Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Whose Truth? Pluralistic Geo-Alignment for (Agentic) AI

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper defines geo-alignment as a measurable property: an AI system is geo-aligned when, for every query and geographic context, its output distribution stays within a tolerance of the locally appropriate distribution.

desk verdict Clean vision paper on geo-alignment with a useful formal definition, but the central claim that alignment norms are spatially learnable is asserted, not demonstrated. read the letter →

arxiv 2508.05432 v1 pith:NX56VRTH submitted 2025-08-07 cs.AI cs.CY

Krzysztof Janowicz , Zilong Liu , Gengchen Mai , Zhangyu Wang , Ivan Majic , Alexandra Fortacz , Grant McKenzie , Song Gao This is my paper
classification cs.AIcs.CY
keywords geo-alignmentpluralisticalignmentAIspatialdependencegeographicbiasagenticGeorepresentationlearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that AI alignment—whether a system behaves in line with societal norms—has a geographic dimension that current alignment research largely ignores. It proposes that a system is geo-aligned when, for every query and every geographic context, the distribution of outputs it produces resembles the distribution a local population would consider appropriate, within a tolerance. The paper formalizes this as a divergence condition and claims that, unlike other forms of pluralism, the local target distribution can be learned because nearby regions tend to share norms, values, and regulations. If the formalization holds, geo-alignment becomes a measurable and trainable property of any generative or agentic AI system, not just a philosophical aspiration. The authors illustrate the need with examples such as region-specific drug regulations and contested borders, where a single global answer is wrong for many users.

What carries the argument

The central object is the locally appropriate output distribution $L(o|q,g)$ paired with the geo-alignment inequality $D(L(\cdot|q,g), S(\cdot|q,g)) < \epsilon$ over all queries and geographic contexts. It gives a quantitative meaning to the question 'whose truth' by making the target a probability distribution over outputs per query and region, with $D$ a divergence such as KL. The companion machinery is spatial autocorrelation: because nearby regions share customs and regulations, $L$ is assumed to vary smoothly in space, which makes the formal definition usable for prediction, interpolation, and learning rather than remaining an unmeasurable ideal.

What would settle it

Take a set of region-sensitive queries with objective legal answers, such as drug regulations by jurisdiction, build L(o|q,g) from statutes, then test whether held-out regions' L values are predicted from neighboring regions by a spatial-autocorrelation model. If prediction is no better than a global frequency baseline, the claim that spatial structure makes geo-alignment learnable fails. A second test: for a contested territorial query, collect local answer distributions within a single region; if they do not stabilize into a distribution, the definition's target L itself is not well-defined.

Watch

Extended reading notes

Core claim

The central claim is Definition 1: an AI system is geo-aligned if for all queries $q \in Q$ and geographic contexts $g \in G$, the dissimilarity $D(L(\cdot|q,g), S(\cdot|q,g))$ stays below a tolerance $\epsilon$, where $L$ is the locally appropriate conditional distribution of outputs and $S$ is the system's output distribution. This converts alignment from a vague societal goal into an evaluable condition on output probabilities. The paper's further claim is that $L$ is not an arbitrary construct: spatial dependence and heterogeneity imply that nearby regions are more likely to have similar alignment needs, so $L$ can be estimated, predicted, and used as a training target. A simplified work

Load-bearing premise

The assumption that there is a stable, locally appropriate answer distribution for each query and place, and that nearby places share it closely enough to predict one another.

