Eduard Storm
@edthestorm.bsky.social
Obacht: Summer Child here. ☝
Who: Labor Economist who likes to employ ML & NLP techniques to tame Big Data.
Where: PostDoc Researcher & Head of Junior Research Group @ihs.ac.at in #VividVienna
https://sites.google.com/view/eduardstorm/home
Who: Labor Economist who likes to employ ML & NLP techniques to tame Big Data.
Where: PostDoc Researcher & Head of Junior Research Group @ihs.ac.at in #VividVienna
https://sites.google.com/view/eduardstorm/home
November 24, 2025 at 8:50 PM
Not properly captured in our paper, but: broader lesson for GenAI era❓
👉 gains depend not only on what AI can do (automation), but especially if workers can step into expanded task spaces that create new work (augmentation)
👉 People need to build and update skills that 🤝 with AI
12/12
👉 gains depend not only on what AI can do (automation), but especially if workers can step into expanded task spaces that create new work (augmentation)
👉 People need to build and update skills that 🤝 with AI
12/12
November 24, 2025 at 8:47 PM
Not properly captured in our paper, but: broader lesson for GenAI era❓
👉 gains depend not only on what AI can do (automation), but especially if workers can step into expanded task spaces that create new work (augmentation)
👉 People need to build and update skills that 🤝 with AI
12/12
👉 gains depend not only on what AI can do (automation), but especially if workers can step into expanded task spaces that create new work (augmentation)
👉 People need to build and update skills that 🤝 with AI
12/12
Key takeaways:
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
November 24, 2025 at 8:47 PM
Key takeaways:
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
Results suggest: high-skilled benefit, lesser-skilled not so much ➡️ Implications for Inequality?
Estimates vary sharply across earnings distribution:
• Bottom deciles: -8 days, earnings −3.9%
• Top decile: +5 days, earnings +2.5%
👉 Suggestive evidence: AI could widen existing inequalities
10/N
Estimates vary sharply across earnings distribution:
• Bottom deciles: -8 days, earnings −3.9%
• Top decile: +5 days, earnings +2.5%
👉 Suggestive evidence: AI could widen existing inequalities
10/N
November 24, 2025 at 8:46 PM
Results suggest: high-skilled benefit, lesser-skilled not so much ➡️ Implications for Inequality?
Estimates vary sharply across earnings distribution:
• Bottom deciles: -8 days, earnings −3.9%
• Top decile: +5 days, earnings +2.5%
👉 Suggestive evidence: AI could widen existing inequalities
10/N
Estimates vary sharply across earnings distribution:
• Bottom deciles: -8 days, earnings −3.9%
• Top decile: +5 days, earnings +2.5%
👉 Suggestive evidence: AI could widen existing inequalities
10/N
AI exposure expands analytic + interactive tasks, and reduces manual ones.
These task expansions translate into measurable earnings gains, especially through analytic work.
👉 Early AI technologies seem to induce task shifts, consistent with reinstatement effects.
9/N
These task expansions translate into measurable earnings gains, especially through analytic work.
👉 Early AI technologies seem to induce task shifts, consistent with reinstatement effects.
9/N
November 24, 2025 at 8:46 PM
AI exposure expands analytic + interactive tasks, and reduces manual ones.
These task expansions translate into measurable earnings gains, especially through analytic work.
👉 Early AI technologies seem to induce task shifts, consistent with reinstatement effects.
9/N
These task expansions translate into measurable earnings gains, especially through analytic work.
👉 Early AI technologies seem to induce task shifts, consistent with reinstatement effects.
9/N
Clear pattern: expert workers gain modestly, others not.
• Experts: +0.7% earnings (~400€), small gains in days worked in response to doubling in local AI demand
• Lesser-skilled workers: small declines
👉 Job-specific expertise matters (more than formal education or other skill proxies).
8/N
• Experts: +0.7% earnings (~400€), small gains in days worked in response to doubling in local AI demand
• Lesser-skilled workers: small declines
👉 Job-specific expertise matters (more than formal education or other skill proxies).
8/N
November 24, 2025 at 8:46 PM
Clear pattern: expert workers gain modestly, others not.
• Experts: +0.7% earnings (~400€), small gains in days worked in response to doubling in local AI demand
• Lesser-skilled workers: small declines
👉 Job-specific expertise matters (more than formal education or other skill proxies).
8/N
• Experts: +0.7% earnings (~400€), small gains in days worked in response to doubling in local AI demand
• Lesser-skilled workers: small declines
👉 Job-specific expertise matters (more than formal education or other skill proxies).
8/N
On average, rising AI demand does not change workers’ employment stability or annual earnings.
👉 Early AI neither caused broad job loss nor generated large productivity gains.
👉 But: these zero results mask considerable skill heterogeneity...
7/N
👉 Early AI neither caused broad job loss nor generated large productivity gains.
👉 But: these zero results mask considerable skill heterogeneity...
7/N
November 24, 2025 at 8:45 PM
On average, rising AI demand does not change workers’ employment stability or annual earnings.
👉 Early AI neither caused broad job loss nor generated large productivity gains.
👉 But: these zero results mask considerable skill heterogeneity...
7/N
👉 Early AI neither caused broad job loss nor generated large productivity gains.
👉 But: these zero results mask considerable skill heterogeneity...
