Germany AI Asylum Law Draft Leans on Staff Training That Review Says Cannot Stop Automation Bias
Reported by Tech Times1 source ↓
Germany’s draft KIMVG would allow migration authorities to use case files to train AI, monitor procedures and compare applicants with open internet sources. The bill relies on manual review and staff training, but a 2010 review found training cannot prevent automation bias.
Key points
- The bill would allow AI training on case data, procedure monitoring and internet comparisons of applicants
- Thirty-seven civil society groups urged the Bundestag to revise the bill before a Nov. 23 committee hearing
- The bill requires manual review of internet-comparison results and special training for staff
- The draft contains no general duty to anonymize data
Details
The KIMVG draft would amend Germany’s Asylum Act and Residence Act. It permits case data to be used to train, validate and test AI systems, allows authorities to monitor how cases move through procedures, and authorizes comparisons with information from open internet sources. People with asylum or residence files at the Federal Office for Migration and Refugees or local foreigners authorities could be affected.
The bill says staff must manually assess internet-comparison results and receive special training for relevant tasks. The government says automated procedures will not replace legal assessment or make binding decisions. But Tech Times reports that a 2010 Human Factors review found automation bias among both novice and expert users and concluded that training or instructions cannot prevent it.
On Monday, 37 civil society organizations, including AlgorithmWatch, appealed to the Bundestag to revise the bill ahead of a public Interior Committee hearing scheduled for Nov. 23. Their requests include making anonymization the general rule, limiting the use of complete case files and central-register data, and setting stricter conditions for internet matching.
The draft has no general anonymization requirement. Its explanatory memorandum says anonymization or pseudonymization may be unnecessary if technically impossible, disproportionate or substantially harmful to result quality; asylum files can contain sensitive information about matters including ethnicity, religion and health. The bill had its first reading on Oct. 8 and was referred to the Interior Committee.
Written from coverage as of
Images
- Tech Times