Editorial extensions

If this is right

  • Geo-alignment becomes an evaluable property, so benchmarks can score models by divergence from region-specific reference distributions instead of a single global answer.
  • Location-aware systems can be audited for legal and cultural correctness before deployment, catching cases where a model returns a US-centric default for a drug that is regulated elsewhere.
  • Spatial autocorrelation lets alignment generalize: norms learned for well-documented regions can be predicted for data-poor or vague-boundary regions.
  • Agentic AI that acts in physical space will need geo-alignment as a design requirement, because its outputs and actions have place-dependent consequences.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The formalism assigns one $L$ per query-region pair, but regions contain internal value pluralism; a natural extension is to model $L$ as a mixture or distribution over subregional distributions, which the paper's hierarchical grid discussion only hints at.
  • Because the definition includes time inside the geographic context but never formalizes temporal change, a direct extension is geo-temporal alignment: $L$ must be updated as laws and norms shift, as the changing pseudoephedrine regulations already show.
  • The learnability claim is empirically testable now: construct $L$ from statute or survey data for a set of regions, then compare spatial-interpolation predictions for held-out regions against a global baseline; no new training is required.
  • Global debiasing 'corrections' can conflict with geo-alignment; the paper's divergence framework offers a way to quantify that conflict as the distance between the debiased output distribution and the locally appropriate one.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This vision paper argues that AI alignment has an underexplored geographic dimension: what counts as appropriate, truthful, or legal output varies across regions and times. The authors introduce and formalize 'geo-alignment' as distributional closeness between a system's output distribution S(o|q,g) and a hypothetical 'locally appropriate' distribution L(o|q,g), measured by a divergence D (Section 3, Eq. 1). They argue that unlike other dimensions of pluralism, spatial structure is predictable and learnable, and they sketch three research directions: geographic alignment benchmarks, neurosymbolic approaches for deliberative geo-alignment, and learning from spatial structure. The paper serves as a position statement connecting GeoAI research to the broader AI alignment literature, with the formalization illustrated by a KL-divergence example about pseudoephedrine regulation.

Significance. If the proposed framework is accepted, geo-alignment would provide a measurable target for evaluating and training generative and agentic systems across geographic contexts, extending pluralistic alignment with a spatial dimension. The paper's explicit linkage of alignment to spatial autocorrelation and to GeoAI techniques (location encoding, geo-knowledge graphs, S2 grids) is a useful contribution, and the authors are transparent about several limitations, including vague region boundaries and sparse queries. The formalization is internally consistent and the numerical KL example is computed correctly. However, the central notions of L and its spatial smoothness are asserted rather than operationalized or empirically supported, and the paper presents no experiments or concrete measurement protocol. The significance therefore rests on a well-motivated but unvalidated research hypothesis.

major comments (3)
  1. [Section 3, Definition 1 / Eq. (1)] Eq. (1) presupposes that for every query q and context g there exists a well-defined 'locally appropriate' distribution L(o|q,g). The paper acknowledges practical estimation difficulties (vague boundaries, sparsity), but the deeper issue is that for many real queries—especially contested territorial, cultural, or political topics—there may be no unique L: local norms are internally plural, vary across demographic groups, and change over time. The paper's own footnote 3 concedes an 'infinitely many' issue, but the proposed remedy (spatial structure / S2 grids) does not resolve non-uniqueness of normative distributions. Without an operational rule for choosing among candidate L distributions, Definition 1 is not testable and geo-alignment becomes an undefined target. The authors should explicitly restrict Definition 1 to a specified reference population or elicitation procedure, or present
  2. [Section 3, pseudoephedrine example] The only concrete numerical illustration assigns L(o|q,g) = {0.8, 0.15, 0.05} by assertion. The paper does not explain how these probabilities would be measured or derived (e.g., from legal text, surveys, or usage statistics), nor how disagreement among sources would be handled. Since this example carries much of the intuitive weight of the formalization, the lack of an operationalization makes the example illustrative rather than a demonstration of how geo-alignment would actually be computed. Please provide a concrete protocol or reference for estimating L in at least one domain, and address sensitivity of D to errors in L.
  3. [Section 4, Vignette 3: Learning from Spatial Structure] The paper's central differentiator from pluralistic alignment is the claim that 'nearby regions are more likely to have similar regulations and customs' and that spatial structure enables prediction. This is an empirical assumption about the spatial autocorrelation of norms and legal regimes, yet it is supported only by analogy to GeoAI scaling laws and latent style transfer, not by data or citations to studies on spatial dependence of alignment-relevant phenomena. Because the learnability claim is the main reason the proposed approach is preferable to generic pluralistic alignment, this is a load-bearing point. The authors should either provide existing evidence, propose a falsifiable statistical test, or explicitly downgrade the claim to a research hypothesis to be validated in future benchmarks.
minor comments (4)
  1. [Section 3, after Eq. (2)] Notation inconsistency: the two systems are introduced as S1 and S2 but written inline as S1(o|q,g) and S2(o|q,g); use subscripts consistently.
  2. [References, [32]] The reference 'Wang, Nemin Wu, ... and and Mai. 2025' has an incomplete author list and a duplicated 'and'. Please complete and correct.
  3. [Footnote 3] The phrase 'there may be infinitely many of them' is ambiguous: it refers to vague boundaries, but the following dependency clause is unclear. Please rephrase to state what exactly is infinite (possible boundary resolutions, candidate regions, etc.).
  4. [Abstract / Title] The title asks 'Whose Truth?' but the body does not explicitly engage with epistemic aspects of truth or with how competing local truths should be adjudicated; the paper is about alignment with local norms. Consider either expanding the discussion or softening the title's promise.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the formalization is a stipulative definition, and the self-citations are not load-bearing.

