Paramveer S. Dhillon

Tuhin Chakrabarty, Paramveer S. Dhillon

Creative writing has long been considered a uniquely human endeavor, requiring voice and style that machines could not replicate. This assumption is challenged by Generative AI that can emulate thousands of author styles in seconds with negligible marginal labor. To understand this better, we conducted a behavioral experiment where 28 MFA writers (experts) competed against three LLMs in emulating 50 critically acclaimed authors. Based on blind pairwise comparisons by 28 expert judges and 131 lay judges, we find that experts preferred human writing in 82.7% of cases under the in-context prompting condition but this reversed to 62% preference for AI after fine-tuning on authors' complete works. Lay judges, however, consistently preferred AI writing. Debrief interviews with expert writers revealed that their preference for AI writing triggered an identity crisis, eroding aesthetic confidence and questioning what constitutes "good writing." These findings challenge discourse about AI's creative limitations and raise fundamental questions about the future of creative labor.

Paramveer S. Dhillon, Somayeh Molaei, Jiaqi Li et al.

Advances in language modeling have paved the way for novel human-AI co-writing experiences. This paper explores how varying levels of scaffolding from large language models (LLMs) shape the co-writing process. Employing a within-subjects field experiment with a Latin square design, we asked participants (N=131) to respond to argumentative writing prompts under three randomly sequenced conditions: no AI assistance (control), next-sentence suggestions (low scaffolding), and next-paragraph suggestions (high scaffolding). Our findings reveal a U-shaped impact of scaffolding on writing quality and productivity (words/time). While low scaffolding did not significantly improve writing quality or productivity, high scaffolding led to significant improvements, especially benefiting non-regular writers and less tech-savvy users. No significant cognitive burden was observed while using the scaffolded writing tools, but a moderate decrease in text ownership and satisfaction was noted. Our results have broad implications for the design of AI-powered writing tools, including the need for personalized scaffolding mechanisms.

Bohan Zhang, Chengke Bu, Paramveer S. Dhillon

AI writing assistants can reduce effort and improve fluency, but they may also weaken writers' sense of authorship. We study this tension with an ownership-aware co-writing editor that offers on-demand, sentence-level suggestions and tests two common design choices: persona-based coaching and style personalization. In an online study (N=176), participants completed three professional writing tasks: an email without AI help, a proposal with generic AI suggestions, and a cover letter with persona-based coaching, while half received suggestions tailored to a brief sample of their prior writing. Across the two AI-assisted tasks, psychological ownership dropped relative to unassisted writing (about 0.85-1.0 points on a 7-point scale), even as cognitive load decreased (about 0.9 points) and quality ratings stayed broadly similar overall. Persona coaching did not prevent the ownership decline. Style personalization partially restored ownership (about +0.43) and increased AI incorporation in text (+5 percentage points). We distill five design patterns: on-demand initiation, micro-suggestions, voice anchoring, audience scaffolds, and point-of-decision provenance, to guide authorship-preserving writing tools.

Bohan Zhang, Yixin Wang, Paramveer S. Dhillon

In this paper, we examine the collaborative dynamics between humans and language models (LMs), where the interactions typically involve LMs proposing text segments and humans editing or responding to these proposals. Productive engagement with LMs in such scenarios necessitates that humans discern effective text-based interaction strategies, such as editing and response styles, from historical human-LM interactions. This objective is inherently causal, driven by the counterfactual `what-if' question: how would the outcome of collaboration change if humans employed a different text editing/refinement strategy? A key challenge in answering this causal inference question is formulating an appropriate causal estimand: the conventional average treatment effect (ATE) estimand is inapplicable to text-based treatments due to their high dimensionality. To address this concern, we introduce a new causal estimand -- Incremental Stylistic Effect (ISE) -- which characterizes the average impact of infinitesimally shifting a text towards a specific style, such as increasing formality. We establish the conditions for the non-parametric identification of ISE. Building on this, we develop CausalCollab, an algorithm designed to estimate the ISE of various interaction strategies in dynamic human-LM collaborations. Our empirical investigations across three distinct human-LM collaboration scenarios reveal that CausalCollab effectively reduces confounding and significantly improves counterfactual estimation over a set of competitive baselines.

