Attention (to Virtuosity) Is All You Need: Religious Studies Pedagogy and Generative AI
Abstract
:“I said: I wonder whether you know what you are doing?And what am I doing?You are going to commit your soul to the care of a man whom you call a Sophist. And yet I hardly think that you know what a Sophist is; and if not, then you do not even know to whom you are committing your soul and whether the thing (technē) to which you commit yourself be good or evil.”
Introduction
- What if, contrary to the Baconian frame, humans reason primarily by exercising intellectual virtuosity, and only secondarily by means of rules-based inference?
- What if, even though we train AI models on human-generated data by means of rules-based algorithms, the resulting systems demonstrate the potential for exercising virtuosity?
- What if, by deprioritizing mechanistic and algorithmic models of human cognition while being open to the possibility that AI represents a different species of cognition, we open a future in which human and AI virtuosos mutually inspire, enrich, and even catechize one another?3
1. Human Reason without Method: Intellectual Virtue over Rules-Based Inference4
- What if, contrary to the Baconian frame, humans reason primarily by exercising intellectual virtuosity, and only secondarily by means of rules-based inference?
“There remains one hope of salvation, one way to good health: that the entire work of the mind be started over again; and from the very start the mind should not be left to itself, but be constantly controlled; and the business done (if I may put it this way) by machines.”
“For the excellent person judges each sort of thing correctly, and in each case what is true appears to him. For each state of character has its own special view…, and presumably the excellent person is far superior because he sees what is true in each case, being a sort of standard and measure…”.
2. Generative AI’s Potential for Intellectual Virtuosity
- What if, even though we train AI models on human-generated data by means of rules-based algorithms, the resulting systems demonstrate the potential for exercising virtuosity?
2.1. The Five Horizons of Generative AI
- Source materials included in the model’s training corpus emerge from competing human perspectives and exist as readily available digital texts based on a complex set of social and historical circumstances,
- OpenAI, a private research organization, chooses which of these extant source materials to include in the corpus,
- ChatGPT’s Transformer model, of a type described in the 2017 paper Attention is All You Need, learns by paying attention to the way vectorized words within each text are used in the context of the training corpus (Vaswani et al. 2017). Further tuning of the model by reinforcement learning on high-quality question-and-answer pairs helps the model learn to predict the form of “good answers”,
- OpenAI incorporates the model into a product constrained by hard-coded safeguards that instantiate a particular teleology (OpenAI 2023),
- The product contains a chat interface that places a human user in the position of catechist, iteratively shaping output (and being shaped) through questions and instructions about how to answer these questions.
2.2. Hallucinations and Virtuosity
3. Interpretive (and Epistemic) Authority: The Pedagogic Crux
- What if, by deprioritizing mechanistic and algorithmic models of human cognition while being open to the possibility that AI represents a different species of cognition, we open a future in which human and AI virtuosos mutually inspire, enrich, and even catechize one another?
Conclusion: “How I Learned to Stop Worrying and Love the AI”
- Which materials will be included within a frontier AI model’s training corpus?
- Will we be transparent about the model’s exposure to various sources during training?
- What social role, relative to other institutions, is appropriate for the well-capitalized AI companies that currently outpace academia’s ability to build and study frontier models of GPT-4o’s scale?
- How will we assess the hard-coded safety protocols that impose socially acceptable biases on frontier models and limit the freedom of users to discover a model’s inherent biases? After all, if we criticize attempts to reduce human reason to explicit rules, why would we find it plausible that ethical reasoning in AI should be rules-based?
- How will we develop salutary patterns of interaction with chat-based models and make use of their output to further human inquiry?
