Unique Identification Authority of India (UIDAI) project - India's reply to SSN

[randomGuy]

You are giving valid concerns but one can't label the project as a failure because of these. Out of the 42 cr or 4200 lacs, 3 lacs records got lost while getting copied from Mumbai regional office to one of the two central data centres. That's bad but that's only 0.07%. This project could only have been taken up with outsourcing of the registration. Otherwise there was no way IMO. Some troll registrar added picture of objects and the people who are supposed to cross-check the docs, photos etc at the back end showed negligence of not checking them. We can'tcall the entire project a failure because of this. For your guard's situation, his card may be stuck with postal department. You can check its status on uidai website. Or you may take the print out from the website by entering details mentioned in acknowledgement receipt which should be with your guard, that's what I did.

[Crookbond]

You are giving valid concerns but one can't label the project as a failure because of these. Out of the 42 cr or 4200 lacs, 3 lacs records got lost while getting copied from Mumbai regional office to one of the two central data centres. That's bad but that's only 0.07%. This project could only have been taken up with outsourcing of the registration. Otherwise there was no way IMO. Some troll registrar added picture of objects and the people who are supposed to cross-check the docs, photos etc at the back end showed negligence of not checking them. We can'tcall the entire project a failure because of this. For your guard's situation, his card may be stuck with postal department. You can check its status on uidai website. Or you may take the print out from the website by entering details mentioned in acknowledgement receipt which should be with your guard, that's what I did.

I think you fail to understand the common pitfall of statistics. 0.07% is not the failure rate (as you make it out to be) - considering the type of data lost (which is said to be lost due to copying - no investigations, no public reports, no follow ups etc.) this is HUGE. You can't just wipe 42 Lakh people out of the face of your country from ANY database. The project is in for a dark period of time. Biometrics is not something one can take lightly. It can be misused and lead you in trouble for the rest of your life. It's not a matter to be taken so casually - only 0.07%, only a region etc. Did you read what I posted? The Aadhar is "enrolled" (issued) but the site gives an error while accessing your e-Aadhar. The postal department is not kind to them and hence, he wanted to avoid the hassle. The reason he refuses to go is because the postal department throws hundreds of Aadhar cards on the ground and asks him to search for it.

[randomGuy]

I think you fail to understand the common pitfall of statistics. 0.07% is not the failure rate (as you make it out to be) - considering the type of data lost (which is said to be lost due to copying - no investigations, no public reports, no follow ups etc.) this is HUGE. You can't just wipe 42 Lakh people out of the face of your country from ANY database. The project is in for a dark period of time. Biometrics is not something one can take lightly. It can be misused and lead you in trouble for the rest of your life. It's not a matter to be taken so casually - only 0.07%, only a region etc. Did you read what I posted? The Aadhar is "enrolled" (issued) but the site gives an error while accessing your e-Aadhar. The postal department is not kind to them and hence, he wanted to avoid the hassle. The reason he refuses to go is because the postal department throws hundreds of Aadhar cards on the ground and asks him to search for it.

Its 3 lac, not 42 lac. They didn't get copied to the data center. It got lost due to unforeseeable technical/infra failure not stolen. There should have been proper clarification after inquiry if there wasn't (sry but i do not have info abt this) My status showed the same, card was stuck with postal. For e-aadhar, things need to be written - ack no. , time stamp of ack generation, DoB If I remember. Format of timestamp is like hr:min:sec, where even ":" (colon) should not be missed. Hope you're doing it fine coz I don't do it right for the 1st time.

[Crookbond]

Its 3 lac, not 42 lac. They didn't get copied to the data center. It got lost due to unforeseeable technical/infra failure not stolen. There should have been proper clarification after inquiry if there wasn't (sry but i do not have info abt this) My status showed the same, card was stuck with postal. For e-aadhar, things need to be written - ack no. , time stamp of ack generation, DoB If I remember. Format of timestamp is like hr:min:sec, where even ":" (colon) should not be missed. Hope you're doing it fine coz I don't do it right for the 1st time.

There are numerous reports where data was lost, stolen, damaged - there have been no follow ups, no inquiries. Even the basics sanity checks of data storage like deduplication etc. The things are written perfectly as they require a strict format adherence. It throws a database error - even a decent error message is not returned :wall:

[randomGuy]

There are 2 data center(in Bangalore and in greater Noida) where all the 1.2 billion records are to be stored, once they are stored there, they wouldn't be lost, rest assured. The loss occurred bcoz newly enrolled records couldn't be copied from the Mumbai regional center and someone at mumbai center hadn't taken backup I suppose. Don't worry abt deduplication. In today's news, gov plans to give legal weight to aadhar through legislation. Privacy concerns also addressed(3 yrs jail time for data thief, up to 1 cr compensation to the sufferer, permissions req. To access data clarified etc) http://www.hindustantimes.com/India-news/NewDelhi/Govt-plans-to--give-aadhaar-legal-weight/Article1-1126786.aspx

[Crookbond]

All these are unacceptable excuses.

