Table 5.

Identification rates for prohibited AI systems by stakeholder category (Phase 3)

Prohibited practice (Art. 5)Civil Society (n = 54) Industry(n = 180)CSO/ind. RatioAcademia (n = 41)Citizens (n = 22)Public Auth. (n = 31)
Manipulation/deception (1)(a)63.0%14.4%4.4×36.6%40.9%22.6%
Exploitation of vulnerabilities (1)(b)59.3%10.6%5.6×26.8%27.3%16.1%
Social scoring (1)(c)40.7%16.1%2.5×26.8%40.9%12.9%
Crime risk prediction (1)(d)51.9%5.6%9.3×9.8%18.2%3.2%
Facial image scraping (1)(e)55.6%6.1%9.1×19.5%13.6%3.2%
Emotion recognition (1)(f)53.7%10.6%5.1×22.0%13.6%16.1%
Mean across prohibitions54.0%10.6%5.1×23.6%25.8%12.4%
Note(s):

Identification rate = percentage of each stakeholder group answering “Yes.” n = 379 (raw CSV data). The mean ratio reported in the surrounding text (5.1×) is a simple arithmetic mean across the six prohibition categories

or Create an Account

Close subscription notice
Close access options