Human Expertise vs. Artificial Intelligence: New evidence on building age estimation - Experts survey and ChatGPT answers

Soot, Matthias; Kretzschmar, Daniel; Eberwein, Johannes; Zaddach, Sebastian; Teuber, Andreas; Weitkamp, Alexandra

Human Expertise vs. Artificial Intelligence: New evidence on building age estimation - Experts survey and ChatGPT answers

Documentation of the data datacite.description.TechnicalInfo	reference.csv: two columns: id and reference year results-survey915272-anonym.csv: All columns of the servey (anonymized) results-survey915272-fragentext-anonym.csv: all columns including the long text question (in German) englishQuestions.txt: every line represents a question ergebnisChatGPT2.xlsx: for every ID there are two columns: year given by ChatGPT (e.g. 1B) and uncertainty given by ChatGPT (e.g. 1U). The three rows are the three different attempts using different accounts with ChatGPT
Additional geographical or spatial references datacite.geolocation	Lower Saxony
Countries to which the data refer datacite.geolocation.iso3166	GERMANY
Description of the data datacite.resourceType	The dataset contains the - reference the year of construction of 100 real estates (single family houses) from the purchase price collection of lower saxony (reference.csv) - The survey asking 300+ experts for the year of construction for the 100 real estate (5 evaluation by expert - random sample for every expert) including certainty of the answer from the expert in the range from 1 (very uncertain) to 5 (very certain) (files results-survey915272-anonym.csv (short titles in headers) + results-survey915272-fragentext-anonym (complete questions in header - in German) - The answer of 3 attempts with different accounts to ask ChatGPT for the year of construction including certainty of the answer from ChatGPT in the range from 1 (very uncertain) to 5 (very certain) (ergebnisChatGPT2.xlsx) -Texfile with the used prompts to ask ChatGPT (prompts.txt) englishQuestions.txt: Textfile containing the english translation of questions from results-survey915272-fragentext-anonym.csv 1-100.zip: - PDFs of the answers of 3 attempts asking ChatGPT (using the prompts from prompts.txt) for all 100 images of buildings
Type of the data datacite.resourceTypeGeneral	Dataset
Type of the data datacite.resourceTypeGeneral	Image
Total size of the dataset datacite.size	376466314
Author dc.contributor.author	Soot, Matthias
Author dc.contributor.author	Kretzschmar, Daniel
Author dc.contributor.author	Eberwein, Johannes
Author dc.contributor.author	Zaddach, Sebastian
Author dc.contributor.author	Teuber, Andreas
Author dc.contributor.author	Weitkamp, Alexandra
Upload date dc.date.accessioned	2025-06-03T14:34:06Z
Publication date dc.date.available	2025-06-03T14:34:06Z
Data of data creation dc.date.created	2024-11
Publication date dc.date.issued	2025-06-03
Abstract of the dataset dc.description.abstract	The construction, operation, and demolition of buildings account for a significant portion of global energy demand and CO2 emissions. Preserving existing buildings and encouraging renovation is crucial for sustainability. However, detailed information about the construction age of buildings is sparse. This study explores the potential of large pre-trained visual language models (VLMs) like ChatGPT to estimate building age, comparing their accuracy to that of real estate experts. The research involves a comprehensive survey of experts and the application of ChatGPT promts to a dataset of building images. The results show that ChatGPT makes more accurate statements about the age of buildings than individual experts. Only the collective intelligence of many experts provides better results than the VLM. This suggests that VLMs should be used more extensively than before to determine the age of buildings. Using this knowledge, urban planning can be tailored more specifically to necessary renovations in order to extend the life cycles of buildings and close loops in line with the principle of circularity.
Public reference to this page dc.identifier.uri	https://opara.zih.tu-dresden.de/handle/123456789/1500
Public reference to this page dc.identifier.uri	https://doi.org/10.25532/OPARA-861
Publisher dc.publisher	Technische Universität Dresden
Licence dc.rights	Attribution-NonCommercial-NoDerivatives 4.0 International	en
URI of the licence text dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/4.0/
Specification of the discipline(s) dc.subject.classification	4
Title of the dataset dc.title	Human Expertise vs. Artificial Intelligence: New evidence on building age estimation - Experts survey and ChatGPT answers
Research instruments opara.descriptionInstrument	LimeSurvey
Research instruments opara.descriptionInstrument	ChatGPT-4o

Files

Original bundle

Now showing 1 - 7 of 7

Name:: ergebnisChatGPT2.xlsx
Size:: 11.78 KB
Format:: Microsoft Excel XML
Description:: The answer of 3 attempts with different accounts to ask ChatGPT for the year of construction including certainty of the answer from ChatGPT in the range from 1 (very uncertain) to 5 (very certain) Structure: For every Building-ID (1-100) there are two columns: year given by ChatGPT (e.g. 1B for ID 1) and uncertainty given by ChatGPT (e.g. 1U for ID 1). The three rows are the three different attempts using different accounts with ChatGPT.

Download

Name:: reference.csv
Size:: 1.18 KB
Format:: Comma-Separated Values
Description:: Reference of the year of construction of 100 real estates (single family houses) selected from the purchase price collection of lower saxony. Structure: two columns: id and reference year

Download

Name:: prompts.txt
Size:: 383 B
Format:: Plain Text
Description:: Textfile with the used prompts to ask ChatGPT (prompts.txt)

Download

Name:: results-survey915272-anonym.csv
Size:: 220.01 KB
Format:: Comma-Separated Values
Description:: The survey results asking 300+ experts for the year of construction for the 100 real estate (max. 5 different real estate by expert - random sample for every expert) including certainty of the answer from the expert in the range from 1 (very uncertain) to 5 (very certain) (files results-survey915272-anonym.csv (short titles in headers). Structure: columns represent questions of the survey (anonymized), rows represent experts answers.

Download

Name:: results-survey915272-fragentext-anonym.csv
Size:: 251.33 KB
Format:: Comma-Separated Values
Description:: The survey asking 300+ experts for the year of construction for the 100 real estate (5 evaluation by expert - random sample for every expert) including certainty of the answer from the expert in the range from 1 (very uncertain) to 5 (very certain) results-survey915272-fragentext-anonym (complete questions in header - in German). Structure: columns represent all questions (long text question) rows represent experts answers

Download

Name:: 1-100.zip
Size:: 358.51 MB
Format:: Unknown data format
Description:: 1-100.zip: for the ID 1-100: - PDFs of the Chat for ID 1-100 generated by ChatGPT (last prompt) for the response to the prompts asked with the accounts organized in folders by ID: Used Accounts and Time span: - MS1 (MS) (28.10.2024-8.11.2024) - LM/DK (8.11.2024-19.11.2024) - lm2 (4.12.2024-6.12.2024)

Download

Name:: englishQuestions.txt
Size:: 34.95 KB
Format:: Plain Text
Description:: File containing questions from the survey translated to english with Deep-L.

Download

License bundle

Now showing 1 - 1 of 1

Name:: license.txt
Size:: 4.66 KB
Format:: Item-specific license agreed to upon submission
Description:

Download

Collections

TU Dresden Data Publications