[Crookbond]

According to a top secret document disclosed by NSA whistleblower Edward Snowden and obtained by The Hindu, the PRISM programme was deployed by the American agency to gather key information from India by tapping directly into the servers of tech giants which provide services such as email, video sharing, voice-over-IPs, online chats, file transfer and social networking services. And, according to the PRISM document seen by The Hindu, much of the communication targeted by the NSA is unrelated to terrorism, contrary to claims of Indian and American officials. Instead, much of the surveillance was focused on India’s domestic politics and the country’s strategic and commercial interests.

http://www.thehindu.com/news/national/nsa-targets-indian-politics-space-nprogrammes/article5161299.ece Aur store kar lo data in "secure" data centers.

[DomainK]

Well, then it should be easy to mislead them with disinformation? We can make them spend millions following up false leads?

[vvvslaxman]

Tamil Nadu leads in recording of biometric details under NPR http://www.thehindu.com/news/cities/chennai/tamil-nadu-leads-in-recording-of-biometric-details-under-npr/article5205082.ece amil Nadu was relatively unenthusiastic when the Aadhaar programme, as designed by the Unique Identification Authority of India (UIDAI), was launched nearly four years ago. Today, it is well ahead of many other States in enrolment under the National Population Register (NPR) scheme. Going by data provided by the office of the Registrar General and Census Commissioner, Tamil Nadu stands first in recording the biometric details of the highest number of people (aged above five years) as part of the NPR enrolment. The procedures prescribed for enrolment under the UIDAI’s programme and the NPR scheme are identical — capturing of iris, 10 fingerprints and visual image. With the Aadhaar programme failing to find favour in the State, there was not much enrolment. In Tamil Nadu, 20 lakh persons were covered under the UIDAI’s scheme (which was abandoned following a Union government decision about one and a half years ago), according to M.R. V. Krishna Rao, joint director at the Chennai Office of Census Operations. The NPR enrolment began in Tamil Nadu in June 2011 in a phased manner. Yet, there are areas that have not yet been brought under the exercise. For instance, in the old limits of Chennai Corporation, 15 wards are yet to be covered. Authorities, however, are planning to wind up the enrolment throughout the State by December. Subsequently, there are plans to set up permanent enrolment centres. Of the targeted 6.74 crore population, 4.28 crore were covered as on September 30. Of this, Aadhaar numbers were generated by the UIDAI for 2.88 crore. Information available with officials of the revenue department, which coordinates with the Census department to facilitate enrolment, indicates that ‘Aadhaar’ letters have been dispatched to 2.38 crore people by the UIDAI through the postal department. Not an easy process Notwithstanding the elaborate arrangements made, those who go to camps for enrolment face hardship. There are long queues. Invariably, one has to visit the camp at least twice — first, to get a token and then to get the biometric details captured. There have been reports of anxious persons going to the camp sites two hours before the scheduled time for commencement of enrolment, which is 9 a.m. Once a resident gets in, it takes 15-20 minutes to complete the process. As the operations are done on the basis of ‘first come first served’, senior citizens find it extremely difficult to wait. The authorities should consider having separate counters, says Devaki, a resident of Velachery. Mr Rao points out that in places surrounding Tambaram, councillors and public-spirited persons volunteer to help senior citizens. While ruling out the possibility of having exclusive enrolment for a group of 30 or more residents in an apartment complex, he says his office would consider the request from a group of senior citizen homes, located near each other, or those whose movement is restricted due to bad health. Another issue is the production of acknowledgement slips, issued in June-July 2010 during the Houselisting and Housing Census, as a pre-requisite to enrolment. For those who have misplaced the slips, the process becomes longer and more cumbersome. They have to get in touch with local officials to know the details contained in the slips. A senior Corporation official says those who do not have the slips can provide the details of their neighbours. Using them, the relevant details can be traced. Residents who were not covered three years ago can take NPR Household Schedule forms from the respective offices of the local body and provide their biographic details. However, authorities should consider providing the data collected through the NPR Household Schedule online, just as draft and final electoral rolls are, he says. But, officials say the data collected through the NPR exercise is more comprehensive compared to that available in the electoral rolls and for security reasons it is not made available online.

