No CrossRef data available.
Article contents
AI/ML Chatbots’ Souls, or Transformers: Less Than Meets the Eye
Published online by Cambridge University Press: 25 January 2024
Abstract
Given the peculiarly linguistic approach that contemporary philosophers use to apply St. Thomas Aquinas’s arguments on the immateriality of the human soul, this paper will present a Thomistic-inspired evaluation of whether artificial intelligence/machine learning (AI/ML) chatbots’ composition and linguistic performance justify the assertion that AI/ML chatbots have immaterial souls. The first section of the paper will present a strong, but ultimately crucially flawed argument that AI/ML chatbots do have souls based on contemporary Thomistic argumentation. The second section of the paper will provide an overview of the actual computer science models that make artificial neural networks and AI/ML chatbots function, which I hope will assist other theologians and philosophers writing about technology, The third section will present some of Emily Bender’s and Alexander Koller’s objections to AI/ML chatbots being able to access meaning from computational linguistics. The final section will highlight the similarities of Bender’s and Koller’s argument to a fuller presentation of St. Thomas Aquinas’s argument for the immateriality of the human soul, ultimately arguing that the current mechanisms and linguistic activity of AI/ML programming do not constitute activity sufficient to conclude that they have immaterial souls on the strength of St. Thomas’s arguments.
Keywords
- Type
- Article
- Information
- Copyright
- © The Author(s) 2024. Published by Cambridge University Press on behalf of Provincial Council of the English Province of the Order of Preachers.
References
1 In this article, particularly in the ‘Basic Overview’ section, I am deeply indebted to Brendon Boldt of Carnegie Mellon University’s Language Technologies Institute for his clarifications, citation support, and conversation about natural language processing and artificial neural networks (ANNs). Any mistakes lay in my own understanding rather than in his guidance.
2 Martin Klimek, ‘The Google Engineer Who Thinks the Company’s AI Has Come to Life’, Washington Post, 11 June 2022, <https://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/> [accessed 26 June 2023].
3 Bobby Allen, ‘The Google Engineer Who Sees Company’s AI as “Sentient” Thinks a Chatbot Has a Soul’, NPR, 16 June 2022, <https://www.npr.org/2022/06/16/1105552435/google-ai-sentient> [accessed 26 June 2023].
4 Blake Lemoine, ‘Is LaMDA Sentient? – An Interview’, Medium, 11 June 2022, <https://cajundiscordian.medium.com/is-lamda-sentient-an-interview-ea64d916d917> [accessed 26 June 2023].
5 Tiffany Wertheimer, ‘Blake Lemoine: Google Fires Engineer Who Said AI Tech Has Feelings’, BBC News, 23 July 2022, <https://www.bbc.com/news/technology-62275326> [accessed 26 June 2023].
6 Andreas Madsen, Siva Reddy, and Sarath Chandar, ‘Post-hoc Interpretability for Neural NLP: A Survey’, ACM Computing Surveys, 55 (2023), 155:2–55:4.
7 Alan Turing, ‘Computational Machinery and Intelligence’, Mind, 59 (1950), 433–42.
8 Samantha Murphy Kelly, ‘ChatGPT Passes Exams from Law and Business Schools’, CNN Business, 26 January 2023, <https://www.cnn.com/2023/01/26/tech/chatgpt-passes-exams/index.html> [accessed 26 June 2023]; As these examples are not the formal blind test proposed by Turing, these instances are not a pass of a full Turing test. See Emily Bender and Alexander Koller, ‘Climbing Towards NLU: On Meaning, Form, and Understanding in the Age of Data’, in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (2020), p. 5188.
9 Such as the testimony of OpenAI CEO Sam Altman’s testimony before the United States Congress. Mohar Chatterjee, ‘AI Hearing Leaves Washington with 3 Big Questions’, Politico, 16 May 2023, <https://www.politico.com/news/2023/05/16/sam-altmans-congress-ai-chatgpt-00097225> [accessed 26 June 2023].
10 Thomas Aquinas, Summa Theologiae, I, Q. 75, article. 2, corpus (henceforth, ST I, Q. 75, art. 2c).
11 ST I, Q. 75, art. 3c; ST III, Q. 34, art. 2 ad 1.
12 ST I, Q. 75, art. 6c.
13 Robert Sokolowski, Phenomenology and the Human Person (Cambridge: Cambridge University Press, 2008), pp. 63–77, pp. 80–96.