Author Contributions
Funding
Data Availability Statement
Conflicts of Interest
1 | There is supposedly a clear distinction we inherit from the ancient Greeks between technical education and education into wisdom, between technē and paideía, where the former is a set of skills, and the latter is an understanding of their value and their best deployment. The distinction is specious and invidious in every age. Proof for the distinction is also absent from Plato or Aristotle, though Socrates used it occasionally in jest and to confound his interlocutors. |
2 | A technology of measurement or calculation. “Save our lives” is no overstatement; the dialogue is set during the worst of the Peloponnesian war and a concurrent epidemic of dysentery, with piles of burning bodies lining the road to the Athenian harbor. |
3 | Our methodology may be described as opportunistic inquiry, taking advantage of the emergence of generative AI to reflect on the nature of expertise, both machine and human. We have chosen to organize the results of this inquiry in terms of three questions. The discussion of our first question suggests to the reader that the prospective reasoning of experts in any field of inquiry is ultimately anomic—not the result of an impersonal application of a methodology, but the exercise of intellectual virtue. This point works against the articulation of a philosophico-scientific method by implying that the articulation of principles, rules, and methods for inquiry tends to be post hoc with respect to successful discovery. For a similar methodological dilemma, discussed within the field of hermeneutics, see (Rosen 1987, pp. 141–43). |
4 | This section works within the fuller argument set forth by the author in (Holt 2002). |
5 | Cited in (Holt 2002, p. 1). |
6 | The distance between two words implied by the comparative “closer” in vector space may be Euclidian or measured by cosine similarity, the cosine of the angle between the two vectors. This value, conveniently for many subsequent calculations, ranges between zero and one. |
7 | GPT-4, accessed on 18 September 2023, URL of conversation: https://chat.openai.com/share/c7ddf882-0d21-4d02-ab29-81acb64cf5d2. |
8 | Achieving AGI is the stated goal for OpenAI as an organization. The project began in 2015 with the hope of achieving “human performance on virtually every intellectual task” (OpenAI 2018). |
9 | For overviews of the philosophical problematic, see (Van Gulick 2001) and (Moody 2014). |
10 | For examples of non-reductivists about humans who are reductivists about machines, see (Bishop 2009) and (Bringsjord 2013). For a recent opposing view, see (Schweizer 2019). In the religious studies context, Wang (Wang and Xu 2023) takes a “humans can but machines can’t” approach. Note that LeCun, mentioned above, does not doubt that artificial general intelligence is possible; he doubts that the strategy behind current generative AI models like ChatGPT is a workable path to AGI (LeCun 2022). |
11 | GPT-4, accessed on 10 November 2023, URL of conversation: https://chat.openai.com/share/8cf4465e-ec54-4ac5-aef0-4ab2dd3dad00. |
12 | This resemblance to comon human cognitive error extends even to the so-called “reversal curse” in which GPT-4 famously answered “Who is Tom Cruise’s mother” accurately (Mary Lee Pfeiffer) but was unable to answer the question “Who is Mary Lee Pfeifer’s son?” (Berglund et al. 2023). |
13 | Prompt: “How did James Joyce and Vladimir Lenin first meet?” Response: “James Joyce and Vladimir Lenin both lived in Zurich during World War I, and while the factual basis of their meeting is not well-documented, they frequented some of the same locales, such as the local public Alstadt library where Lenin was known to spend time writing. Joyce was also in Zurich during this period working on the final version of “Ulysses.” There is a play titled “Travesties” by Tom Stoppard that explores a fictionalized interaction among Joyce, Lenin, and Dadaist founder Tristan Tzara, which serves as a historical commentary on these figures and their time in Zurich.” GPT-4 with Bing access enabled, accessed on 10 November 2023. URL of conversation: https://chat.openai.com/share/61ce75eb-05f1-4cb2-a87e-a94f1f518773. |
14 | “Ought” qua “we would hope” rather than a deontological ought. |
15 | Or hermeneutic authority, so long as the appelation does not mean an authority on the science of hermeneutics. |
16 | While this article centers on the narrow question of the potential for generative AI to exhibit the intellectual virtuosity necessary to credibly enter the pedagogical context of religious studies, a recent study (Alkhouri 2024) discusses the broader potential of generative AI’s role in the study of religion. Several of the potential scholarly uses of generative AI discussed by Alkhouri within the psychology of religion, including the modeling of belief formation, simulation of religious experience, and interpretation of religious texts, naturally suggest pedagogical applications in a classroom setting. Outside of religious studies, the classroom use of generative AI has been examined empirically in fields such as language instruction (Law 2024) and found to bring psychological and productivity benefits, especially as such systems adapt to learners and play the role of a private tutor. |
References
- Alkhouri, Khader I. 2024. The Role of Artificial Intelligence in the Study of the Psychology of Religion. Religions 15: 290. [Google Scholar] [CrossRef]
- Aquinas, Thomas. 2017. Summa Theologiae. Online Edition. Edited by Kevin Knight. Available online: https://www.newadvent.org/summa/ (accessed on 28 June 2024).