[Crookbond]

A recommended read for people who want to understand more about the technical side of Aadhar. Aadhaar de-duplication myth busted http://moneylife.in/article/aadhaar-de-duplication-myth-busted-any-answers-mr-nilekani/34884.html Highlighting this one in reference to my earlier comments on this thread as to why percentage based statistics may not reveal the gravity of the problem at hand.

The case of Narayanan also mocks the false positive identification rate (FIPR) theory of UIDAI. Earlier, speaking about the FIPR, the UIDAI had said, "We will look at the point where the FPIR (i.e. the possibility that a person is mistaken to be a different person) is 0.0025%". This means, for every 1 lakh comparisons, there would be two and a half false positives. On a large scale, it means for a population of over 120 crore, there would be 18 lakh crore false positives, or, for every single Indian resident there would be 15,000 false positives!

[Detonator]

I have Aadhaar Card :proud:

[randomGuy]

http://forbesindia.com/article/big-bet/how-nandan-nilekani-took-aadhaar-past-the-tipping-point/36259/0

[randomGuy]

A recommended read for people who want to understand more about the technical side of Aadhar. Aadhaar de-duplication myth busted http://moneylife.in/article/aadhaar-de-duplication-myth-busted-any-answers-mr-nilekani/34884.html Highlighting this one in reference to my earlier comments on this thread as to why percentage based statistics may not reveal the gravity of the problem at hand.

This 0.0025% looks a bogus number yar. check this out - http://www.planetbiometrics.com/creo_files/upload/article-files/India_boldly_takes_biometrics_where_no_country_has_gone_before.pdf

I. Two Biometric Modalities The UID system uses ten fingerprints and two iris images for identification. It will accept a minimum of one fingerprint or iris image for verification. Both modalities are treated equally and identically in the system. Few have understood the true impact of the iris decision. NIST reports FPIR rate of ten-finger identification to be between 1.5 to 3.5%ii on a gallery size of approximately one million. UIDAI reports FPIR rate of 0.057% over a gallery size of 100 million. This is a 50 times accuracy improvement in a 100 times larger database. There is another way of looking at the impact. UIDAI reports 2.9% of people have biometrically poor quality fingerprints but only 0.23% have biometrically poor quality fingerprints and iris. Since accuracy deteriorates precipitously even with a small number of poor quality images, an overall ten-fold reduction in poor quality enrollments is the root cause for exceptional FPIR and FNIR rates. A third metric would reinforce this point. It is not uncommon in the literature to see estimations of 1 to 5% failure to enroll (FTE) fingerprint rate. UIDAI reports FTE rate of 0.14%, another 10X improvement. Iris has also helped in “fraud†and unintentional error detection. Reviewing UIDAI results, it appears that 40% of correctly found duplicates had more than one person’s biometrics. These duplicates are about 0.2% (or 2,000/day) of the enrollment, a very significant number. They were easier to detect and eliminate because multi-modal gives higher confidence levels for detecting duplicates. Whichever way performance is measured, iris capture has improved the system 10 to 1000 timesiii . It is simple to verify the scale of the improvement using a back of the envelope calculation. We know that iris and fingerprint are two completely independent modalitiesiv . If they are combined (i.e., AND operation), the resulting FNIR is the multiplication of the individual FNIRs. For example, if FNIR for FP and iris were 1 in 1,000 (0.1%), the combined FNIR would be 1 in 1,000,000 (0.0001%). Multi-modal systems thus get a much larger flexibility to trade off FPIR and FNIR and arrive at an operating point that is several orders of magnitude better than single mode system. In this author’s opinion, the iris decision alone turned the UID system into a roaring biometrics success and averted a potentially catastrophic failure. In hindsight, academics had quantified increased performance of two independent modalities over one modality a long time ago. The UIDAI results should not come as a surprise to them. The author believes that the industry, specifically the buyers and their consultants, were too cautious in the past and waited for someone else to take lead. Then came a newcomer, some would say a naïve UIDAI, and went with the academics' predictions. Whatever the historical reasons for slow adaptation of multiple modalities, it is truly gratifying to see Indonesia and Mexico, two of the larger developing countries using the same approach in their national ID project. Let us hope others too will follow this now “not so new†approach.