full rationale

The paper is a vision/position paper rather than a derivation of predictions from fitted parameters. Definition 1 (Eq. 1) stipulates that a system is geo-aligned when D(L(·|q,g), S(·|q,g)) < epsilon, where L is introduced as the '(hypothetical) locally appropriate distribution of outputs'. This is a definitional target, not a claim derived from something else; it does not reduce a prediction to its inputs because no prediction is derived from Eq. (1) beyond the definition itself. The pseudoephedrine example explicitly uses a 'simplified example' with an assumed L = {0.8, 0.15, 0.05}; these are illustrative values, not fitted parameters, and the subsequent KL computations are arithmetic demonstrations, not empirical predictions. The paper's central forward-looking claim—that nearby regions are more likely to have similar alignment needs and that spatial structure is learnable—is an empirical assumption/analogy, not a circular step. It is supported by analogy to spatial autocorrelation and by citations to prior work including the authors' own GeoAI papers ([16], [31]), but the cited results concern spatial dependence measures and location encoding, not alignment; they are used as motivation, not as a self-referential justification of geo-alignment. No uniqueness theorem, hidden ansatz, or fitted input renamed as prediction appears. The paper thus contains no exhibited reduction of a claimed result to its own premises, and no load-bearing self-citation chain.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central claim rests on two domain assumptions about the nature of norms and space; no free parameters or invented entities are introduced because the paper is a vision paper with no fitted model. The hand-chosen pseudoephedrine probabilities are illustrative examples, not fitted parameters.

assumptions (3)
  • domain assumption For every query q and geographic context g, a well-defined locally appropriate output distribution L(o|q,g) exists and represents shared societal norms.
    Section 3, Definition 1 and the discussion of L estimation difficulties. The entire formalization is vacuous if L cannot be defined for contested regions or intra-region value pluralism.
  • domain assumption Alignment norms are spatially autocorrelated: nearby regions are more likely to have similar alignment needs, and this relation decays with distance in a learnable way.
    Section 1, paragraph starting 'Such a geo-alignment goes beyond pluralistic approaches'. This premise is required for the proposed 'learning from spatial structure' direction; the paper hedges with 'may' but does not test it.
  • domain assumption Geographic context can be discretized (e.g., via hierarchical global grids like S2) so that a calculus of containment holds for norms.
    Section 4, Vignette 3. Needed for the proposal to infer context for subregions; assumes regulatory or normative nesting that may not hold for cultural norms.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Whose Truth? Pluralistic Geo-Alignment for (Agentic) AI." pith.science (2026). https://pith.science/paper/NX56VRTH

@misc{pith2026250805432,
  author       = {Pith},
  title        = {Pith review of: Whose Truth? Pluralistic Geo-Alignment for (Agentic) AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NX56VRTH}},
  note         = {Machine review of arXiv:2508.05432}
}
read the original abstract

