Empirical Data Explorer

Key Findings Explorer

Explore the quantitative findings behind our analysis, drawn from WHO, UNESCO, NIST, Stanford HAI, and Wang et al. (2024), across employment, sustainability, and healthcare equity.

Expect AI to Impact Jobs
57%
Within the next five years (Stanford HAI, 2024)
Fear AI Job Replacement
36%
Within the next five years (Stanford HAI, 2024)
AI Training Energy Use
1.287 GWh
To train a model like ChatGPT (Wang et al., 2024)
Carbon Change per 1% AI Growth
-0.0013%
National carbon emissions (Wang et al., 2024)
Research Coverage
67 countries
1,809 observations from 1993–2019 (Wang et al., 2024)
Algorithms Flagged
18
Identified by 42 stakeholder respondents (Jain et al., 2023)
Additional research context

Useful findings beyond the final paper

These figures come from research files used during the project but were not quoted in the submitted manuscript because of the paper's page-limit restrictions.

66%

expect AI to dramatically affect their lives within three to five years

Stanford HAI, 2024
52%

express nervousness about AI products and services

Stanford HAI, 2024
55%

of organizations use AI in at least one business function

Stanford HAI, 2024
42% / 59%

report AI-related cost reductions / revenue increases

Stanford HAI, 2024
25

U.S. AI-related regulations in 2023, compared with one in 2016

Stanford HAI, 2024
139

AI-related medical devices approved by the FDA in 2022

Stanford HAI, 2024
14 of 84

AI ethics guidelines referenced sustainability

Jobin et al., 2019
$78M / $191M

estimated compute cost for GPT-4 / Gemini Ultra training

Stanford HAI, 2024 estimates
View Stanford AI Index source View Jobin et al. source
Source Data

Empirical Findings Sample

Source Metric Value Implication
Stanford HAI (2024) Expect AI to affect their job within 5 years 57% Widespread anticipation of workforce change
Stanford HAI (2024) Fear their job could be replaced by AI 36% Signals the need for reskilling investment
Wang et al. (2024) Energy to train a model like ChatGPT 1.287 GWh Roughly equal to 120 American households a year
Wang et al. (2024) Carbon emissions change per 1% AI growth -0.0013% AI can support decarbonization when developed responsibly
Wang et al. (2024) Energy transition change per 1% AI growth +0.0025% Largest coefficient observed among the effects studied
Wang et al. (2024) Ecological footprint change per 1% AI growth -0.0018% Average result across 67 countries from 1993–2019
Wang et al. (2024) Panel observations 1,809 Longitudinal evidence covering 67 countries
Jain et al. (2023) Healthcare algorithm bias against Black patients Underestimated need Real world evidence of bias from proxy variables
Jain et al. (2023) Stakeholder responses analyzed 42 Representatives and individuals contributed 485 pages of evidence
Jain et al. (2023) Algorithms with potential for bias 18 Bias can arise whether or not race is explicitly included
Jain et al. (2023) Qualitative findings 7 themes / 31 subthemes Shows bias can enter every stage of algorithm development
Stanford HAI (2024)Expect AI to dramatically affect their lives in 3 to 5 years66%Up from 60% in the previous year
Stanford HAI (2024)Nervous about AI products and services52%A 13 percentage point increase from 2022
Stanford HAI (2024)Organizations using AI in at least one function55%Up from 20% in 2017
Stanford HAI (2024)Organizations reporting AI cost reductions / revenue increases42% / 59%Evidence of economic benefits alongside social risks
Stanford HAI (2024)U.S. AI-related regulations in 202325Up from one in 2016; 56.3% annual growth
Stanford HAI (2024)FDA-approved AI-related medical devices in 202213912.1% more than 2021 and over 45 times the 2012 total
Jobin et al. (2019)AI ethics guidelines mentioning sustainability14 of 84Far fewer than transparency, which appeared in 73 of 84
Stanford HAI (2024)Estimated training compute cost: GPT-4 / Gemini Ultra$78M / $191MIllustrates the growing resource intensity of frontier models