[Crookbond]

This 0.0025% looks a bogus number yar. check this out - http://www.planetbiometrics.com/creo_files/upload/article-files/India_boldly_takes_biometrics_where_no_country_has_gone_before.pdf

:wall: Official, Proof of Concept Report form UIDAI website -

we will look at the point where the FPIR (i.e. the possibility that a person is mistaken to be a different person) is 0.0025 %.

http://uidai.gov.in/images/FrontPageUpdates/uid_enrolment_poc_report.pdf Page 23 - still bogus? The less said about the technological/scientific aspect the better.

[randomGuy]

:wall: Official, Proof of Concept Report form UIDAI website - http://uidai.gov.in/images/FrontPageUpdates/uid_enrolment_poc_report.pdf Page 23 - still bogus? The less said about the technological/scientific aspect the better.

What's was banging the head smiley yaar? The link that i gave shows Uidai reports FPIR of 0.057% for 100 million registerations IN THE ACTUAL PROJECT. 0.0025% fpir was for 20k or 40k records as per the PDF you shared, maybe that cannot be extrapolated to the number of records in order of 100 millions coz otherwise the fpir for 100 millions would have been near 100%. This is slightly awkward to imagine but imo, thats what is the actual case.

[randomGuy]

To explain a bit more - Lets take your case of FPIR of 0.0025% on 40k unique records. Now, lets say 40k people from the street came to test. So, 0.0025% of 40k = 1 person should match the existing 40k unique records. Reject that 1 person* and add the other 39,999 persons to your database of unique records. Now you have 79,999 records. the mistake you are making imo, is to assume that the FPIR on 79,999 would have become nearly 2X0.0025%=0.0050%, which is wrong. I mean this would have increased slightly(say this became 0.0026% for ~80k records) but not linearly with the number of records as you imagine. Hope this clears the issue somewhat. *that person can give his biometric maybe one more time or maybe wont be able to ever enroll.

[randomGuy]

And since this is awkward to imagine, I was thinking about the possible reasons which may justify the observations. I came out with one possibility. Maybe the duplicates are matching mainly 1 (or 1 particular group of) record that we've stored and that one record(or group of records) is the one associated with poorly-defined/erased fingerprints. Like in the figure below, person 15000, 65456, and record 'c' are all associated with poorly-defined/erased fingerprints and hence matching. This is one possible explanation of why we are getting FPIR of 0.0025% when comparing against 40k unique records and FPIR of JUST 0.057% when comparing against 100 million unique records. Hope this helps.

[Crookbond]

What's was banging the head smiley yaar? The link that i gave shows Uidai reports FPIR of 0.057% for 100 million registerations IN THE ACTUAL PROJECT. 0.0025% fpir was for 20k or 40k records as per the PDF you shared' date=' maybe that cannot be extrapolated to the number of records in order of 100 millions coz otherwise the fpir for 100 millions would have been near 100%. This is slightly awkward to imagine but imo, thats what is the actual case.[/quote'] Phew! You mised the larger point - something which I wrote a comment when I shared the initial article. You need to understand that 0.057% or 0.0025% looks small in general but in the given context these numbers are BIG and just completely pulls of all claims on the project. First, You said the number is bogus - I said it is NOT "bogus" but a number reported by UIDAI in the PC. Second, since, you are interested in this I have decided to take out some time to show you some calculations about FPIR. Let us assume for sake of simplicity let us consider that the numbers reported are correct and let us perform our calculations for both the scenarios. FPIR1 - 0.0025% FPIR2 - 0.057% Let us also observe, the effect of both these FPIRs on two sets of populations - Population1 = 100M (actual Aadhar registrations), Population2 = 1.2 Billion (actual Indian population). In case of Population 1, there can be (100M * 100M-1)/2=4.99×10^15 unique biometric possible pairs i.e there are 4.99×10^15 comparisons required to prove that a biometric is unique. This would be if were doing this comparisons manually but hey, we have a super fast computer now! However, hang on the computer is not perfect and has a False Postive Identification Rate (FPIR) of FPIR1 and FPIR2. Well, that reduces our manual work but we would still like to know how much manual work we want to do. Now, for Population 1 -- FPIR1 would require 4.99×10^15*(0.0025/100) = 1.25 × 10^11 combinations and FPIR2 would require 4.99×10^15*(0.057/100) = 2.85 × 10^12 combinations to manually check and resolve by hand. In other words, you would have to investigate 1250 False Positives manually for each Indian in case of FPIR1 and 28,500 False Positives manually for each Indian in case of FPIR2. In case of Population 2, there can be (1.2Billion * 1.2Billion-1)/2=7.2 x 10^17 unique biometric possible pairs . Therefore, there are 7.2 x 10^17 comparisons required to prove that a biometric is unique. But, that would be with if the FPIR were 100%. However, the FPIR is FPIR1 and FPIR2. Now, for Population 2 -- -- FPIR1 would require 7.2 x 10^17*(0.0025/100) = 1.8 x 10^13 combinations and FPIR2 would require 7.2 x 10^17*(0.057/100) = 4.1 × 10^14 combinations to manually check and resolve by hand. In other words, you would have to investigate 15,000 False Positives manually for each Indian in case of FPIR1 and 341,666 False Positives manually for each Indian in case of FPIR2. No matter any four combination of numbers (which ever FPIR-Population ratio you take) it is practically infeasible to ensure deduplication. What seems to be an "odd 1 or there" kind of scenario is not that odd. Think of it this way - in the numbers you seem to report are accurate (FPIR2, Population1), one person can have 28500 Aadhar cards (at max.) before he can be detected by the system. And if you didn't notice, and increase in FPIR is a bad thing - notice how the number of combinations required by FPIR2(0.057% - as said by you) over FPIR1 (0.0025% as said by me) for Population1. Hence, the wall smiley.