Tuhin Chakrabarty, Jane C. Ginsburg, Paramveer Dhillon

The use of copyrighted books for training AI models has led to numerous lawsuits from authors concerned about AI's ability to generate derivative content. Yet it's unclear if these models can generate high quality literary text while emulating authors' styles. To answer this we conducted a preregistered study comparing MFA-trained expert writers with three frontier AI models: ChatGPT, Claude & Gemini in writing up to 450 word excerpts emulating 50 award-winning authors' diverse styles. In blind pairwise evaluations by 159 representative expert & lay readers, AI-generated text from in-context prompting was strongly disfavored by experts for both stylistic fidelity (OR=0.16, p<10^-8) & writing quality (OR=0.13, p<10^-7) but showed mixed results with lay readers. However, fine-tuning ChatGPT on individual authors' complete works completely reversed these findings: experts now favored AI-generated text for stylistic fidelity (OR=8.16, p<10^-13) & writing quality (OR=1.87, p=0.010), with lay readers showing similar shifts. These effects generalize across authors & styles. The fine-tuned outputs were rarely flagged as AI-generated (3% rate v. 97% for in-context prompting) by best AI detectors. Mediation analysis shows this reversal occurs because fine-tuning eliminates detectable AI stylistic quirks (e.g., cliche density) that penalize in-context outputs. While we do not account for additional costs of human effort required to transform raw AI output into cohesive, publishable prose, the median fine-tuning & inference cost of $81 per author represents a dramatic 99.7% reduction compared to typical professional writer compensation. Author-specific fine-tuning thus enables non-verbatim AI writing that readers prefer to expert human writing, providing empirical evidence directly relevant to copyright's fourth fair-use factor, the "effect upon the potential market or value" of the source works.

Yachuan Liu, Xiaochun Wei, Lin Shi et al.

Large language models (LLMs) face significant challenges in ex-ante reasoning, where analysis, inference, or predictions must be made without access to information from future events. Even with explicit prompts enforcing temporal cutoffs, LLMs often generate outputs influenced by internalized knowledge of events beyond the specified cutoff. This paper introduces a novel task and benchmark designed to evaluate the ability of LLMs to reason while adhering to such temporal constraints. The benchmark includes a variety of tasks: stock prediction, Wikipedia event prediction, scientific publication prediction, and Question Answering (QA), designed to assess factual knowledge under temporal cutoff constraints. We use leakage rate to quantify models' reliance on future information beyond cutoff timestamps. Experimental results reveal that LLMs struggle to consistently adhere to temporal cutoffs across common prompting strategies and tasks, demonstrating persistent challenges in ex-ante reasoning. This benchmark provides a potential evaluation framework to advance the development of LLMs' temporal reasoning ability for time-sensitive applications.

Bohan Zhang, Yixin Wang, Paramveer S. Dhillon

This paper introduces a novel causal framework for multi-stage decision-making in natural language action spaces where outcomes are only observed after a sequence of actions. While recent approaches like Proximal Policy Optimization (PPO) can handle such delayed-reward settings in high-dimensional action spaces, they typically require multiple models (policy, value, and reward) and substantial training data. Our approach employs Q-learning to estimate Dynamic Treatment Regimes (DTR) through a single model, enabling data-efficient policy learning via gradient ascent on language embeddings. A key technical contribution of our approach is a decoding strategy that translates optimized embeddings back into coherent natural language. We evaluate our approach on mental health intervention, hate speech countering, and sentiment transfer tasks, demonstrating significant improvements over competitive baselines across multiple metrics. Notably, our method achieves superior transfer strength while maintaining content preservation and fluency, as validated through human evaluation. Our work provides a practical foundation for learning optimal policies in complex language tasks where training data is limited.

Yachuan Liu, Bohan Zhang, Qiaozhu Mei et al.

