Rob Coleman OBE BEng
24 September 2008
Dear Mr Coleman
T19669/8 and 10178
The 21 August email was addressed to Mr Rejman-Greene in his capacity as Head of the Home Office Biometrics Centre of Expertise. He has not replied. It would be interesting to know why, on this occasion, someone else has to speak for him.
The suggestion made in the email is that Mr Rejman-Greene explain to the public why facial recognition technology is now being introduced in the UK given that he has previously advised the Home Office that it doesn't work. It would be interesting to know how, on this occasion, he has overcome his earlier reservations. Is there some reason why the public should, after all, be confident that facial recognition technology works reliably?
As you say in your 18 September letter, the poor performance of facial recognition technology was well known in 2003, when Messrs Mansfield and Rejman-Greene published their report to the Home Office, and for years before that. The question is, has the technology improved?
You mention that there has been continued research and development since 2003. I welcome that. You give no reason, however, to believe that there has been any progress. There is no reason to believe that the technology now works any better than it did five years ago. There is no reason to believe, therefore, that it is worth spending money on trialling it again now. It has already been trialled, several times, and failed.
There is an alternative. China is currently conducting a "10 million faces" test as part of its operation Golden Shield. Why not wait until the results of that test are published?
If the technology passes China's test, well and good. The UK could then usefully conduct its own trial to double check. But if, as I suspect it will, the technology proves once again to fail, then the UK could save its money. We do not need to conduct yet another trial when China are already doing it for us and when, as I suspect, Mr Rejman-Greene already knows perfectly well that the technology still doesn't work.
The fact that there is an international standard for passport photographs does not imply that facial recognition technology has improved. It is not clear why you include this fact in your letter on behalf of Mr Rejman-Greene.
You refer to the distinction between one-to-many matches and one-to-one matches. You say that the 2003 report by Messrs Mansfield and Rejman-Greene suggests that one-to-one matching is likely to be considerably more accurate than one-to-many.
Certainly, their report says of one-to-many matching that "facial recognition is not a feasible option" (para.55). Facial recognition technology does not provide the basis for applications like watchlists, according to their report. The National Audit Office made the same point in their February 2007 report on ePassports (para.3.4).
We agree, then, that facial recognition technology is not the answer to one-to-many questions. But do Messrs Mansfield and Rejman-Greene suggest that performance is better in the case of one-to-one matching?
The answer is yes, they do, but they shouldn't. It is illogical of them. At para.58 they say "for one-to-one identity verification, it is only necessary to use a single finger, a single iris or face recognition. The matching error rates achievable are those previously mentioned in Section 3.1". And at Section 3.1 they say "even under relatively good conditions, face recognition fails to approach the required performance" (para.52). Para.52 flatly contradicts para.58.
The 2004 UKPS biometrics enrolment trial results fully support Mr Rejman-Greene's earlier reservations. Facial recognition technology failed 31% of the time with able-bodied participants and 52% of the time with disabled participants (para.18.104.22.168).
Two years later, the House of Commons Science and Technology Committee reported as follows, again confirming Mr Rejman-Greene's views in 2003:
We also note an apparent discrepancy between the advice offered to us during our visit to the United States in March 2006 and the advice subsequently provided to the identity cards programme team. On 6 March 2006, we met informally a group of senior policy advisers from the Department of Homeland Security to discuss the identity cards programme. When questioned about the maturity of biometric technologies, the advisers agreed that currently the technology was probably not as reliable or as accurate as it might need to be for a national identity card scheme (para.81).
They are not alone. The UK agencies who will have to rely on these biometrics share a lack of confidence in them. The Commissioner of the Metropolitan Police does not believe that facial recognition technology is good enough. And according to the Daily Telegraph:
The trials are already being boycotted by members of the Government's new border force because of concerns they might not stop terrorists coming into the UK.
The current trials of biometrics based on facial recognition at Manchester airport and other places need, therefore, to be conducted to the highest standards, if they need to be conducted at all. Confidence needs to be restored. Could you or Mr Rejman-Greene please tell me or the public how the trials are being controlled. Are there independent observers monitoring the process? How long are the trials due to go on? How many people do you intend to go through the trials before declaring them a failure (or a success)? When will the results be published and who by? What acceptance tests have been set in advance?
This business of pre-set acceptance tests is important. The Identity & Passport Service (IPS) told the Science and Technology Committee that they required biometrics based on fingerprints to fail less than 1% of the time (para.18). In the UKPS biometrics enrolment trial, the fingerprint technology failed 19% of the time (para.22.214.171.124). Despite the technology being 19 times worse than the lowest acceptable level, IPS are still proposing to use it. The public wants, needs, deserves and pays for a firmer understanding of arithmetic and logic.
While you emphasise the point quite correctly that this technology is only being trialled in the UK at the moment, politicians are pre-empting the results and saying that the technology will work. The Prime Minister, for example, said on 17 January 2008 that biometrics "will make it possible to securely link an individual to a unique identity".
The public hear these messages from politicians and may take them to be statements of incontestable, scientific fact. It would, to repeat, be appreciated if Mr Rejman-Greene would tell the public that facial recognition technology is still on trial. It may not work. When it comes to facial recognition, the biometrics emperor may turn out to have no clothes. It is currently no more than wishful thinking. People should not allow their hopes to be raised that their security can be improved by the use of facial recognition technology.
The unreliability of the biometrics being endorsed by the Home Office will be revealed sooner or later. When people find that they no longer have the right to work in the UK, according to IPS, there will be fireworks. That situation can be defused now by correcting the false impression that these biometrics are reliable.
I heartily endorse your statement that "in keeping with the spirit and effect of the Freedom of Information Act, all information is assumed to be releasable to the public".