Eamon Duede

Zhi Hong, Aswathy Ajith, Gregory Pauloski et al.

Transformer-based masked language models such as BERT, trained on general corpora, have shown impressive performance on downstream tasks. It has also been demonstrated that the downstream task performance of such models can be improved by pretraining larger models for longer on more data. In this work, we empirically evaluate the extent to which these results extend to tasks in science. We use 14 domain-specific transformer-based models (including ScholarBERT, a new 770M-parameter science-focused masked language model pretrained on up to 225B tokens) to evaluate the impact of training data, model size, pretraining and finetuning time on 12 downstream scientific tasks. Interestingly, we find that increasing model sizes, training data, or compute time does not always lead to significant improvements (i.e., >1% F1), if at all, in scientific information extraction tasks and offered possible explanations for the surprising performance differences.

Eamon Duede

Philosophers have recently focused on critical, epistemological challenges that arise from the opacity of deep neural networks. One might conclude from this literature that doing good science with opaque models is exceptionally challenging, if not impossible. Yet, this is hard to square with the recent boom in optimism for AI in science alongside a flood of recent scientific breakthroughs driven by AI methods. In this paper, I argue that the disconnect between philosophical pessimism and scientific optimism is driven by a failure to examine how AI is actually used in science. I show that, in order to understand the epistemic justification for AI-powered breakthroughs, philosophers must examine the role played by deep learning as part of a wider process of discovery. The philosophical distinction between the 'context of discovery' and the 'context of justification' is helpful in this regard. I demonstrate the importance of attending to this distinction with two cases drawn from the scientific literature, and show that epistemic opacity need not diminish AI's capacity to lead scientists to significant and justifiable breakthroughs.

Eamon Duede

This paper aims to clarify the representational status of Deep Learning Models (DLMs). While commonly referred to as 'representations', what this entails is ambiguous due to a conflation of functional and relational conceptions of representation. This paper argues that while DLMs represent their targets in a relational sense, in general, we have no good reason to believe that DLMs encode locally semantically decomposable representations of their targets. That is, the representational capacity these models have is largely global, rather than decomposable into stable, local subrepresentations. This result has immediate implications for explainable AI (XAI) and directs attention toward exploring the global relational nature of deep learning representations and their relationship both to models more generally to understand their potential role in future scientific inquiry.

Peiqi Sui, Eamon Duede, Hoyt Long et al.

LLMs hallucinate, yet some confabulations can have social affordances if carefully bounded. We propose critical confabulation (inspired by critical fabulation from literary and social theory), the use of LLM hallucinations to "fill-in-the-gap" for omissions in archives due to social and political inequality, and reconstruct divergent yet evidence-bound narratives for history's "hidden figures". We simulate these gaps with an open-ended narrative cloze task: asking LLMs to generate a masked event in a character-centric timeline sourced from a novel corpus of unpublished texts. We evaluate audited (for data contamination), fully-open models (the OLMo-2 family) and unaudited open-weight and proprietary baselines under a range of prompts designed to elicit controlled and useful hallucinations. Our findings validate LLMs' foundational narrative understanding capabilities to perform critical confabulation, and show how controlled and well-specified hallucinations can support LLM applications for knowledge production without collapsing speculation into a lack of historical accuracy and fidelity.

Eamon Duede, Kevin Davey

Can we acquire apriori knowledge of mathematical facts from the outputs of computer programs? People like Burge have argued (correctly in our opinion) that, for example, Appel and Haken acquired apriori knowledge of the Four Color Theorem from their computer program insofar as their program simply automated human forms of mathematical reasoning. However, unlike such programs, we argue that the opacity of modern LLMs and DNNs creates obstacles in obtaining apriori mathematical knowledge from them in similar ways. We claim though that if a proof-checker automating human forms of proof-checking is attached to such machines, then we can obtain apriori mathematical knowledge from them after all, even though the original machines are entirely opaque to us and the proofs they output may not, themselves, be human-surveyable.

Eamon Duede

This paper examines the role of artificial intelligence in scientific problem-solving, with a focus on its implications for disciplinary creativity. Drawing on recent work in the philosophy of creativity, I distinguish between creative approaches and creative products, and introduce the concept of disciplinary creativity -the creative application of discipline-specific expertise to a valued problem within that field. Through two cases in mathematics, I show that while computation can extend disciplinary creativity, certain approaches involving AI can serve to displace it. This displacement has the potential to alter (and, perhaps, diminish) the value of scientific pursuit.

Peiqi Sui, Eamon Duede, Sophie Wu et al.

This paper presents a systematic defense of large language model (LLM) hallucinations or 'confabulations' as a potential resource instead of a categorically negative pitfall. The standard view is that confabulations are inherently problematic and AI research should eliminate this flaw. In this paper, we argue and empirically demonstrate that measurable semantic characteristics of LLM confabulations mirror a human propensity to utilize increased narrativity as a cognitive resource for sense-making and communication. In other words, it has potential value. Specifically, we analyze popular hallucination benchmarks and reveal that hallucinated outputs display increased levels of narrativity and semantic coherence relative to veridical outputs. This finding reveals a tension in our usually dismissive understandings of confabulation. It suggests, counter-intuitively, that the tendency for LLMs to confabulate may be intimately associated with a positive capacity for coherent narrative-text generation.

Feng Shi, Misha Teplitskiy, Eamon Duede et al.

As political polarization in the United States continues to rise, the question of whether polarized individuals can fruitfully cooperate becomes pressing. Although diversity of individual perspectives typically leads to superior team performance on complex tasks, strong political perspectives have been associated with conflict, misinformation and a reluctance to engage with people and perspectives beyond one's echo chamber. It is unclear whether self-selected teams of politically diverse individuals will create higher or lower quality outcomes. In this paper, we explore the effect of team political composition on performance through analysis of millions of edits to Wikipedia's Political, Social Issues, and Science articles. We measure editors' political alignments by their contributions to conservative versus liberal articles. A survey of editors validates that those who primarily edit liberal articles identify more strongly with the Democratic party and those who edit conservative ones with the Republican party. Our analysis then reveals that polarized teams---those consisting of a balanced set of politically diverse editors---create articles of higher quality than politically homogeneous teams. The effect appears most strongly in Wikipedia's Political articles, but is also observed in Social Issues and even Science articles. Analysis of article "talk pages" reveals that politically polarized teams engage in longer, more constructive, competitive, and substantively focused but linguistically diverse debates than political moderates. More intense use of Wikipedia policies by politically diverse teams suggests institutional design principles to help unleash the power of politically polarized teams.