Recent work has shown that standard training via empirical risk minimization (ERM) can produce models that achieve high accuracy on average but low accuracy on underrepresented groups due to the prevalence of spurious features. A predominant approach to tackle this group robustness problem minimizes the worst group error (akin to a minimax strategy) on the training data, hoping it will generalize well on the testing data. However, this is often suboptimal, especially when the out-of-distribution (OOD) test data contains previously unseen groups. Inspired by ideas from the information retrieval and learning-to-rank literature, this paper first proposes to use Discounted Cumulative Gain (DCG) as a metric of model quality for facilitating better hyperparameter tuning and model selection. Being a ranking-based metric, DCG weights multiple poorly-performing groups (instead of considering just the group with the worst performance). As a natural next step, we build on our results to propose a ranking-based training method called Discounted Rank Upweighting (DRU), which differentially reweights a ranked list of poorly-performing groups in the training data to learn models that exhibit strong OOD performance on the test data. Results on several synthetic and real-world datasets highlight the superior generalization ability of our group-ranking-based (akin to soft-minimax) approach in selecting and learning models that are robust to group distributional shifts.

Paramveer Dhillon, Sinan Aral

In recent years, there has been significant interest in understanding users' online content consumption patterns. But, the unstructured, high-dimensional, and dynamic nature of such data makes extracting valuable insights challenging. Here we propose a model that combines the simplicity of matrix factorization with the flexibility of neural networks to efficiently extract nonlinear patterns from massive text data collections relevant to consumers' online consumption patterns. Our model decomposes a user's content consumption journey into nonlinear user and content factors that are used to model their dynamic interests. This natural decomposition allows us to summarize each user's content consumption journey with a dynamic probabilistic weighting over a set of underlying content attributes. The model is fast to estimate, easy to interpret and can harness external data sources as an empirical prior. These advantages make our method well suited to the challenges posed by modern datasets. We use our model to understand the dynamic news consumption interests of Boston Globe readers over five years. Thorough qualitative studies, including a crowdsourced evaluation, highlight our model's ability to accurately identify nuanced and coherent consumption patterns. These results are supported by our model's superior and robust predictive performance over several competitive baseline methods.

Jeremy Yang, Dean Eckles, Paramveer Dhillon et al.

Decision makers often want to target interventions so as to maximize an outcome that is observed only in the long-term. This typically requires delaying decisions until the outcome is observed or relying on simple short-term proxies for the long-term outcome. Here we build on the statistical surrogacy and policy learning literatures to impute the missing long-term outcomes and then approximate the optimal targeting policy on the imputed outcomes via a doubly-robust approach. We first show that conditions for the validity of average treatment effect estimation with imputed outcomes are also sufficient for valid policy evaluation and optimization; furthermore, these conditions can be somewhat relaxed for policy optimization. We apply our approach in two large-scale proactive churn management experiments at The Boston Globe by targeting optimal discounts to its digital subscribers with the aim of maximizing long-term revenue. Using the first experiment, we evaluate this approach empirically by comparing the policy learned using imputed outcomes with a policy learned on the ground-truth, long-term outcomes. The performance of these two policies is statistically indistinguishable, and we rule out large losses from relying on surrogates. Our approach also outperforms a policy learned on short-term proxies for the long-term outcome. In a second field experiment, we implement the optimal targeting policy with additional randomized exploration, which allows us to update the optimal policy for future subscribers. Over three years, our approach had a net-positive revenue impact in the range of $4-5 million compared to the status quo.

Paramveer Dhillon, Jordan Rodu, Dean Foster et al.

Unlabeled data is often used to learn representations which can be used to supplement baseline features in a supervised learner. For example, for text applications where the words lie in a very high dimensional space (the size of the vocabulary), one can learn a low rank "dictionary" by an eigen-decomposition of the word co-occurrence matrix (e.g. using PCA or CCA). In this paper, we present a new spectral method based on CCA to learn an eigenword dictionary. Our improved procedure computes two set of CCAs, the first one between the left and right contexts of the given word and the second one between the projections resulting from this CCA and the word itself. We prove theoretically that this two-step procedure has lower sample complexity than the simple single step procedure and also illustrate the empirical efficacy of our approach and the richness of representations learned by our Two Step CCA (TSCCA) procedure on the tasks of POS tagging and sentiment classification.