14 David Braine, The Human Person: Animal and Spirit (Notre Dame, IN: University of Notre Dame Press, 1992), pp. 412–20, p. 450.
15 Gyula Klima, ‘Aquinas vs. Buridan on the Universality of Human Concepts and the Immateriality of the Human Intellect’, Philosophica, 47 (2022), 15. <https://doi.org/10.5840/philosophica20228163>.
16 Eleonore Stump, ‘Emergence, Causal Powers, and Aristotelianism in Metaphysics’, in Powers and Capacities in Philosophy: The New Aristotelianism, ed. by John Greco and Ruth Groff (London: Routledge, 2013), pp. 48–68.
17 Frank Rosenblatt, Principles of Neurodynamics (New York City: Spartan Books, 1962); J. J. Hopfield, ‘Neural Networks and Physical Systems with Emergent Collective Computational Abilities’, Proceedings of the National Academy of Sciences of the United States of America, 79 (1982), 2554–58.
18 Citing Mortimer Adler, Klima dismisses this possibility, Karl D. Stephan and Gyula Klima, ‘Artificial Intelligence and Its Natural Limits’, AI & Society, 36 (2021), 13. <https://doi.org/10.1007/s00146-020-00995-z>.
19 Emily Bender et al., ‘On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?’ FAccT (2021), 610–23. <https://doi.org/10.1145/3442188.3445922>.
20 Ashish Vaswani et al., ‘Attention Is All You Need’, 31st Conference on Neural Information Processing Systems (NIPS, 2017), arXiv:1706.03762v5 (2017), pp. 1–15.
21 Jurgen Schmidhuber, ‘Deep Learning in Neural Networks: An Overview’, Technical Report IDSIA-03-14 / arXiv:1404.7828 v4 (2014), pp. 4–5.
22 Ian Goodfellow, Yoshua Bengio, and Aaron Courville, Deep Learning (Cambridge, MA: MIT Press, 2016), pp. 163–64. <https://www.deeplearningbook.org/> [accessed 26 June 2023].
23 Schmidhuber, ‘Deep Learning’, 33.
24 For a very nuanced treatment of terminology in AI/ML chatbots, see Andrew Davison, ‘Machine Learning and Theological Traditions of Analogy’, Modern Theology, 37 (2021), 254–74. <https://doi.org/10.1111/moth.12682>.
25 Goodfellow et al., Deep Learning, pp. 293–94.
26 Goodfellow et al., Deep Learning, p. 171.
27 Yann Lecun, Léon Bottou, Yoshua Bengio, and Patrick Haffner, ‘Gradient-Based Learning Applied to Document Recognition’, Proceedings of the IEEE, 86 (1988), 2279–82.
28 Chiyuan Zhang et al., ‘Understanding Deep Learning Requires Re-Thinking Generalization’, ICLR 5 (2017), 1–15. <https://doi.org/10.48550/arXiv.1611.03530>.
29 Dan Jurafsky and James H. Martin, Speech and Language Processing, 3rd edn (draft), (Stanford, CA: Stanford University Press, 2023), Ch. 3, pp. 1–2. <https://web.stanford.edu/~jurafsky/slp3/3.pdf> [accessed 26 June 2023].
30 Christopher D. Manning, ‘Human Language Understanding & Reasoning’, Dædalus: The Journal of the American Academy of Arts & Sciences 151 (2022), 127. <https://doi.org/10.1162/DAED_a_01905>.
31 Heng-Tze Cheng, ‘LaMDA: Towards Safe, Grounded, and High-Quality Dialog Models for Everything’, GoogleBlog, 21 January 2022, <https://ai.googleblog.com/2022/01/lamda-towards-safe-grounded-and-high.html> [accessed 30 June 2023].
32 Jennifer Elias, ‘Google’s Newest A.I. Model Uses Nearly Five Times More Text Training Data Than Predecessor’, CNBC News, 16 May 2023, <https://www.cnbc.com/2023/05/16/googles-palm-2-uses-nearly-five-times-more-text-data-than-predecessor.html> [accessed 30 June 2023].
33 ChatGPT is based on GPT-3.5, known to have 175 billion weights (Tom B. Brown et al., ‘Language Models Are Few-Shot Learners’, arXiv:2005.14165v4 (2020), p. 8, <https://arxiv.org/abs/2005.14165v4> [accessed 30 June 2023] and GPT-4, whose weight count is not disclosed but is certainly higher than GPT-3.5.