AI (super) alignment describes the challenge of ensuring (future) AI systems behave in accordance with societal norms and goals. While a quickly evolving literature is addressing biases and inequalities, the geographic variability of alignment remains underexplored. Simply put, what is considered appropriate, truthful, or legal can differ widely across regions due to cultural norms, political realities, and legislation. Alignment measures applied to AI/ML workflows can sometimes produce outcomes that diverge from statistical realities, such as text-to-image models depicting balanced gender ratios in company leadership despite existing imbalances. Crucially, some model outputs are globally acceptable, while others, e.g., questions about Kashmir, depend on knowing the user's location and their context. This geographic sensitivity is not new. For instance, Google Maps renders Kashmir's borders differently based on user location. What is new is the unprecedented scale and automation with which AI now mediates knowledge, expresses opinions, and represents geographic reality to millions of users worldwide, often with little transparency about how context is managed. As we approach Agentic AI, the need for spatio-temporally aware alignment, rather than one-size-fits-all approaches, is increasingly urgent. This paper reviews key geographic research problems, suggests topics for future work, and outlines methods for assessing alignment sensitivity.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Agentic AI in Remote Sensing: Foundations, Taxonomy, and Emerging Systems

    cs.CV 2026-01 unverdicted novelty 7.0 of 10

    The paper delivers the first comprehensive review and unified taxonomy of agentic AI in remote sensing, covering single-agent copilots, multi-agent systems, planning mechanisms, benchmarks, and a roadmap while noting ...

Reference graph

Works this paper leans on

41 extracted references · 30 canonical work pages · cited by 1 Pith paper

  1. [1]

    Andrea Ballatore, Mark Graham, and Shilad Sen. 2017. Digital hegemonies: the localness of search engine results. Annals of the American Association of Geographers 107, 5 (2017), 1194–1215

    work page 2017

  2. [2]

    Brandon Bennett. 2010. Spatial vagueness. In Methods for Handling Imperfect Spatial Information. Springer, 15–47

    work page 2010

  3. [3]

    Antonio A Casilli, Paola Tubaro, Maxime Cornet, Clément Le Ludec, Juana Torres- Cierpe, and Matheus Viana Braz. 2025. Global Inequalities in the Production of Artificial Intelligence: A Four-Country Study on Data Work. The Handbook of Digital Labor (2025), 217–232. Whose Truth? Pluralistic Geo-Alignment for (Agentic) AI

    work page 2025

  4. [4]

    Fahim Faisal and Antonios Anastasopoulos. 2023. Geographic and Geopolitical Biases of Language Models. In Proceedings of the 3rd Workshop on Multi-lingual Representation Learning (MRL). 139–163

    work page 2023

  5. [5]

    Iason Gabriel. 2020. Artificial intelligence, values, and alignment. Minds and machines 30, 3 (2020), 411–437

    2020

  6. [6]

    Melody Y. Guan, Manas Joglekar, Eric Wallace, Saachi Jain, Boaz Barak, Alec Helyar, Rachel Dias, Andrea Vallone, Hongyu Ren, Jason Wei, Hyung Won Chung, Sam Toyer, Johannes Heidecke, Alex Beutel, and Amelia Glaese. 2025. Deliberative Alignment: Reasoning Enables Safer Language Models. https: //doi.org/10.48550/arXiv.2412.16339 arXiv:2412.16339 [cs]

    arXiv doi:10.48550/arxiv.2412.16339 2025

  7. [7]

    Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola. 2024. The platonic representation hypothesis. arXiv preprint arXiv:2405.07987 (2024)

    arXiv 2024

  8. [8]

    Krzysztof Janowicz. 2023. Philosophical foundations of geoai: Exploring sus- tainability, diversity, and bias in geoai and spatial data science. In Handbook of Geospatial Artificial Intelligence. CRC Press, 26–42

    work page 2023

Show all 41 references
  1. [9]

    Krzysztof Janowicz, Song Gao, Grant McKenzie, Yingjie Hu, and Budhendra Bhaduri. 2020. GeoAI: spatially explicit artificial intelligence techniques for geographic knowledge discovery and beyond. , 625–636 pages

    2020

  2. [10]

    Viet Dac Lai, Chien Van Nguyen, Nghia Trung Ngo, Thuat Nguyen, Franck Dernoncourt, Ryan A Rossi, and Thien Huu Nguyen. 2023. Okapi: Instruction- tuned large language models in multiple languages with reinforcement learning from human feedback. arXiv preprint arXiv:2307.16039 (2023)

    2023 arXiv

  3. [11]

    Bryan Li, Samar Haider, and Chris Callison-Burch. 2024. This land is Your, My land: Evaluating geopolitical bias in language models through territorial disputes. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistic...

