Throwing light on black boxes: emergence of visual categories from deep learning

Abstract

One of the best known arguments against the connectionist approach to artificial intelligence and cognitive science is that neural networks are black boxes, i.e., there is no understandable account of their operation. This difficulty has impeded efforts to explain how categories arise from raw sensory data. Moreover, it has complicated investigation about the role of symbols and language in cognition. This state of things has been radically changed by recent experimental findings in artificial deep learning research. Two kinds of artificial deep learning networks, namely the convolutional neural network and the generative adversarial network have been found to possess the capability to build internal states that are interpreted by humans as complex visual categories, without any specific hints or any grammatical processing. This emergent ability suggests that those categories do not depend on human knowledge or the syntactic structure of language, while they do rely on their visual context. This supports a mild form of empiricism, while it does not assume that computational functionalism is true. Some consequences are extracted regarding the debate about amodal and grounded representations in the human brain. Furthermore, new avenues for research on cognitive science are open.

This is a preview of subscription content, log in to check access.

Fig. 1
Fig. 2

Notes

  1. 1.

    An online demonstration is available at http://ganpaint.io/.

  2. 2.

    GANs have been called “the coolest idea in deep learning in the last 20 years” (Metz 2017) by Yann LeCun, recipient of the Turing award, the computer science equivalent to the Nobel prize.

  3. 3.

    There are artificial deep neural networks that do manage sequential linguistic information, such as deep recurrent neural networks (DRNNs), including long short-term memory units (LSTMs), but they are not considered in this work.

References

  1. Barsalou, L. (2016). On staying grounded and avoiding quixotic dead ends. Psychonomic Bulletin and Review, 23(4), 1122–1142.

    Article  Google Scholar 

  2. Bau, D., Strobelt, H., Peebles, W., Wulff, J., Zhou, B., Zhu, J., et al. (2019). Semantic photo manipulation with a generative image prior. ACM Transactions on Graphics, 38(4), 59.

    Article  Google Scholar 

  3. Bau, D., Zhou, B., Khosla, A., Oliva, A., & Torralba, A. (2017). Network dissection: Quantifying interpretability of deep visual representations. CoRR arxiv:1704.05796.

  4. Bau, D., Zhu, J. Y., Strobelt, H., Zhou, B., Tenenbaum, J. B., Freeman, W. T., & Torralba, A. (2018). GAN dissection: Visualizing and understanding generative adversarial networks. CoRR arxiv:1811.10597.

  5. Bechtel, W. (1993). The case for connectionism. Philosophical Studies, 71(2), 119–154.

    Article  Google Scholar 

  6. Benitez, J. M., Castro, J. L., & Requena, I. (1997). Are artificial neural networks black boxes? IEEE Transactions on Neural Networks, 8(5), 1156–1164.

    Article  Google Scholar 

  7. Berkeley, I. S. N. (2019). The curious case of connectionism. Open Philosophy, 2(1), 190–205.

    Article  Google Scholar 

  8. Brainerd, C. J., & Reyna, V. F. (2002). Fuzzy-trace theory and false memory. Current Directions in Psychological Science, 11(5), 164–169.

    Article  Google Scholar 

  9. Buckner, C. (2015). A property cluster theory of cognition. Philosophical Psychology, 28(3), 307–336.

    Article  Google Scholar 

  10. Buckner, C. (2018). Empiricism without magic: Transformational abstraction in deep convolutional neural networks. Synthese, 195(12), 5339–5372.

    Article  Google Scholar 

  11. Buckner, C. (2019). Deep learning: A philosophical introduction. Philosophy Compass,. https://doi.org/10.1111/phc3.12625.

    Article  Google Scholar 

  12. Buckner, C., & Garson, J. (2019). Connectionism and post-connectionist models. In M. Sprevak & M. Colombo (Eds.), The Routledge handbook of the computational mind (pp. 76–90). New York: Routledge.

    Google Scholar 

  13. Butz, M. V., & Kutter, E. F. (2017). How the mind comes into being. Oxford: Oxford University Press.

    Google Scholar 

  14. Cantwell Smith, B. (2019). The promise of artificial intelligence. Cambridge: The MIT Press.

    Google Scholar 

  15. Carabantes, M. (2019). Black-box artificial intelligence: An epistemological and critical analysis. AI & Society. https://doi.org/10.1007/s00146-019-00888-w.

  16. Chin-Parker, S., & Ross, B. H. (2002). The effect of category learning on sensitivity to within-category correlations. Memory & Cognition, 30(3), 353–362.

