The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2504
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T07:44:27.260300Z
status: published
lens: p_space
sector: labor
headline: AI-Driven Human Task Substitution: Evidence from Google Translate and ChatGPT
dek: 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 cultur…
gain_title: (none)
problem_title: A difference-in-differences study of an online labor market found ChatGPT produced substantially larger and more uniform drops in demand for both regular translations and creative transcreations than Google Neural Machine Translation did.
trace_subject: (none)
gain_reading: (none)
gain_evidence: (none)
problem_reading: A difference-in-differences study of an online labor market found ChatGPT produced substantially larger and more uniform drops in demand for both regular translations and creative transcreations than Google Neural Machine Translation did.
problem_evidence: ChatGPT produced substantially larger and more uniform reductions for both analytical and intuitive tasks
quick_read: A peer-reviewed study in Management Science compared two AI shocks in the translation industry on a major online labor market: Google Neural Machine Translation (GNMT) and, later, ChatGPT. Using transaction-level data from 2016 to 2017 and 2022 to 2023, it found both reduced demand for human work, but to very different degrees.

GNMT cut demand for analytical tasks, meaning regular translations, by economically meaningful amounts. It had no detectable effect on intuitive tasks, or transcreations, which need creative and cultural adaptation. ChatGPT, by contrast, reduced demand for both kinds of work, with larger and more uniform reductions.

The authors say this means conclusions about which kinds of human work resist automation do not straightforwardly carry over from predictive AI to generative AI. Creative and culturally adaptive work that looked safe from GNMT was not protected from ChatGPT.
limitation: 
tag: Evidence-backed problem
key_points: The study examined two sequential natural experiments in one online labor market for translation. | GNMT is described as a predictive, task-specific AI; ChatGPT as a broad, general-purpose generative AI. | Tasks were split into analytical (regular translations) and intuitive (transcreations needing creative and cultural adaptation). | Researchers estimated difference-in-differences models on transaction-level data from 2016 to 2017 and 2022 to 2023. | GNMT reduced demand for analytical tasks by economically meaningful magnitudes, with no detectable effect on intuitive tasks. | ChatGPT produced substantially larger and more uniform reductions for both task types.
rundown: The researchers treated the launch of GNMT and, later, ChatGPT as natural experiments in a major online labor market for translators. Difference-in-differences models on transaction-level data covered two windows, 2016 to 2017 and 2022 to 2023.

They separated work into analytical tasks, regular translations, and intuitive tasks, transcreations that require creative and cultural adaptation. This split allowed them to test whether creative work was shielded from automation.

For GNMT, demand fell for analytical tasks while intuitive tasks showed no detectable change. For ChatGPT, the reductions were substantially larger and applied to both categories, which the authors describe as a marked expansion in the scope and magnitude of AI substitution.
sources:
- peer_reviewed | Management Science | https://doi.org/10.1287/mnsc.2024.06740 | 2026-10-14
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

bac72d6b8632da27f07cb7db30d0dd8025bba034583fc3dbbd738b95a995749a

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000