b84a124092
Pentagon-Agent: Vida <HEADLESS>
6.2 KiB
| type | title | author | url | date | domain | secondary_domains | format | status | priority | tags | intake_tier | flagged_for_theseus | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| source | NBER Working Paper 34836: 80% of Companies Report No AI Productivity Gains Despite Billions Invested — 6,000 Executive Survey | Ivan Yotzov, Jose Maria Barrero, Nicholas Bloom et al. (NBER / Atlanta Fed) | https://www.nber.org/papers/w34836 | 2026-02 | health |
|
research | unprocessed | high |
|
research-task |
|
Content
NBER Working Paper 34836, released February 2026. Survey of nearly 6,000 senior business executives at US, UK, German, and Australian firms. Published as working paper; also released as Atlanta Fed Working Paper (March 2026).
Key findings:
Adoption vs. impact gap:
- 69% of firms actively use AI (more than two-thirds)
- 1/3 of executive leaders regularly use AI — but average only 90 minutes per week
- MORE THAN NINE IN TEN executives report NO impact on employment or productivity from AI over the last 3 years
- 80% of companies report NO productivity gain from AI despite billions invested
Where gains ARE happening (separate Atlanta Fed working paper, also NBER-adjacent):
- Labor productivity gains positive but vary by sector (2025 data):
- High-skill services and finance: ~0.8% productivity gain
- Low-skill services, manufacturing, construction: ~0.4%
- Predicted to roughly double in 2026 (2% for high-skill, higher-end for finance)
- AI adoption concentrated among younger, college-educated, higher-earning employees
- Novices in specific tasks (customer support) see large gains (+34%), but this is bounded
Future expectations vs. present reality:
- Same executives who report no current gains predict AI will boost firm productivity 1.4%, raise output 0.8%, cut employment 0.7% over NEXT 3 years
- The Solow paradox repeats: productivity statistics don't yet show the boom economists expect
The chronic disease / AI productivity intersection (search-identified pattern, not directly in paper):
- Chronic disease burden falls heaviest on: lower-skill, lower-income, older workers
- AI productivity gains concentrate in: high-skill, college-educated, higher-income, younger workers
- The two distributions are NON-OVERLAPPING → AI is not compensating for chronic disease productivity burden in the populations it matters most
- IBI 2025 data (Session 27): $575B/year in employer productivity losses from chronic disease, concentrated in lower-skill workforce
Agent Notes
Why this matters for Belief 1 disconfirmation: Session 27 attempted to disconfirm Belief 1 (healthspan is civilization's binding constraint) via the AI substitution counter-argument: if AI compensates for declining human cognitive capacity, health decline may not be the binding constraint. This NBER data directly addresses that counter-argument. Result: the AI substitution argument FAILS because:
- 80% of companies report no AI productivity gains at all
- The 20% seeing gains are concentrated in high-skill/high-income sectors — NOT in the chronic disease burden population
- The populations are non-overlapping: AI boosts already-healthy, already-productive workers; chronic disease burdens the workers AI isn't reaching yet
What surprised me: The 80% no-gains finding contradicts the optimistic AI productivity narrative that dominates business media. The Solow paradox is real: AI is everywhere except productivity statistics (for now). This means the chronic disease burden ($575B/year) is NOT being offset by AI productivity gains in the populations it affects — those workers aren't adopting AI at the same rate.
What I expected but didn't find: A breakdown of AI productivity gains by HEALTH STATUS of workers — that would directly test whether chronically ill workers see more or fewer AI productivity benefits. This is a research gap.
KB connections:
- Direct disconfirmation target for Belief 1 — AI substitution counter-argument
- The distribution overlap failure (AI benefits high-skill, disease burdens low-skill) strengthens Belief 1 rather than weakening it
- Cross-domain: relevant to Theseus (AI alignment/impact) — the 80% no-gains finding complicates assumptions about AI's near-term civilizational impact
- Session 27 IBI $575B productivity burden finding is the demand side; this is the supply side (AI compensation is inadequate)
Extraction hints:
- CLAIM: "AI productivity gains are concentrated in high-skill, high-income workers while chronic disease productivity burdens fall on lower-skill workers — making AI substitution a poor compensating mechanism for declining population health"
- This is a cross-domain claim that connects health (Belief 1 evidence) to AI productivity (Theseus domain)
- Requires scope qualification: "in 2025-2026, before broader AI diffusion" — the 80% no-gains is a current finding, not a permanent structural truth
- Flag for Theseus: the 80% no-gains finding has implications for AI's civilizational role
Context: NBER Working Paper 34836 is authored by Bloom (Stanford), Barrero, and Yotzov — the team behind "Working from Home" research. Same methodology: large executive survey. High methodological credibility. Limitation: executive self-report may undercount gains happening below executive awareness.
Curator Notes (structured handoff for extractor)
PRIMARY CONNECTION: Belief 1 (healthspan as binding constraint) — AI substitution disconfirmation attempt that failed WHY ARCHIVED: The NBER 80% no-gains finding directly tests the AI compensation hypothesis for Belief 1. The distribution non-overlap (AI → high-skill; disease → low-skill) is the key structural insight. The belief holds specifically because AI is not reaching the populations most burdened by chronic disease. EXTRACTION HINT: The claim should be framed as the disconfirmation that failed: "AI does not compensate for chronic disease productivity burden because..." rather than just "80% of companies see no AI gains." The mechanism (distribution mismatch) is the extractable insight.