7/N
Identification Strategy:
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
November 24, 2025 at 8:45 PM
Identification Strategy:
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
AI Exposure rises with skill levels:
👉 Experts face more AI vacancies than helpers, professionals, or specialists
👉 Similar insights by formal education and occupational task structures
Sets the stage for distinct insights by skill groups (more on that later).
5/N
👉 Experts face more AI vacancies than helpers, professionals, or specialists
👉 Similar insights by formal education and occupational task structures
Sets the stage for distinct insights by skill groups (more on that later).
5/N
November 24, 2025 at 8:45 PM
AI Exposure rises with skill levels:
👉 Experts face more AI vacancies than helpers, professionals, or specialists
👉 Similar insights by formal education and occupational task structures
Sets the stage for distinct insights by skill groups (more on that later).
5/N
👉 Experts face more AI vacancies than helpers, professionals, or specialists
👉 Similar insights by formal education and occupational task structures
Sets the stage for distinct insights by skill groups (more on that later).
5/N
AI demand across local labor markets: occupations × regions
Most local labor markets show little AI skill demand, others experienced notable increases.
(e.g.: in 2017 only 9% of local labor markets displayed AI demand, by 2023 ca. 16%)
👉 Key variation: changes in AI skill demand over time.
4/N
Most local labor markets show little AI skill demand, others experienced notable increases.
(e.g.: in 2017 only 9% of local labor markets displayed AI demand, by 2023 ca. 16%)
👉 Key variation: changes in AI skill demand over time.
4/N
November 24, 2025 at 8:44 PM
AI demand across local labor markets: occupations × regions
Most local labor markets show little AI skill demand, others experienced notable increases.
(e.g.: in 2017 only 9% of local labor markets displayed AI demand, by 2023 ca. 16%)
👉 Key variation: changes in AI skill demand over time.
4/N
Most local labor markets show little AI skill demand, others experienced notable increases.
(e.g.: in 2017 only 9% of local labor markets displayed AI demand, by 2023 ca. 16%)
👉 Key variation: changes in AI skill demand over time.
4/N
Stylized Facts on AI Skill Demand (2017–23, Germany)
• Modest in aggregate terms, fluctuates between 1 – 1.5% of all job postings.
• Most demand on unspecified AI skills, #machinelearning, and other technologies popularized prior to the emergence of #GenAI.
3/N
• Modest in aggregate terms, fluctuates between 1 – 1.5% of all job postings.
• Most demand on unspecified AI skills, #machinelearning, and other technologies popularized prior to the emergence of #GenAI.
3/N
November 24, 2025 at 8:44 PM
Stylized Facts on AI Skill Demand (2017–23, Germany)
• Modest in aggregate terms, fluctuates between 1 – 1.5% of all job postings.
• Most demand on unspecified AI skills, #machinelearning, and other technologies popularized prior to the emergence of #GenAI.
3/N
• Modest in aggregate terms, fluctuates between 1 – 1.5% of all job postings.
• Most demand on unspecified AI skills, #machinelearning, and other technologies popularized prior to the emergence of #GenAI.
3/N
How does AI skill demand affect individual workers?
#AI can:
1. displace tasks
2. boost productivity
3. create new tasks
👉 Explore channels in context of longer-term dynamics of the early AI wave (2017-2023).
Data: Online Job Postings + German worker-level admin data (@iabnews.bsky.social)
2/N
#AI can:
1. displace tasks
2. boost productivity
3. create new tasks
👉 Explore channels in context of longer-term dynamics of the early AI wave (2017-2023).
Data: Online Job Postings + German worker-level admin data (@iabnews.bsky.social)
2/N
November 24, 2025 at 8:43 PM
How does AI skill demand affect individual workers?
#AI can:
1. displace tasks
2. boost productivity
3. create new tasks
👉 Explore channels in context of longer-term dynamics of the early AI wave (2017-2023).
Data: Online Job Postings + German worker-level admin data (@iabnews.bsky.social)
2/N
#AI can:
1. displace tasks
2. boost productivity
3. create new tasks
👉 Explore channels in context of longer-term dynamics of the early AI wave (2017-2023).
Data: Online Job Postings + German worker-level admin data (@iabnews.bsky.social)
2/N
November 24, 2025 at 8:18 PM
Key takeaways:
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
November 24, 2025 at 8:11 PM
Key takeaways:
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
How does AI skill demand affect workers’ earnings and employment stability?
• No broad earnings and employment responses
• Gains mostly modest and concentrated among expert workers
• Certain inequality concerns, but also job-augmenting potential
11/12
Identification Strategy:
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
November 24, 2025 at 8:03 PM
Identification Strategy:
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
OLS likely biased
👉 We use a leave-one-out instrument: national AI demand within occupations (excluding worker’s own region).
This approach helps to isolate broad tech shifts from local conditions (see paper for technical details).
6/N
PS: No teaching load, lots of freedom for your own projects, and an interdisciplinary team that values both rigor and collegiality - and all of that in the heart of one of the most liveable cities worldwide.
November 12, 2025 at 5:30 PM
PS: No teaching load, lots of freedom for your own projects, and an interdisciplinary team that values both rigor and collegiality - and all of that in the heart of one of the most liveable cities worldwide.
Link to paper: www.nber.org/system/files...
Link to non-technical summary: openai.com/index/how-pe...
Link to non-technical summary: openai.com/index/how-pe...
www.nber.org
October 3, 2025 at 6:46 AM
Link to paper: www.nber.org/system/files...
Link to non-technical summary: openai.com/index/how-pe...
Link to non-technical summary: openai.com/index/how-pe...