Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2022.12.23.521610

Comparing scientific abstracts generated by ChatGPT to original abstracts using an artificial intelligence output detector, plagiarism detector, and blinded human reviewers

Gao, C. A.·Howard, F. M.·Markov, N. S.·Dyer, E. C.·Ramesh, S.·Luo, Y.·Pearson, A. T.
View original source ↗Export metadata ↓

Abstract

BackgroundLarge language models such as ChatGPT can produce increasingly realistic text, with unknown information on the accuracy and integrity of using these models in scientific writing. MethodsWe gathered ten research abstracts from five high impact factor medical journals (n=50) and asked ChatGPT to generate research abstracts based on their titles and journals. We evaluated the abstracts using an artificial intelligence (AI) output detector, plagiarism detector, and had blinded human reviewers try to distinguish whether abstracts were original or generated. ResultsAll ChatGPT-generated abstracts were written clearly but only 8% correctly followed the specific journals formatting requirements. Most generated abstracts were detected using the AI output detector, with scores (higher meaning more likely to be generated) of median [interquartile range] of 99.98% [12.73, 99.98] compared with very low probability of AI-generated output in the original abstracts of 0.02% [0.02, 0.09]. The AUROC of the AI output detector was 0.94. Generated abstracts scored very high on originality using the plagiarism detector (100% [100, 100] originality). Generated abstracts had a similar patient cohort size as original abstracts, though the exact numbers were fabricated. When given a mixture of original and general abstracts, blinded human reviewers correctly identified 68% of generated abstracts as being generated by ChatGPT, but incorrectly identified 14% of original abstracts as being generated. Reviewers indicated that it was surprisingly difficult to differentiate between the two, but that the generated abstracts were vaguer and had a formulaic feel to the writing. ConclusionChatGPT writes believable scientific abstracts, though with completely generated data. These are original without any plagiarism detected but are often identifiable using an AI output detector and skeptical human reviewers. Abstract evaluation for journals and medical conferences must adapt policy and practice to maintain rigorous scientific standards; we suggest inclusion of AI output detectors in the editorial process and clear disclosure if these technologies are used. The boundaries of ethical and acceptable use of large language models to help scientific writing remain to be determined.

Source connections

Explore related subjects

scientific communication and educationPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Gao, C. A., Howard, F. M., Markov, N. S., Dyer, E. C., Ramesh, S., Luo, Y., Pearson, A. T.. 2022-12-27. Comparing scientific abstracts generated by ChatGPT to original abstracts using an artificial intelligence output detector, plagiarism detector, and blinded human reviewers. https://doi.org/10.1101/2022.12.23.521610

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Evaluating Large Language Models as Tools to Navigate Researchers in Rapidly Evolving Research Landscapes: A Case Study in Cancer Drug Response Prediction

Large Language Models (LLMs) have emerged as promising tools for assisting researchers in automating and accelerating the synthesis of literature reviews. However, their reliability is a significant concern due to issues like factual inaccuracies and hallucinations. The key question is whether LLMs can reliably provide comprehensive, up-to-date overviews and analyses. This study evaluates the performance of three leading LLMs (OpenAI's ChatGPT, Google's Gemini, and DeepSeek) on the complex task of generating a comprehensive survey paper on deep learning for cancer Drug Response Prediction (DRP). By testing both standard and Deep Research (DR) / Deep Think (DT) modes of LLMs with prompts of varying detail, this paper assesses key academic dimensions, including reference management, content quality, and analytical depth. Key findings reveal that while DR modes of LLMs significantly improve reliability by eliminating hallucinations, performance variations exist across models and prompts. A trade-off between reference quantity and integration quality was observed, and even the best-performing models lacked the analytical depth of human experts, often requiring extensive human supervision. The study concludes that LLMs currently serve as powerful assistive tools but still cannot replace the critical validation and synthesis provided by human researchers. Choosing the best LLM to use depends on the task in hand, while several strategies can be implemented to improve the produced output.

scientific communication and education↗
bioRxiv2024

Retrospective Analysis of the Effects of BWF Interdisciplinary Postdoctoral to Faculty Transition Awards on Future Funding Success

Established by the Burroughs Wellcome Fund (BWF) in 2001, the Career Award at the Scientific Interface (CASI) is a career development award for scientists with doctoral training in the physical/mathematical/computational sciences or engineering conducting postdoctoral research in the biological sciences. The goal of the program is to support early career scientists interested in pursuing an independent research career with an interdisciplinary focus. In order to assess the benefit of the CASI award on recipients, the authors undertook a retrospective analysis of the funding data for CASI recipients to evaluate success against matching cohorts. These cohorts included applicants who succeeded to the final interview stage but were ultimately unsuccessful (interviewed), applicants who submitted proposals but did not make it to the final interview stage (proposal declined), and a randomly selected dataset of researchers from a comparable program, the highly competitive Pathway to Independence Award (K99/R00) from the National Institutes of Health (NIH). The results indicate that CASI recipients outperformed unsuccessful applicants and their K99/R00 counterparts in federal grant rates and overall grant dollars. The authors conclusion affirms that the CASI mechanism and BWF support successfully achieve the objective of invigorating the careers of young investigators, resulting in tangible downstream long-term effects.

scientific communication and education↗
bioRxiv2024

Mapping the Learning Curves of Deep Learning Networks

There is an important challenge in systematically interpreting the internal representations of deep neural networks. This study introduces a multi-dimensional quantification and visualization approach which can capture two temporal dimensions of a model learning experience: the "information processing trajectory" and the "developmental trajectory." The former represents the influence of incoming signals on an agents decision-making, while the latter conceptualizes the gradual improvement in an agents performance throughout its lifespan. Tracking the learning curves of a DNN enables researchers to explicitly identify the model appropriateness of a given task, examine the properties of the underlying input signals, and assess the models alignment (or lack thereof) with human learning experiences. To illustrate the method, we conducted 750 runs of simulations on two temporal tasks: gesture detection and natural language processing (NLP) classification, showcasing its applicability across a spectrum of deep learning tasks. Based on the quantitative analysis of the learning curves across two distinct datasets, we have identified three insights gained from mapping these curves: nonlinearity, pairwise comparisons, and domain distinctions. We reflect on the theoretical implications of this method for cognitive processing, language models and multimodal representation. Author summaryDeep learning networks, specifically recurrent neural networks (RNNs), are designed for processing incoming signals sequentially, making them intuitive computational systems for studying cognitive processing that involves dynamic contexts. There has been a tradition in the fields of machine learning and neuro-cognitive science to examine how a system (either humans or models) represents information through various computational and statistical techniques. Our study takes this one step further by devising a technique for examining the "learning curves" of deep learning networks utilizing the sequential representations as part of RNNs architectures. Just as humans develop learning curves when solving problems, the introduced method captures both how incoming signals help improve decision-making and how a systems problem-solving abilities enhance when encountering the same situation multiple times throughout its lifespan. Our study selected two distinct tasks: gesture detection and emotion tweet classification, to illustrate the insights researchers can draw from mapping models learning curves. The proposed method hinted that gesture learning experiences are smoother, while language learning relies on sudden knowledge gains during processing, corroborating the findings from previous literature.

scientific communication and education↗