    Article  Google Scholar 

  17. Chollet, F. (2018). Deep learning with Python. Shelter Island, NY: Manning.

    Google Scholar 

  18. Clark, A. (1993). Associative engines: Connectionism, concepts, and representational change. Cambridge, MA: The MIT Press.

    Google Scholar 

  19. Creel, K. A. (2020). Transparency in complex computational systems. Philosophy of Science. https://doi.org/10.1086/709729.

  20. Dawson, M. R. W. (2004). Minds and machines: Connectionism and psychological modeling. Malden, MA: Blackwell Publishing.

    Google Scholar 

  21. Dove, G. (2009). Beyond perceptual symbols: A call for representational pluralism. Cognition, 110(3), 412–431.

    Article  Google Scholar 

  22. Dove, G. (2016). Three symbol ungrounding problems: Abstract concepts and the future of embodied cognition. Psychonomic Bulletin and Review, 23(4), 1109–1121.

    Article  Google Scholar 

  23. Elber-Dorozko, L., & Shagrir, O. (2019). Computation and levels in the cognitive and neural sciences. In M. Sprevak & M. Colombo (Eds.), The Routledge handbook of the computational mind (pp. 205–222). New York: Routledge.

    Google Scholar 

  24. Foster, D. (2019). Generative deep learning. Sebastopol, CA: O’Reilly.

    Google Scholar 

  25. Gärdenfors, P. (2000). Conceptual spaces: The geometry of thought. Cambridge, MA: The MIT Press.

    Google Scholar 

  26. Gärdenfors, P. (2014). The geometry of meaning: Semantics based on conceptual spaces. Cambridge, MA: The MIT Press.

    Google Scholar 

  27. Gomes, L. (2014). Machine-learning maestro Michael Jordan on the delusions of Big Data and other huge engineering efforts. In IEEE spectrum, October 20, 2014.

  28. Gonzalez-Garcia, A., Modolo, D., & Ferrari, V. (2018). Do semantic parts emerge in convolutional neural networks? International Journal of Computer Vision, 126(5), 476–494.

    Article  Google Scholar 

  29. Gulli, A., & Pal, S. (2017). Deep learning with keras. Birmingham: Packt Publishing Ltd.

    Google Scholar 

  30. He, K., Zhang, X., Ren, S., & Sun, J. (2015). Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification. In Proceedings of the 2015 IEEE international conference on computer vision (ICCV) (pp. 1026–1034). IEEE Computer Society.

  31. Horgan, T., & Tienson, J. (1996). Connectionism and the philosophy of psychology. Cambridge, MA: The MIT Press.

    Google Scholar 

  32. Kaipainen, M., & Hautamäki, A. (2015). A perspectivist approach to conceptual spaces. In F. Zenker & P. Gärdenfors (Eds.), Applications of conceptual spaces: The case for geometric knowledge representation (pp. 245–258). Cham: Springer.

    Google Scholar 

  33. Ketkar, N. (2017). Deep learning with Python. New York: Apress.

    Google Scholar 

  34. Kiefer, M., & Pulvermüller, F. (2012). Conceptual representations in mind and brain: Theoretical developments, current evidence and future directions. Cortex, 48(7), 805–825.

    Article  Google Scholar 

  35. Langr, J., & Bok, V. (2019). GANs in action. Shelter Island, NY: Manning.

    Google Scholar 

  36. Leshinskaya, A., & Caramazza, A. (2016). For a cognitive neuroscience of concepts: Moving beyond the grounding issue. Psychonomic Bulletin & Review, 23(4), 991–1001.

    Article  Google Scholar 

  37. Machery, E. (2005). Concepts are not a natural kind. Philosophy of Science, 72(3), 444–467.

    Article  Google Scholar 

  38. Machery, E. (2007). Concept empiricism: A methodological critique. Cognition, 104(1), 19–46.

    Article  Google Scholar 

  39. Machery, E. (2016). The amodal brain and the offloading hypothesis. Psychonomic Bulletin and Review, 23(4), 1090–1095.

    Article  Google Scholar 

  40. Mahon, B., & Caramazza, A. (2009). Concepts and categories: A cognitive neuropsychological perspective. Annual Review of Psychology, 60, 27–51.

    Article  Google Scholar 

  41. Marcus, G. (2018). Innateness, AlphaZero, and artificial intelligence. CoRR arxiv:1801.05667.

  42. Mayor, J., Gomez, P., Chang, F., & Lupyan, G. (2014). Connectionism coming of age: Legacy and future challenges. Frontiers in Psychology, 5, 187.

