Theory vs Practice
Where does real-world AI usage diverge most from theoretical exposure? Ranked by the absolute gap between Anthropic's observed AI usage percentile and the theoretical AIOE percentile. Red rows = usage exceeds theory. Blue rows = theory exceeds usage.
| # | Occupation | Gap (pts) | Direction | AIOE Theory | Net Risk | Risk | Impact |
|---|---|---|---|---|---|---|---|
| 1 | Building painter 71311 | +75 | Above theory | -57% | 33.6% | High | At Risk |
| 2 | Audiologist 22661 | -65 | Below theory | 115% | 2.8% | Very Low | Augmented |
| 3 | Speech therapist 22662 | -65 | Below theory | 115% | 2.8% | Very Low | Augmented |
| 4 | Library officer 34331 | -64 | Below theory | 112% | 16.4% | Moderate | Stable |
| 5 | Website administrator/Webmaster 35140 | +60 | Above theory | -3% | 51.5% | Very High | At Risk |
| 6 | Environmental officer (environmental protection) 21331 | -59 | Below theory | 136% | 7.6% | Low | Augmented |
| 7 | IT Infrastructure technician 35121 | +59 | Above theory | -3% | 27.9% | Moderate | Mixed |
| 8 | IT security technician 35122 | +59 | Above theory | -3% | 24.6% | Moderate | Augmented |
| 9 | IT support technician (including IT user helpdesk technician) 35123 | +59 | Above theory | -3% | 25.3% | Moderate | Mixed |
| 10 | Landscape architect 21621 | -59 | Below theory | 103% | 14.4% | Low | Augmented |
| 11 | Automation engineer (including robotics engineer) 21413 | -56 | Below theory | 128% | 9.7% | Low | Augmented |
| 12 | Manufacturing engineer 21411 | -56 | Below theory | 128% | 11.0% | Low | Augmented |
| 13 | Process engineer 21415 | -56 | Below theory | 128% | 10.4% | Low | Augmented |
| 14 | Production engineer 21412 | -56 | Below theory | 128% | 10.6% | Low | Augmented |
| 15 | Quality control/assurance engineer 21414 | -56 | Below theory | 128% | 10.9% | Low | Augmented |
| 16 | Chemical engineering technician 31161 | +51 | Above theory | 6% | 31.3% | High | At Risk |
| 17 | Chemical engineering technician (petrochemicals) 31163 | +51 | Above theory | 6% | 31.0% | High | At Risk |
| 18 | Environmental engineer 21430 | -50 | Below theory | 134% | 8.5% | Low | Augmented |
| 19 | Sales supervisor 52201 | +48 | Above theory | 5% | 32.8% | High | At Risk |
| 20 | Shop sales assistant 52202 | +48 | Above theory | 5% | 34.3% | High | At Risk |
| 21 | Photographer 34310 | +48 | Above theory | -17% | 17.3% | Moderate | Augmented |
| 22 | Data entry clerk 41320 | +48 | Above theory | 47% | 68.5% | Very High | At Risk |
| 23 | Data processing control clerk 43151 | +48 | Above theory | 47% | 61.5% | Very High | At Risk |
| 24 | Travel consultant/Reservation executive 42210 | +43 | Above theory | -21% | 27.2% | Moderate | At Risk |
| 25 | Cabin attendant/steward 51112 | +43 | Above theory | -21% | 17.1% | Moderate | Stable |
Gap = Anthropic observed usage percentile minus theoretical AIOE percentile. Positive means more AI adoption than theory predicts. Learn more
Frequently asked questions
Where does AI theory diverge from actual usage?
Academic AI exposure indices measure theoretical task automation potential, while Anthropic's observed usage data shows what people actually use AI for. The biggest gaps reveal where adoption lags or leads predictions.
Why do some jobs have high theoretical AI exposure but low real usage?
Regulatory barriers, trust requirements, or workflow integration costs can slow adoption even when tasks are technically automatable. Conversely, some low-exposure roles adopt AI tools faster than predicted.