34 GPT-4 (OpenAI, ‘GPT-4 Technical Report’, arXiv:2303.08774 (2023), p. 1, <https://arxiv.org/abs/2303.08774> [accessed 30 June 2023], for example, is based on the Transformer which was introduced in 2017 Vashwani et al., ‘Attention’.
35 Jurgen Schmidhuber, ‘Annotated History of Modern AI and Deep Learning’, Technical Report IDSIA-22-22, arXiv:2212.11279v2 (2022), pp. 5–8, <https://arxiv.org/abs/2212.11279v2> [accessed 30 June 2023].
36 Jason Wei et al., ‘Emergent Abilities of Large Language Models’, Transactions on Machine Learning Research (2022), 1–2. <https://openreview.net/forum?id=yzkSU5zdwD> [accessed 30 June 2023].
37 Bender and Koller, ‘Climbing Towards NLU’, pp. 5186–87.
38 Ibid.
39 Ibid., p. 5187.
40 Ibid.
41 Ibid., p. 5188; John Searle, ‘Minds, Brains, and Programs’, Behavioral and Brain Sciences, 3 (1980), 417–57.
42 Searle, ‘Minds, Brains, and Programs’, 417–20.
43 Bender and Koller, ‘Climbing Towards NLU’, 5188; Searle, ‘Minds, Brains, and Programs’, 420–27.
44 Searle, ‘Minds, Brains, and Programs’, pp. 428–45. For the many dimensions and responses to the Chinese room, see David Cole, ‘The Chinese Room Argument’, in The Stanford Encyclopedia of Philosophy, ed. by Edward N. Zalta and Uri Nodelman (2023 Edition), <https://plato.stanford.edu/archives/sum2023/entries/chinese-room/> [accessed 4 August 2023].
45 Bender and Koller, ‘Climbing Towards NLU’, pp. 5189–90.
46 Ibid.
47 Ibid., p. 5190.
48 Ibid.
49 Bender et al., ‘Stochastic Parrots’, 616.
50 Bender and Koller, ‘Climbing Towards NLU’, p. 5187.
51 St. Thomas De Principiis Naturae, cc. 1-3; David Oderberg, Real Essentialism (London: Routledge, 2009), pp. 47–52; Stump, ‘Emergence’, pp. 48–68.
52 St. Thomas, De Ente et Essentia, c. 1.
53 ST I, Q. 84, art. 4c.
54 For details on this process, see ST I, QQ. 75–79, 84–89; Benjamin Block, ‘Thomas Aquinas on How We Know Essences: The Formation and Perfection of Concepts in the Human Intellect’ (PhD dissertation, The Catholic University of America, 2019), pp. 131–296.
55 ST I Q. 75, art. 2c.
56 Sokolowski, Phenomenology, p. 53.
57 Ibid., p. 55.
58 Ibid., p. 56.
59 Braine, Human Person, p. 353.
60 Ibid., p. 471.
61 Ibid., pp. 471–72.
62 Bender and Koller, ‘Climbing Towards NLU’, 5188; Ned Block, ‘Psychologism and Behaviorism’, The Philosophical Review, 90 (1981), 8–10.
63 Bender et al., ‘Stochastic Parrots’, 617–18.
64 This conclusion also excludes some kinds of human speech production from the category of authentic predication. Having memorized some stock phrases in a language (such as for purposes of tourism), but without a real understanding of the individual words or syntax, would authentically have communicative intent and expression, but would not be authentically predicating; one does not really know the language. In such a case, one does have some authentically-predicated thought to share, but that thought is authentically-predicated in the speaker’s original language before using the stock translation.
Another kind of exclusion involves early speech acquisition or the acquisition of a new kind of knowledge. A mere parroting of phrases one has heard in order to provoke a desired effect is not an authentic predication and is often prone to error. A toddler, for example, simply repeating ‘yes’ in an attempt to get food does not truly express what he is asking for. Political discourse or undergraduate papers may also be examples of parroting phrases without understanding in order to provoke a desired effect. Not all words produced by human beings have authentic predication and communicative intent.
65 Jude Chua Soo Meng, ‘Artificial Intelligence and Thomistic Angelology: A Rejoinder’, Quodlibet, 3 (2001), 3–6.