    2024

  4. [12]

    Zhenlong Li, Huan Ning, Song Gao, Krzysztof Janowicz, Wenwen Li, Samantha T Arundel, Chaowei Yang, Budhendra Bhaduri, Shaowen Wang, A Zhu, et al

  5. [13]

    Zilong Liu, Krzysztof Janowicz, Ling Cai, Rui Zhu, Gengchen Mai, and Meilin Shi. 2022. Geoparsing: Solved or biased? an evaluation of geographic biases in geoparsing. AGILE: GIScience Series 3 (2022), 9

    2022

  6. [14]

    Zilong Liu, Krzysztof Janowicz, Ivan Majic, Meilin Shi, Alexandra Fortacz, Mina Karimi, Gengchen Mai, and Kitty Currier. 2025. Operationalizing Geographic Diversity for the Evaluation of AI-Generated Content. Transactions in GIS 29, 3 (2025), e70057

    2025

  7. [15]

    Gengchen Mai, Yingjie Hu, Song Gao, Ling Cai, Bruno Martins, Johannes Scholz, Jing Gao, and Krzysztof Janowicz. 2022. Symbolic and subsymbolic GeoAI: Geospatial knowledge graphs and spatially explicit machine learning. Trans. GIS 26, 8 (2022), 3118–3124

    2022

  8. [16]

    Gengchen Mai, Krzysztof Janowicz, Yingjie Hu, Song Gao, Bo Yan, Rui Zhu, Ling Cai, and Ni Lao. 2022. A review of location encoding for GeoAI: methods and applications. International Journal of Geographical Information Science 36, 4 (2022), 639–673

    2022

  9. [17]

    Gengchen Mai, Xiaobai Yao, Yiqun Xie, Jinmeng Rao, Hao Li, Qing Zhu, Ziyuan Li, and Ni Lao. 2024. SRL: Towards a general-purpose framework for spatial representation learning. In Proceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems...

    2024

  10. [18]

    Rohin Manvi, Samar Khanna, Marshall Burke, David Lobell, and Stefano Ermon

  11. [19]

    Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan. 2021. A survey on bias and fairness in machine learning. ACM computing surveys (CSUR) 54, 6 (2021), 1–35

    2021

  12. [20]

    Sonia K Murthy, Tomer Ullman, and Jennifer Hu. 2024. One fish, two fish, but not the whole sea: Alignment reduces language models’ conceptual diversity. arXiv preprint arXiv:2411.04427 (2024)

    2024 arXiv

  13. [21]

    Maximilian Nickel, Kevin Murphy, Volker Tresp, and Evgeniy Gabrilovich. 2015. A review of relational machine learning for knowledge graphs. Proc. IEEE 104, 1 (2015), 11–33

    2015

  14. [22]

    Rida Qadri, Renee Shelby, Cynthia L Bennett, and Emily Denton. 2023. Ai’s regimes of representation: A community-centered study of text-to-image models in south asia. In Proceedings of the 2023 ACM Conference on Fairness, Account- ability, and Transparency. 506–517

    2023

  15. [23]

    Vikram V Ramaswamy, Sing Yu Lin, Dora Zhao, Aaron Adcock, Laurens van der Maaten, Deepti Ghadiyaram, and Olga Russakovsky. 2023. Geode: a geograph- ically diverse evaluation dataset for object recognition. Advances in Neural Information Processing Systems 36 (2023), 66127–66137

    2023

  16. [24]

    Jinmeng Rao, Song Gao, Gengchen Mai, and Krzysztof Janowicz. 2023. Building privacy-preserving and secure geospatial artificial intelligence foundation models (vision paper). InProceedings of the 31st ACM International Conference on Advances in Geographic Information Systems . 1–4

    2023

  17. [25]

    Jinmeng Rao, Song Gao, and Sijia Zhu. 2023. CATS: Conditional Adversarial Trajectory Synthesis for privacy-preserving trajectory data publication using deep learning approaches. International Journal of Geographical Information Science 37, 12 (2023), 2538–2574

