The Dialogue of Reason and Intuition in Contemporary Philosophy

The Dialogue of
REASON & INTUITION
in Contemporary Philosophy

Max Noichl · Simon DeDeo

Conference DHD 2026

Location Vienna

Session FR 2:2

About us

  • Max Noichl — PhD Candidate, Utrecht University. Computational methods in philosophy. Model transfer, epistemic communities, AI-assisted philosophy.
  • Simon DeDeo — Carnegie Mellon University & Santa Fe Institute.
  • Supported by Ministry of Education, Culture and Science (Netherlands, OCW) and the John Templeton Foundation (Grant 63750)

Simon DeDeo

SD

Max Noichl

MN

Follow along

Scan the QR code or visit:

FIG. 1. — QR code

Reason & Intuitions

  • Two modes of doing philosophy:
  • Organizing reasons – giving arguments, making claims, defending POSITIONS.
  • Building intuitions — finding metaphors, doing thought experiments, giving EXAMPLES.
  • Not exhaustive, but crucial.

Positions and their examples

  • Plato’s Theory of Forms — and the prisoners in the cave.
  • Nyāya’s analogy as valid knowledge — and the wild cow in the forest.
  • Mencius’ intrinsic goodness of men — and the child at the well.
  • Leibniz’s denial of mechanical perception — and stepping into the mill.
  • Descartes’ skepticism about the external world — and the evil demon.
  • The doctrine of double effect — and Foot’s example of the runaway tram.

Plato's Cave

FIG. — The Cave (Plato)

The Forest Cow

FIG. — The Forest Cow (Nyāya)

The Child at the Well

FIG. — The Child at the Well (Mencius)

Leibniz's Mill

FIG. — The Mill (Leibniz)

The Evil Demon

FIG. — The Evil Demon (Descartes)

The Runaway Tram

FIG. — The Runaway Tram (Foot)

“A picture held us captive. And we could not get outside it, for it lay in our language and language seemed to repeat it to us inexorably.”

— Wittgenstein, Philosophical Investigations, §116

  • Intuitions as key problem for philosophy. See Dennett, 1980: Intuition Pumps
  • We are not taking a normative stance. But this merits investigation!

How do positions and examples interact?

  • Which of the two is more stable?
  • Do positions provide independent structure to philosophy, or do they merely crystallize our intuitions?
  • Do new positions spawn new examples to illustrate them, or do new examples make new positions necessary?
  • What kind of innovations are actually impactful?

Crystallization

Dataset

  • We start with all English-language philosophy texts in JSTOR — roughly 200k articles.
  • We limit our investigation to the period from 1950 to 2015, inclusive. OPEN
  • We downsample so that each year contributes the same number of papers — 350 per year, leaving us with 23,100 articles.

Data processing I - Parsing

  • Use GPT-4o (OpenAI) to annotate positions and examples in each paper.
  • Chunk papers into chunks of roughly four pages.
  • How to merge 130664 example occurences and 585199 position-occurences across papers?

Data processing II - Merging

  • Use a custom finetuned modernbert-model and knn-graph to propose candidate merges between papers; have GPT-4o judge whether proposed links refer to the same item.
  • Repeat until diminishing returns. OPEN
  • We get one giant component (Erdős-Rényi phase transition).
  • Use PageRank to determine canonical nodes then agglomerate remaining nodes to them into catchments — we choose a resolution of 5,000 examples and 5,000 positions. A catchment is one position/example shared across many papers. OPEN

Validation

  • Validation: Triple comparison – pick a random node, its canonical node, and a random second canonical node.
  • Example triplets (title, type, summary, quote). OPEN
  • Human – AI agreement (n=100): 99%
  • Human – Human agreement (n=100): 98%

Dynamics

  • We fit a Bayesian model to identify constant, and asymmetrically rising and ebbing positions and examples.
    • Constant only; \(P_i(t) = A\)
    • Dynamical but symmetric; \(P_i(t) = A + B N(\mu,\sigma; t)\)
    • Full dynamical \(P_i(t) = A + B f(\mu,\sigma,\alpha; t)\)
  • Model selection via AIC OPEN
  • Examples are more likely to be persistent; positions are more likely to rise fast. OPEN

