,
17 tweets,
7 min read
Read on Twitter
Recently, the National Crime Records Bureau of India published a tender for the Automated Facial Recognition System (AFRS). Some background (B), and key takeaways (T). #AI #FacialRecognition #ML #India 1/n
B1. The AFRS is conceptualised to modernize the police force, for “criminal identification, verification and its dissemination among various police organizations and units across the country.” 2/n
B2. The rationale: face recognition will help with “facilitating easy recording, analysis, retrieval and sharing of Information” and help in identification of: criminals, missing children/persons, unidentified dead bodies and unknown traced children/persons. 3/n
B3. The government has clarified that this will not violate privacy given the scope mentioned in B2. The nature of FR systems + reading the tender indicates otherwise (thread) 4/n
T1. The AFRS is intended to capture face images from CCTV feeds, newspapers, raids etc, and generate alerts if matches are found w.r.t the Crime and Criminal Tracking Network and Systems. Privacy is definitely at stake. 5/n
T2. AFRS will be able to match a suspect’s photograph using images from Passport, CCTNS, Prisons, Ministry of Women and Child Development, State/National Automated Fingerprint Identification System, or “any other image database”. No purpose limitation, infinite function creep.6/n
T3. CCTVs include those installed in “critical identified locations” as well as video feeds from private or public organization’s video feeds. 7/n
T4. There is also a requirement that the AFRS will provide the ability to search using demographics (!). 8/n
T5. Mandatory features include identification + authentication of facial features, 1:1 verification and 1:N matching. This could mean disastrous consequences if you’re in the wrong place, at the wrong time, and..if you look like a suspect -- the wrong face? 9/n
T6. Data within the AFRS will need to be protected against unauthorized access, accidental modifications, deletions, must be encrypted etc. Nothing about purpose limitation as of now. 10/n
T7. Data collected at each police station and from surveillance cameras will all be stored at the national crime records bureau – again, this is about *everyone*. 11/n
T8. Performance will be tested against parameters like response time, work flow, disaster recovery, etc. Discrimination is not a parameter. 12/n
T9. This is worrying. Facial recognition systems are a fundamental threat to privacy by their very nature. These systems collect data (your face) often without consent, without your knowledge, and mines it to a point of relevance that individuals have no control over. 13/n
T10. Additionally, even if efficiency + justice are the values driving use of facial recognition, this is not the case in practice. Accuracy rates are shockingly low (Delhi police’s FR was reported to be 2% last year), and.. 14/n
T11. To be clear – perfectly accurate and theoretically fair facial recognition systems are *still* problematic. Still inconsistent with the right to privacy under the Indian constitution.. (this deserves a separate thread though..) 15/15