    2023

  18. [26]

    Stuart Russell. 2019. Human compatible: AI and the problem of control . Penguin

    2019

  19. [27]

    Shreya Shankar, Yoni Halpern, Eric Breck, James Atwood, Jimbo Wilson, and D Sculley. 2017. No classification without representation: Assessing geodiversity issues in open data sets for the developing world. arXiv preprint arXiv:1711.08536 (2017)

    2017 arXiv

  20. [28]

    Tianxiao Shen, Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2017. Style transfer from non-parallel text by cross-alignment. Advances in neural information processing systems 30 (2017)

    2017

  21. [29]

    Taylor Sorensen, Jared Moore, Jillian Fisher, Mitchell Gordon, Niloofar Mireshghallah, Christopher Michael Rytting, Andre Ye, Liwei Jiang, Ximing Lu, Nouha Dziri, Tim Althoff, and Yejin Choi. 2024. Position: a roadmap to pluralistic alignment. In Proceedings of the 41st Intern...

    2024

  22. [30]

    Yi-Fu Tuan. 1979. Space and place: Humanistic perspective. In Philosophy in geography. Springer, 387–427

    1979

  23. [31]

    Zhangyu Wang, Krzysztof Janowicz, Gengchen Mai, and Ivan Majic. 2024. Probing the Information Theoretical Roots of Spatial Dependence Measures. arXiv:2405.18459 [cs.IT]

    2024 arXiv

  24. [32]

    Zhangyu Wang, Nemin Wu, Qian Cao, Jiangnan Xia, Zeping Liu, Yiqun Xie, Akshay Nambi, Tanuja Ganu, Ni Lao, Ninghao Liu, and and Mai. 2025. GeoBS: Information-Theoretic Quantification of Geographic Bias in AI Models. Under Review (2025)

    2025

  25. [33]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reason- ing in large language models. Advances in neural information processing systems 35 (2022), 24824–24837

    2022

  26. [34]

    Sarah Myers West, Meredith Whittaker, and Kate Crawford. 2019. Discriminating systems. AI Now 2019 (2019), 1–33

    2019

  27. [35]

    Norbert Wiener. 1960. Some Moral and Technical Consequences of Automation:. Science 131, 3410 (1960), 1355–1358

    1960

  28. [36]

    Nemin Wu, Qian Cao, Zhangyu Wang, Zeping Liu, Yanlin Qi, Jielu Zhang, Joshua Ni, X Angela Yao, Hongxu Ma, Lan Mu, et al. 2024. TorchSpatial: A Location Encoding Framework and Benchmark for Spatial Representation Learning. In The Thirty-eight Conference on Neural Information Pr...

    2024

  29. [37]

    Da Yin, Hritik Bansal, Masoud Monajatipoor, Liunian Harold Li, and Kai-Wei Chang. 2022. GeoMLAMA: Geo-Diverse Commonsense Probing on Multilingual Pre-Trained Language Models. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing . 2039–2055

    2022

  30. [38]

    Rui Zhu, Cogan Shimizu, Shirly Stephen, Colby K Fisher, Thomas Thelen, Kitty Currier, Krzysztof Janowicz, Pascal Hitzler, Mark Schildhauer, Wenwen Li, et al

  31. [41]

    arXiv preprint arXiv:2502.13874 (2025)

    The KnowWhereGraph: A Large-Scale Geo-Knowledge Graph for In- terdisciplinary Knowledge Discovery and Geo-Enrichment. arXiv preprint arXiv:2502.13874 (2025)

    2025 arXiv

  32. [2024]

    In Proceedings of the 41st International Conference on Machine Learning (ICML’24)

    Large language models are geographically biased. In Proceedings of the 41st International Conference on Machine Learning (ICML’24) . Article 1409, 16 pages

  33. [2025]

    arXiv preprint arXiv:2503.23633 (2025)

    Giscience in the era of artificial intelligence: A research agenda towards autonomous gis. arXiv preprint arXiv:2503.23633 (2025)

    2025 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.