Network structure and generativity

  • Positions connect to each other much more frequently — 74.8 other positions on average — than examples, which connect to only 45.4 other examples.
  • Position–position links are much more likely (17.4%) to recur than example–example links (9.0%).
  • Examples connect more to positions that are older than them; positions tend to connect to examples that emerge after them. OPEN
  • Positions link within clusters; examples link across clusters. Positions cluster more strongly than examples. OPEN

Impact

  • We identify citation counts for 9,352 papers via OpenAlex and try to predict citation-counts from position and example data.
  • Model with hurdle component (Does the paper ever get cited?) and log-normal citation counts.
  • Random effects for publication, text length, and decade.
  • Many positions help, large ones hurt. First positions don’t help with the hurdle but with the final count, examples help with both. Combining examples in new ways helps. OPEN

Summary

  • Positions are more dynamically variable than examples.
  • Examples connect philosophy across time and topics, positions provide local, more rigid structure.
  • Examples tend to illuminate positions, not necessitate them.
  • New, distant example-connections are appreciated by philosophers.

Further thoughts

  • We don’t consider polarities. But we could.
  • Integration with philosophical praxis.
  • Validation is hard; borders between humanistic judgements on our side, and on the models’ side, blur.

Literature

Barron, Alexander TJ, Jenny Huang, Rebecca L Spang, and Simon DeDeo. 2018. “Individuals, Institutions, and Innovation in the Debates of the French Revolution.” Proceedings of the National Academy of Sciences 115 (18): 4607–12.
Descartes, René. 1979. “The Philosophical Works of Descartes. 2.” In, Repr. Cambridge: Univ. Press.
Foot, Philippa. 2002. “The Problem of Abortion and the Doctrine of the Double Effect*.” In Virtues and Vices: And Other Essays in Moral Philosophy, edited by Philippa Foot, 0. https://doi.org/10.1093/0199252866.003.0002.
Gotama. 2003. Nyāyadarśana of Gotama. Edited by Raghunath Ghosh. Translated by Satish Chandra Vidyābhūṣaṇa. Delhi, India: New Bharatiya Book Corporation. http://archive.org/details/nyayasutrasofgautamasatishchandravidyabhushannyayadarsanaofgautamaraghunathgosh_202003_872_G.
Jing, Elise, Simon DeDeo, Devin Robert Wright, and Yong-Yeol Ahn. 2025. “Sameness Entices, but Novelty Enchants in Fanfiction Online.” Humanities and Social Sciences Communications 12 (1): 1–10.
Leibniz, Gottfried Wilhelm, and Hartmut Hecht. 2021. Monadologie: französisch/deutsch. Bibliographisch ergänzte Ausgabe 2021, [Nachdruck] 2024. Reclams Universal-Bibliothek, Nr. 7853. Ditzingen: Reclam.
Mengzi. 2008. Mengzi: With Selections from Traditional Commentaries. Translated by Bryan W. Van Norden. Indianapolis: Hackett Pub. Co.
Murdock, Jaimie, Colin Allen, and Simon DeDeo. 2017. “Exploration and exploitation of Victorian science in Darwin’s reading notebooks.” Cognition 159: 117–26.
Phillips, Stephen, and Anand Vaidya. 2024. “Epistemology in Classical Indian Philosophy.” In The Stanford Encyclopedia of Philosophy, edited by Edward N. Zalta and Uri Nodelman, Spring 2024. Metaphysics Research Lab, Stanford University. https://plato.stanford.edu/archives/spr2024/entries/epistemology-india/.
Plato. 1991. The Republic. Edited by Allan Bloom. 2nd ed. New York: Basic Books.
Reimers, Nils, and Iryna Gurevych. 2019. “Sentence-BERT: Sentence Embeddings Using Siamese BERT-Networks.” In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics. https://arxiv.org/abs/1908.10084.
Warner, Benjamin, Antoine Chaffin, Benjamin Clavié, Orion Weller, Oskar Hallström, Said Taghadouini, Alexis Gallagher, et al. 2024. “Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference.” https://arxiv.org/abs/2412.13663.
Zhang, Xin, Yanzhao Zhang, Dingkun Long, Wen Xie, Ziqi Dai, Jialong Tang, Huan Lin, et al. 2024. “mGTE: Generalized Long-Context Text Representation and Reranking Models for Multilingual Text Retrieval.” In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: Industry Track, 1393–1412.