    Article  Google Scholar 

  43. Metz, C. (2017). Google’s dueling neural networks spar to get smarter, no humans required. https://www.wired.com/2017/04/googles-dueling-neural-networks-spar-get-smarter-no-humans-required/.

  44. Neudeck, P., & Wittchen, H. U. (2012). Introduction: Rethinking the model-refining the method. In P. Neudeck & H. U. Wittchen (Eds.), Exposure therapy (pp. 1–8). New York, NY: Springer.

    Google Scholar 

  45. Newell, A., & Simon, H. A. (1976). Computer science as empirical inquiry: Symbols and search. Communications of the ACM, 19(3), 113–126.

    Article  Google Scholar 

  46. Pearl, J., & Mackenzie, D. (2018). The book of why: The new science of cause and effect. New York: Basic Books.

    Google Scholar 

  47. Piccinini, G., & Bahar, S. (2013). Neural computation and the computational theory of cognition. Cognitive Science, 37(3), 453–488.

    Article  Google Scholar 

  48. Piccinini, G., & Scarantino, A. (2011). Information processing, computation, and cognition. Journal of Biological Physics, 37(1), 1–38.

    Article  Google Scholar 

  49. Qin, Z., Yu, F., Liu, C., & Chen, X. (2018). How convolutional neural networks see the world—A survey of convolutional neural network visualization methods. Mathematical Foundations of Computing, 1(2), 149–180.

    Article  Google Scholar 

  50. Ross, B. (2001). Natural concepts, psychology of. In N. J. Smelser & P. B. Baltes (Eds.), International encyclopedia of the social & behavioral sciences (pp. 10380–10384). Oxford: Pergamon. https://doi.org/10.1016/B0-08-043076-7/01489-3.

    Google Scholar 

  51. Salisbury, J., & Schneider, S. (2019). Concepts, symbols and computation. In M. Sprevak & M. Colombo (Eds.), The Routledge handbook of the computational mind (pp. 310–322). New York: Routledge.

    Google Scholar 

  52. Shum, J., Hermes, D., Foster, B. L., Dastjerdi, M., Rangarajan, V., Winawer, J., et al. (2013). A brain area for visual numerals. Journal of Neuroscience, 33(16), 6709–6715.

    Article  Google Scholar 

  53. Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., et al. (2017). Mastering the game of Go without human knowledge. Nature, 550, 354–359.

    Article  Google Scholar 

  54. Stephan, A. (2006). The dual role of ‘emergence’ in the philosophy of mind and in cognitive science. Synthese, 151(3), 485.

    Article  Google Scholar 

  55. Tu, J. (1996). Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. Journal of Clinical Epidemiology, 49(11), 1225–1231.

    Article  Google Scholar 

  56. Vondrick, C., Pirsiavash, H., & Torralba, A. (2016). Generating videos with scene dynamics. arxiv:1609.02612.

  57. Wajnerman Paz, A. (2018). An efficient coding approach to the debate on grounded cognition. Synthese, 195(12), 5245–5269.

    Article  Google Scholar 

  58. Zhang, Q. S., & Zhu, S. C. (2018). Visual interpretability for deep learning: A survey. Frontiers of Information Technology & Electronic Engineering, 19(1), 27–39.

    Article  Google Scholar 

  59. Zhou, B., Khosla, A., Lapedriza, A., Oliva, A., & Torralba, A. (2014). Object detectors emerge in deep scene CNNs. arxiv:1412.6856.

Download references

Acknowledgements

The author is deeply indebted to Dr. Rosa María Ruiz-Domínguez (Universidad de Málaga, Málaga, Spain), for her insightful comments about the connections of this work with psychology and cognitive science. He is also grateful to David Teira (Universidad Nacional de Educación a Distancia, Madrid, Spain) and Emanuele Ratti (University of Notre Dame, Notre Dame, Indiana, USA) for their valuable comments. Finally, he would like to thank the anonymous reviewers for their constructive suggestions, which have greatly improved the original manuscript.

Author information

Affiliations

Authors

Corresponding author

Correspondence to Ezequiel López-Rubio.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

López-Rubio, E. Throwing light on black boxes: emergence of visual categories from deep learning. Synthese (2020). https://doi.org/10.1007/s11229-020-02700-5

Download citation

Keywords

  • Deep learning
  • Visual categories
  • Machine learning
  • Cognitive science