[randomGuy]

Let me explain a bit more - Point 1 - FPIR1 = 0.0025% when comparing against 40K unique records. This means that there is 0.0025% chance that if we pick up a random person from the street (who is not registered), his/her biometric would falsely match one of the 40k unique records. Now, take FPIR2 = 0.057% when comparing against 100 million unique records. Important thing to note here is that the number of unique records to be compared with, have increased 2500 times (100million/40k). compare this to the increase in FPIR increased by ~22 times. So when you talk about FPIR, you should mention that that FPIR is for comparison against how many unique records. Point 2 - Lets assume the average FPIR when comparing a random person(who isnt enrolled) picked from street against 1 unique record,2 unique records, 3 unique records, 4 unique records, 5 unique records,.....1.2 billion unique records is 0.060% .(ofcourse it will be near 0.060% coz it HAS to be close to 0.057% from what I've tried to explain in past 3 posts) It simply means that - 0.06% of 1.2billion = 7,20,000 people will NOT be able to enroll because their biometric matches falsely with some other unique record. So, basically 7.2 lac of 12,000 lac (i.e. 1.2 billion) people will not be able to enroll. This figure would be acceptable to most concerned with UID as well as with the gov.

[Crookbond]

Let me explain a bit more - Point 1 - FPIR1 = 0.0025% when comparing against 40K unique records. This means that there is 0.0025% chance that if we pick up a random person from the street (who is not registered), his/her biometric would falsely match one of the 40k unique records. Now, take FPIR2 = 0.057% when comparing against 100 million unique records. Important thing to note here is that the number of unique records to be compared with, have increased 2500 times (100million/40k). compare this to the increase in FPIR increased by ~22 times. So when you talk about FPIR, you should mention that that FPIR is for comparison against how many unique records.

This is totally random! FPIR shouldn't increase with the size of dataset - if it does, it shows that the initial experiments conducted were faulty! Moreover, even if it did one should know "Why" - there is no explanation from UIDAI on this front.

Point 2 - Lets assume the average FPIR when comparing a random person(who isnt enrolled) picked from street against 1 unique record,2 unique records, 3 unique records, 4 unique records, 5 unique records,.....1.2 billion unique records is 0.060% .(ofcourse it will be near 0.060% coz it HAS to be close to 0.057% from what I've tried to explain in past 3 posts) It simply means that - 0.06% of 1.2billion = 7,20,000 people will NOT be able to enroll because their biometric matches falsely with some other unique record. So, basically 7.2 lac of 12,000 lac (i.e. 1.2 billion) people will not be able to enroll. This figure would be acceptable to most concerned with UID as well as with the gov.

Uffo! It doesn't mean the rest 1.2Billion - 7.2 Lkh people are immune to FPIR!!!!! FPIR is not a function of "who" registers in the gallery. For each one of those remaining enrolled person there is a chance that 341,666 people who have similar biometrics. Every time you want to verify your biometrics for PDS (ration bank etc.) there can be ~341,666 people who have the similar biometric data as you and people have to sit and manually make ~4.1 × 10^14 comparisons to ensure you are the RIGHT person. Imagine you're in court tomorrow and a biometric fingerprint is found on the scene. Judge - Whose fingerprint is it? Plaintiff Lawyer - We can say that this biometric belongs to 'k' people. Judge - How many are these 'k' people? Plaintiff Lawyer - Approx. 341,666 people all over India Judge - :facepalm: