Tag: Philosophy of Science

Posted on

(Work in Progress) Scientific Theory and the Epistemology of Neural Networks

Scientific Theory and the Epistemology of Neural Networks

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

Posted on

(Featured) Limits of conceivability in the study of the future. Lessons from philosophy of science

Limits of conceivability in the study of the future. Lessons from philosophy of science

Veli Virmajoki explores the epistemological and conceptual limitations of futures studies, and offers an enlightening perspective in the philosophical discourse on the conceivability of future possibilities. Utilizing three case studies from the philosophy of science as the crux of its argument, the paper meticulously dissects how these limitations pose significant obstacles in envisaging alternatives to the present state of affairs. The author poses a thought-provoking argument centered on the constraints imposed by our current understanding of reality and the mechanisms it employs to reinforce its own continuity and inevitability.

The backbone of this philosophical inquiry lies in the robust debate between inevitabilism, a stance asserting the inevitable development of specific scientific theories, and contingentism, a view that endorses the potentiality of genuinely alternative scientific trajectories. The exploration of this contentious issue facilitates a deeper understanding of the constraints in predicting future scenarios, as our ability to conceptualize these alternatives is bound by our understanding of past and present realities. The paper deftly argues that the choice between inevitabilism and contingentism is fundamentally intertwined with our personal intuition about the range of genuine possibilities, thereby asserting the subjective nature of perceived futurity. As such, the article offers a fresh, critical lens to scrutinize the underpinnings of futures studies, and instigates a profound rethinking of our philosophical approach to anticipating what lies ahead.

Unconceived Possibilities and their Consequences

The author asserts that our conception of potential futures is significantly limited by profound epistemological and conceptual factors. They draw on the case study of the late 19th-century ether theories in physics, where, despite the existence of genuinely alternative theories, only a limited number of possibilities were conceived due to prevailing scientific practices and principles. The author uses this historical case to illustrate that while some futures may seem inconceivable from our present vantage point, they may still fall within the realm of genuine possibilities.

Moreover, the author argues that the potential impact of these unconceived possibilities extends beyond the localized elements of a system to reverberate throughout its entirety. This underlines the complexity of the task in futures studies; any unconceived alternatives in one sector of a system can trigger significant, far-reaching consequences for the entire system. Therefore, the research warns against oversimplification in predicting future scenarios and emphasizes the need for a nuanced approach that recognizes the interconnectedness of elements within any given system. This presents a remarkable challenge for futures studies, highlighting the depth of the iceberg that lies beneath the surface of our current epistemological and conceptual understanding.

Historical Trajectories and Justification of Future Possibilities

In the examination of plausibility and the justification of future possibilities, the article underscores the fundamental epistemological and conceptual challenges that limit our capability to predict alternative futures. The author refers to historical episodes like the case of Soviet cybernetics, where the existence of plausible alternative futures was not recognized, due to the collective failure to see past the status quo. It brings to light the inherent difficulties in justifying the plausibility or even the possibility of certain futures, where our current knowledge systems and conceptual frameworks may blind us to divergent scenarios. This observation raises pertinent questions about the inherent biases of our epistemic practices, as well as the potential for deeply entrenched beliefs to restrict our ability to imagine and evaluate a broader range of future possibilities. Hence, this line of inquiry necessitates the careful examination of the underlying assumptions that might constrain the scope of our foresight and deliberations on future possibilities.

The article further discusses the concept of historical trajectories and their connection to future possibilities, offering a philosophical lens into the entanglement of past, present, and future. It argues that our understanding of history and future possibilities, and our interpretation of the present’s robustness and inevitability, are inextricably linked through a complex web of modal considerations. The author emphasizes the interconnectedness of past trajectories and future possibilities, arguing that the way we perceive historical possibilities affects how we anticipate future outcomes. This perspective allows us to examine whether it is the deterministic view of history (inevitabilism) or the contingency of events (contingentism) that should be the default position, a determination that would have profound implications for our understanding of future possibilities.

Inevitabilism vs. Contingentism

Tthe author elaborates on a crucial dichotomy in philosophy of science: inevitabilism versus contingentism. Inevitabilism implies a deterministic understanding of scientific and historical development, where the present state of affairs appears as the unique and necessary outcome of the past. Contingentism, on the other hand, endorses the idea of multiple genuine alternatives to the current state, thus opening the space of historical and future possibilities. The article underscores that these positions are not simply academic disputes but carry substantial implications for how we conceive possibilities for the future. Moreover, these divergent outlooks reflect the individual’s inherent beliefs and intuitions about the range of possibilities within human affairs. The author contends that these perspectives cannot conclusively advocate for or against alternative futures because one’s stance on the inevitabilism versus contingentism debate inherently relies on their preconceived notions of the scope of historical and future possibilities.

Future Research Avenues

In light of the research as presented, promising avenues for future research emerge. The author suggests a systematic examination of the epistemological and conceptual boundaries of our ability to conceive and reason about potential futures. Such an investigation is not limited to philosophical discourse but requires interdisciplinary dialogue with a myriad of fields, as these boundaries are, in part, shaped by our social and scientific structures. This method of research would offer a comprehensive understanding of the creative and critical capacities of futures studies and aid us in recognizing our epistemological and conceptual predicament concerning future possibilities. Furthermore, it could potentially expose the manner in which these boundaries are historically mutable, opening up a discussion about the renegotiation of the boundaries of conceivability.

Abstract

In this paper, the epistemological and conceptual limits of our ability to conceive and reason about future possibilities are analyzed. It is argued that more attention should be paid in futures studies on these epistemological and conceptual limits. Drawing on three cases from philosophy of science, the paper argues that there are deep epistemological and conceptual limits in our ability to conceive and reason about alternatives to the current world. The nature and existence of these limits are far from obvious and become visible only through careful investigation. The cases establish that we often are unable to conceive relevant alternatives; that historical and counterfactual considerations are more limited than has been suggested; and that the present state of affairs reinforces its hegemony through multiple conceptual and epistemological mechanisms. The paper discusses the reasons behind the limits of the conceivability and the consequences that follow from the considerations that make the limits visible. The paper suggests that the epistemological and conceptual limits in our ability to conceive and reason about possible futures should be mapped systematically. The mapping would provide a better understanding of the creative and critical bite of futures studies.

Limits of conceivability in the study of the future. Lessons from philosophy of science
Posted on

(Featured) Epistemic diversity and industrial selection bias

Epistemic diversity and industrial selection bias

Manuela Fernández Pinto and Daniel Fernández Pinto offer a compelling examination of the role that funding sources play in shaping scientific consensus, focusing specifically on the influence of private industry. Drawing on the work of Holman and Bruner (2017), the authors use a reinforcement learning model, known as a Q-learning model, to explore industrial selection. The central concept of industrial selection posits that, rather than corrupting individual scientists, private industry can subtly steer scientific outcomes towards their interests by selectively funding research. In the authors’ simulation, three different funding scenarios are considered: research funded solely by industry, research funded solely by a random agent, and research jointly funded by industry and a random agent.

Results from the simulations reinforce the effects of industrial selection observed by Holman and Bruner, showing a divergence from correct scientific hypotheses under sole industry funding. When scientists are funded solely by a random agent, the outcomes are closer to the correct hypothesis. Most notably, when funding is a mix of industry and random allocation, the random element appears to counteract, or at least delay, the bias introduced by industry funding. The authors further observe an unexpected and somewhat paradoxical interaction with methodological diversity, a factor traditionally seen as a strength in scientific communities. Industrial funding effectively exploits this diversity to skew consensus towards industry-friendly outcomes.

The authors then introduce a provocative and potentially contentious suggestion based on their simulations. They propose that a random allocation of funding might be a more effective countermeasure against industrial selection bias than the commonly held belief in meritocratic funding systems, which might inadvertently perpetuate industry bias. This suggestion arises from their observation that a random funding agent in the simulation effectively obstructs industrial selection bias. They also consider the merits and drawbacks of a two-stage random allocation system, wherein only research proposals that pass an initial quality assessment are subject to a subsequent funding lottery.

The study raises compelling philosophical considerations on the influence of funding on the direction and integrity of scientific research. It challenges the common narrative that methodological diversity and meritocracy inherently lead to unbiased, high-quality science, suggesting instead that these can be co-opted by industry to push scientific consensus toward commercially advantageous outcomes. It further incites reflection on the ethical implications of allowing commercial interests to potentially manipulate scientific consensus and the responsibility of society to ensure the pursuit of truth in science. The research also ties into broader discussions on the balance between rational decision-making and randomness, and the potential role of randomness as a mitigating factor in decision-making processes rife with bias or undue influence.

Future research could delve deeper into how a random allocation system might be implemented in practice, particularly regarding the initial quality assessment process. It would also be beneficial to explore how such a system could coexist with traditional funding sources, and what percentage of overall funding would need to be randomly allocated to effectively mitigate industrial selection bias. Additionally, more nuanced simulations could help further untangle the complex relationship between methodological diversity and industrial bias, and identify other possible factors that may be manipulated to sway scientific consensus. Ultimately, this research presents a provocative stepping stone for further exploration into the complex and subtle ways commercial interests may influence scientific research, and potential innovative strategies to counteract such influences.

Abstract

Philosophers of science have argued that epistemic diversity is an asset for the production of scientific knowledge, guarding against the effects of biases, among other advantages. The growing privatization of scientific research, on the contrary, has raised important concerns for philosophers of science, especially with respect to the growing sources of biases in research that it seems to promote. Recently, Holman and Bruner (2017) have shown, using a modified version of Zollman (2010) social network model, that an industrial selection bias can emerge in a scientific community, without corrupting any individual scientist, if the community is epistemically diverse. In this paper, we examine the strength of industrial selection using a reinforcement learning model, which simulates the process of industrial decision-making when allocating funding to scientific projects. Contrary to Holman and Bruner’s model, in which the probability of success of the agents when performing an action is given a priori, in our model the industry learns about the success rate of individual scientists and updates the probability of success on each round. The results of our simulations show that even without previous knowledge of the probability of success of an individual scientist, the industry is still able to disrupt scientific consensus. In fact, the more epistemically diverse the scientific community, the easier it is for the industry to move scientific consensus to the opposite conclusion. Interestingly, our model also shows that having a random funding agent seems to effectively counteract industrial selection bias. Accordingly, we consider the random allocation of funding for research projects as a strategy to counteract industrial selection bias, avoiding commercial exploitation of epistemically diverse communities.

Epistemic diversity and industrial selection bias