining the Predictions of Any Classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1135–1144. Presented at the San Francisco, California, USA. doi:10.1145/2939672.2939778

Rudin, C. (05 2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1, 206–215. doi:10.1038/s42256-019-0048-x

Samek, W., Wiegand, T., & Müller, K.-R. (10 2017). Explainable Artificial Intelligence: Understanding, Visualizing and Interpreting Deep Learning Models. ITU Journal: ICT Discoveries – Special Issue 1 – The Impact of Artificial Intelligence (AI) on Communication Networks and Services, 1, 1–10.

Sanchez-Gonzalez, A., Godwin, J., Pfaff, T., Ying, R., Leskovec, J., & Battaglia, P. W. (2020). Learning to Simulate Complex Physics with Graph Networks. CoRR, abs/2002.09405. Retrieved from https://arxiv.org/abs/2002.09405

Seshia, S. A., Desai, A., Dreossi, T., Fremont, D. J., Ghosh, S., Kim, E., … Yue, X. (2018). Formal Specification for Deep Neural Networks. In S. K. Lahiri & C. Wang (Eds.), Automated Technology for Verification and Analysis (pp. 20–34). Cham: Springer International Publishing.

Srećković, S., Berber, A. & Filipović, N. The Automated Laplacean Demon: How ML Challenges Our Views on Prediction and Explanation. Minds & Machines 32, 159–183 (2022). https://doi.org/10.1007/s11023-021-09575-6

Szymanski, L., & McCane, B. (2012). Deep, super-narrow neural network is a universal classifier. The 2012 International Joint Conference on Neural Networks (IJCNN), 1–8. doi:10.1109/IJCNN.2012.6252513

Taylor, B. J., & Darrah, M. A. (2005). Rule extraction as a formal method for the verification and validation of neural networks. Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., 5, 2915–2920 vol. 5. doi:10.1109/IJCNN.2005.1556388

Walmsley, J. (06 2021). Artificial intelligence and the value of transparency. AI & SOCIETY, 36, 1–11. doi:10.1007/s00146-020-01066-z

Wheeler, G. R., & Pereira, L. M. (2004). Epistemology and artificial intelligence. Journal of Applied Logic, 2(4), 469–493. doi:10.1016/j.jal.2004.07.007

Xu, T., Zhan, J., Garrod, O. G. B., Torr, P. H. S., Zhu, S.-C., Ince, R. A. A., & Schyns, P. G. (2018). Deeper Interpretability of Deep Networks. arXiv [Cs.CV]. Retrieved from http://arxiv.org/abs/1811.07807

Zednik, C. (06 2021). Solving the Black Box Problem: A Normative Framework for Explainable Artificial Intelligence. Philosophy & Technology, 34. doi:10.1007/s13347-019-00382-7

Zerilli, J., Knott, A., Maclaurin, J., & Gavaghan, C. (12 2019). Transparency in Algorithmic and Human Decision-Making: Is There a Double Standard? Philosophy & Technology, 32. doi:10.1007/s13347-018-0330-6

Highly accurate protein structure prediction with AlphaFold. (08 2021). Nature, 596, 583–589. doi:10.1038/s41586-021-03819-2