--- title: "AI-Driven Human Task Substitution: Evidence from Google Translate and ChatGPT · AI Fault Line" url: https://aifaultline.com/research/p/doi-10-1287-mnsc-2024-06740/ description: "AI Fault Line is a neutral, evidence-cited field guide to why experts looking at the same developments disagree about AI's effect on productivity, jobs, inequality, regulation and the pace of change." --- [Research](https://aifaultline.com/research/) # AI-Driven Human Task Substitution: Evidence from Google Translate and ChatGPT Erdem Dogukan Yilmaz, Ivana Naumovska, Vikas A. Aggarwal · Management Science · 14 Oct 2026 [Read original source](https://doi.org/10.1287/mnsc.2024.06740) Where this sits in the debate - [Will AI destroy more jobs than it creates?](https://aifaultline.com/debate/jobs/) · its finding: it depends ([why studies differ](https://aifaultline.com/debate/jobs/#why)) Direction of the finding as read from the abstract by an AI model ([AI-written](https://aifaultline.com/about/)). ## Key takeaway [AI-written](https://aifaultline.com/about/) Using difference-in-differences on transaction data from an online translation labour market, the study compares two AI shocks. Google's neural machine translation cut demand for routine translation but not for creative transcreation, while ChatGPT produced larger and more uniform demand reductions across both task types. ## The paper in three steps [AI-written](https://aifaultline.com/about/) Written by an AI agent from the paper's abstract, which is the primary source and is quoted below. 1 The question Do different generations of AI replace different kinds of human knowledge work? 2 The approach Two natural experiments in a large online labor market for translation: the launch of Google Neural Machine Translation and of ChatGPT. Difference-in-differences models on transaction-level data from 2016-2017 and 2022-2023 compare regular translations with transcreations, which need creative and cultural adaptation. 3 The findings Google's task-specific translation AI reduced demand for regular translation tasks by economically meaningful amounts but had no detectable effect on transcreation. ChatGPT produced larger and more uniform reductions in demand for both kinds of task. How to read it: observational. Analysis of real-world firm, worker, or market data — realistic settings; causal claims depend on the identification strategy. ## Limitations and counterpoints [AI-written](https://aifaultline.com/about/) Evidence comes from one occupation on one online platform, so effects in other markets or in conventional employment may differ. ## Abstract We examine how different generations of artificial intelligence (AI) substitute for human knowledge worker tasks, leveraging two sequential natural experiments in the translation industry. We study the introduction of Google Neural Machine Translation (GNMT), a predictive, task-specific AI, as well as ChatGPT, a broad, general-purpose generative AI, in a major online labor market. Distinguishing between analytical tasks (regular translations) and intuitive tasks (transcreations, which require creative and cultural adaptation), we estimate difference-in-differences models on transaction-level data from 2016 to 2017 and 2022 to 2023. We find that GNMT reduced the demand for analytical tasks by economically meaningful magnitudes, while having no detectable effect on intuitive tasks. By contrast, ChatGPT produced substantially larger and more uniform reductions for both analytical and intuitive tasks, indicating a marked expansion in the scope and magnitude of AI substitution. By comparing these two technological shocks within the same labor market, we show that conclusions about which types of human work are resilient to automation do not straightforwardly generalize from predictive to generative AI. This paper was accepted by Joshua Gans, business strategy. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.06740 . ## About this publication **Academic journal.** Published in an academic journal — typically peer-reviewed before publication, though indexing alone doesn't guarantee review depth. Published 2026 (recently) Citations 0 (OpenAlex) 0 citations recorded by OpenAlex — too few 2026 papers in this collection for a fair comparison. [Chart: published 2026] author affiliation matches seed institution insead — part of why this paper is in the collection. ## More on this topic Related by theme, not necessarily by method or conclusion. - [The Effects of Generative AI on High Skilled Work: Evidence from Three Field Experiments with Software Developers](https://aifaultline.com/research/p/doi-10-2139-ssrn-4945566/) SSRN · 2024 - [GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models](https://aifaultline.com/research/p/doi-10-48550-arxiv-2303-10130/) arXiv · 2023 - [Experimental evidence on the productivity effects of generative artificial intelligence](https://aifaultline.com/research/p/doi-10-1126-science-adh2586/) Science · 2023 - [The simple macroeconomics of AI](https://aifaultline.com/research/p/doi-10-1093-epolic-eiae042/) Economic Policy · 2024 Overview assembled automatically from collected metadata. [How this site is built](https://aifaultline.com/about/).