← Back to Concept Contrast

Concept Contrast Fable

AI impact assessment and data protection impact assessment: one asks whom AI will harm, the other asks how data will leak

Conclusion first:An AI impact assessment focuses on the systemic impact that deployment of an AI system may have on individuals and society — bias, discrimination, and the fairness and transparency of automated decision-making. A data protection impact assessment focuses on privacy risks in personal data processing activities — whether data collection is excessive, whether storage is secure, and whether data subject rights are protected. AIIA asks, "Could AI unfairly deny a loan?" DPIA asks, "Is the user's income data encrypted in transit?"

An AI impact assessment focuses on the systemic impact that deployment of an AI system may have on individuals and society — bias, discrimination, and the fairness and transparency of automated decision-making. A data protection impact assessment focuses on privacy risks in personal data processing activities — whether data collection is excessive, whether storage is secure, and whether data subject rights are protected. AIIA asks, "Could AI unfairly deny a loan?" DPIA asks, "Is the user's income data encrypted in transit?"

Fable

The yamen established two new sets of case-review procedures: one looked at the consequences of a ruling, the other at the rules for keeping the case files.

During the Tianhe era, each prefecture tried out a bronze judgment mirror that could draft verdicts for judicial officials based on old cases. The Court of Judicial Review ordered that two volumes be submitted before the mirror was opened. Most prefectures and counties found this troublesome, so they wrote both volumes as "Use the bronze mirror with caution; keep the key with a designated person," and hastily submitted them. Some clerks did not understand, thinking that since both were meant to prevent trouble, one combined volume should be enough. Shen Heng, however, worried that once combined into one, the questions that most needed to be pressed would be covered up by elegant phrasing.

Shen Heng, the prefectural judge of Qingzhou, did not copy it directly. He first tested the bronze mirror in three types of cases: land disputes, commercial debt disputes, and marital property division. He found that in certain widow inheritance cases, the bronze mirror often followed old rulings that favored the clan; and out-of-town merchants whose accents were recorded vaguely by the clerk were also more likely to be disadvantaged. Shen Heng wrote a separate volume on these people's circumstances, remedies after misjudgment, appeal windows, and deactivation conditions. His first draft was also insufficient. It only listed case types, without writing who would be misjudged and how they could appeal. The Court of Judicial Review returned the volume, requiring that affected persons, remedy deadlines, and deactivation conditions be clearly stated.

In the second volume, he examined the case files themselves. Which files could be shown to the bronze mirror, which names had to be masked, which medical records, family letters, and child testimony could not be shared; when the copyist took the files, how many days until return, who could browse them—all had to leave a trace. The file review also went through rework. After masking the names, Shen Heng found that place names, family birth order, and rare disease names could still point back to the person, so he had to keep weakening the clues.

The neighboring prefecture submitted only a volume of fine words, and three months after opening the mirror trouble broke out: the bronze mirror did not leak records, but it pushed several groups of people toward heavier consequences; at the same time, the names in old cases were pieced together from torn pages by teahouse storytellers. Both matters were serious, but they were not the same kind of error.

Later, when the grand court reviewed the files, it first asked whether the two ledgers each had their own owner. If the same person wrote only one line, "be cautious," it was sent back for rework. Because the harm paths were different, the investigation methods, remediation owners and shutdown conditions were also different.

The Grand Court later made the rules clear: one volume asked how new instruments would change people and affairs; the other asked where the case files came from, how they would be used, and whether they would leak out. The former looked at consequences and responsibility, the latter at materials and boundaries. If either volume was missing, the bronze mirror could not be opened.

AI impact assessment

Systematically assess the social impact of AI systems, focusing on fairness, bias, transparency, and the impact of autonomous decision-making on individual rights. The core question answered is "will this AI system harm certain groups of people."

Data protection impact assessment

Conduct a privacy risk assessment for activities involving personal data processing, focusing on the necessity and proportionality of data collection, security measures, and protection of data subject rights. The core question it answers is "will this data processing activity infringe on personal privacy?"

Story Mapping

  • The machine-judgment office and the iron-judgment instrument: AI systems that trigger both AIIA and DPIA
  • Consequence assessment with instruments: AIIA—measuring the differential impact of automated decisions on specific groups
  • The iron judge weighs women's property cases 15% heavier: a classic scenario of data bias causing unfair decisions
  • Case file use assessment: DPIA—checking the security, de-identification, and access control of data processing
  • Judge Shen investigated the two reports separately: the only case correctly distinguished
  • AI Act + GDPR dual trigger: the same product, two reports, and a real compliance requirement that cannot be merged
  • The intersection of data quality: DPIA governs compliance, AIIA governs consequence—the same data across two dimensions

Applied to AI product compliance documentation and regulatory filings

If your AI product falls under both the AI Act’s high-risk classification and the GDPR’s data processing scope, you must write two separate reports rather than one combined document. The core section of the AIIA is the differential impact analysis—using segmented data to test how the model performs across different groups and providing mitigation measures. The core section of the DPIA is the data flow diagram and security control matrix—describing where data comes from, who processes it, where it is stored, and how it is protected. The two documents may reference each other, but their logic is independent. If you submit an “AI impact assessment” to regulators that contains only data security content and no differential impact analysis, regulators can determine that you have not fulfilled your assessment obligation.

京ICP备2026014272号