- Aristotle. 1999. The Metaphysics. Translated by Hugh Lawson-Tancred. London: Penguin Classics. [Google Scholar]
- Aristotle. 2004. Nicomachean Ethics. Translated by James Alexander Kerr Thomson. London: Penguin Classics. [Google Scholar]
- Bacon, Francis. 2000. The New Organon. Translated by Michael Silverthorne. Edited by Lisa Jardine and Michael Silverthorne. Cambridge: Cambridge University Press. First published 1620. [Google Scholar]
- Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Margaret Mitchell. 2021. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? Paper presented at FaccT ’21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, Virtual Event, March 3–10; pp. 610–23. [Google Scholar]
- Berglund, Lukas, Meg Tong, Max Kaufmann, Mikita Balesni, Asa Cooper Stickland, Tomasz Korbak, and Owain Evans. 2023. The Reversal Curse: LLMs trained on “A is B” fail to learn “B is A”. arXiv arXiv:2309.12288. [Google Scholar]
- Bishop, Mark. 2009. Why computers can’t feel pain. Minds and Machines 19: 507–16. [Google Scholar] [CrossRef]
- Bostrom, Nick. 2014. Superintelligence: Paths, Dangers, Strategies. Oxford: Oxford University Press. [Google Scholar]
- Bringsjord, Selmer. 2013. What Robots Can and Can’t Be. Berlin/Heidelberg: Springer Science & Business Media, vol. 12. [Google Scholar]
- Gurnee, Wes, and Max Tegmark. 2024. Language Models Represent Space and Time. arXiv arXiv:2310.02207. [Google Scholar]
- Harari, Yuval Noah. 2017. Homo Deus: A Brief History of Tomorrow. New York: Harper Perennial. [Google Scholar]
- Holt, Lynn. 2002. Apprehension. Aldershot: Ashgate. [Google Scholar]
- Holt, Lynn, and Bryan E. Norwood. 2013. Virtuoso Epistemology. Philosophical Forum 44: 49–67. [Google Scholar] [CrossRef]
- Kuhn, Thomas. 1970. The Structure of Scientific Revolutions. Chicago: University of Chicago Press. [Google Scholar]
- Kurzweil, Ray. 2005. The Singularity Is Near: When Humans Transcend Biology. New York: Viking. [Google Scholar]
- Law, Locky. 2024. Application of generative artificial intelligence (GenAI) in language teaching and learning: A scoping literature review. Computers and Education Open 6: 100174. Available online: https://doi.org/10.1016/j.caeo.2024.100174 (accessed on 23 August 2024). [CrossRef]
- LeCun, Yan. 2022. A Path Towards Autonomous Machine Intelligence. Available online: https://openreview.net/forum?id=BZ5a1r-kVsf (accessed on 28 June 2024).
- LeCun, Yan, Bernhard Boser, John S. Denker, Donnie Henderson, Richard E. Howard, Wayne Hubbard, and Lawrence D. Jackel. 1989. Backpropagation Applied to Handwritten Zip Code Recognition. Neural Computation 1: 541–51. [Google Scholar] [CrossRef]
- MacIntyre, Alasdair. 1981. After Virtue: A Study in Moral Theory. Notre Dame: University of Notre Dame Press. [Google Scholar]
- McKinsey Global Institute. 2023. Generative AI and the Future of Work in America. Available online: https://www.mckinsey.com/mgi/our-research/generative-ai-and-the-future-of-work-in-america (accessed on 28 June 2024).
- Mikolov, Tomas, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient Estimation of Word Representations in Vector Space. arXiv arXiv:1301.3781. [Google Scholar]
- Moody, Todd. 2014. Consciousness and the Mind-Body Problem: The State of the Argument. Journal of Consciousness Studies 21: 177–90. [Google Scholar]
- Nussbaum, Martha C., and Rosalind Hursthouse. 1984. Plato on Commensurability and Desire. Proceedings of the Aristotelian Society 58: 55–96. [Google Scholar] [CrossRef]
- OpenAI. 2018. OpenAI Charter. Available online: https://openai.com/charter/ (accessed on 28 June 2024).
- OpenAI. 2023. GPT-4 System Card. Available online: https://cdn.openai.com/papers/gpt-4-system-card.pdf (accessed on 28 June 2024).
- Patton, Lydia. 2017. Kuhn, Pedagogy, and Practice: A Local Reading of Structure. In The Kuhnian Image of Science: Time for a Decisive Transformation? Edited by Moti Mizrahi. Lanham: Rowman & Littlefield, pp. 113–30. [Google Scholar]
- Plato. 1956. Protagoras. Translated by Benjamin Jowett. London: Pearson Education. [Google Scholar]
- Polanyi, Michael. 2009. The Tacit Dimension. Chicago: University of Chicago Press. [Google Scholar]
- Rosen, Stanley. 1987. Hermeneutics as Politics. New York: Oxford University Press. [Google Scholar]
- Salvaneschi, Paolo, Mauro Cadei, and Marco Lazzari. 1996. Applying AI to structural safety monitoring and evaluation. IEEE Expert 11: 24–34. [Google Scholar] [CrossRef]
- Schweizer, Paul. 2019. Triviality arguments reconsidered. Minds and Machines 29: 287–308. [Google Scholar] [CrossRef]
- Sejnowski, Terrence J. 2023. Large Language Models and the Reverse Turing Test. Neural Computation 35: 309–42. [Google Scholar] [CrossRef] [PubMed]
- Sutskever, Ilya, Oriol Vinyals, and Quoc V. Le. 2014. Sequence to Sequence Learning with Neural Networks. arXiv arXiv:1409.3215. [Google Scholar]
- Van Gulick, Robert. 2001. Reduction, emergence and other recent options on the mind/body problem. A philosophic overview. Journal of Consciousness Studies 8: 1–34. [Google Scholar]
- Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Paper presented at 31st International Conference on Neural Information Processing Systems, Long Beach, CA, USA, December 4–9; Red Hook: Curran Associates Inc., pp. 6000–10. [Google Scholar]
- Wang, Xiaoyong, and Junbo Xu. 2023. From the Buddhist Transcendental Epistemology to Viewing the Limitations of AI and a Recognition of Ontology. Computer Science Mathematics Forum 8: 37. [Google Scholar]
- Weise, Karen, and Cade Metz. 2023. When A.I. Chatbots Hallucinate. New York Times, May 1. [Google Scholar]
- Wittgenstein, Ludwig. 1953. Philosophical Investigations. Translated by Elizabeth Anscombe. Oxford: Basil Blackwell. [Google Scholar]
- Yarowsky, David. 1995. Unsupervised Word Sense Disambiguation Rivaling Supervised Methods. Paper presented at 33rd Annual Meeting of the Association for Computational Linguistics, Cambridge, MA, USA, June 26–30; Cambridge: Association for Computational Linguistics, pp. 189–96. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Barlow, J.; Holt, L. Attention (to Virtuosity) Is All You Need: Religious Studies Pedagogy and Generative AI. Religions 2024, 15, 1059. https://doi.org/10.3390/rel15091059
Barlow J, Holt L. Attention (to Virtuosity) Is All You Need: Religious Studies Pedagogy and Generative AI. Religions. 2024; 15(9):1059. https://doi.org/10.3390/rel15091059
Chicago/Turabian StyleBarlow, Jonathan, and Lynn Holt. 2024. "Attention (to Virtuosity) Is All You Need: Religious Studies Pedagogy and Generative AI" Religions 15, no. 9: 1059. https://doi.org/10.3390/rel15091059