[MUSIC] Welcome to this third video, where we will discuss the case of the wrong identification against Brandon Mayfield. Let me give you some elements of the context and Christophe will develop further the issue relation to the fingerprint evidence. By the way, the whole ordeal that the Mayfield family has went through, is described in the book written by Brandon Mayfield and his daughter Sharia, age of 12 at the time. The case start with the Madrid bombing incident on March 11, 2004. Four commuter train were bombed in a coordinated attack that led to the death of nearly 200 person and many hundred wounded. The Spanish police discovered an abandoned utility car. A Reno Canggu. In the car, a blue plastic bag, like this one, containing seven detonators, traces of explosive and a cassette tape with verses of the Quran. The bag was taken to the laboratory for examination. Using superglue fuming followed by dye staining, a series of papillary marks were detected. We have replicated the process for you using a blue plastic bag and the fuming cabinet of this lab. These marks were searched without success against the Spanish fingerprint collection using an automatic system called AFIS that we will discuss later. They then searched these marks internationally through the special Interpol channel. On March 13, 2004, the marks were transmitted to the FBI and searched against their database. At the time, 47 million candidate for one of the Marks called LFP17 there was a candidate in the AFIS search list. Candidate number four was therefore checked by the FBI Latin Print fingerprint expert and identified to Brandon Mayfield. This identification was subsequently verified by two other examiner of the Latin Print unit of the FBI. Brandon Mayfield had his fingerprint in the FBI fingerprint repository because he served in the US Army for ten years. He was a lawyer running an office in Portland, Oregon with his wife Mona of Egyptian origin. Brandon Mayfield was converted to Islam, working mainly on domestic relation suits. The experts concluded with an excess of 15 matching characteristics between the mark and Mayfield's print. Experts declared that they were absolutely confident, Mayfield had left this mark. The official documents state that the identification was a 100% match. The Mayfield family were undercover Soviets. On that basis, on May 6, 2004, the FBI issues a warrant to detain Brandon may fit as a material witness. That is, a person alleged to have crucial information concerning their criminal proceedings. Since September 11th, the statue has been used to detain suspect without charge. He was arrested and locked in a security prison on suspicion of terrorism. A search warrant was executed at his office and at his home. During these searches, documents related to Spain were discovered in the family computer. He had also listed his services as a lawyer in an Islamic directory. He was then presented to a judge in court. The judge asked for a court appointed expert to review the fingerprint evidence. The expert confirmed the identification made by the FBI. So how the problem, if any, was detected. >> Well, the first explanations that came to the mind of Brandon Mayfield was the first a typo in the text reading 100% match. Was it 1% or 10%? The second explanation was a forgery. The error was luckily discovered by the Spanish National Police when they arrested a Spanish Algerian individual named Dawud Hunani. When his prints were compared to disputed marks and others, they identified him and notified the FBI. That was the 19th of May, 2004, the same day, the court appointed expert confirmed the FBI initial identification. >> Talk about coincidence, as soon as the print from Dawud where checked by the FBI, their examiner concur with their identification and recognized their mistake. This case is the first known case of error by the FBI in this area of expertise. On the May 24th, 2004 the FBI apologized to Mr. Mayfield and his family and announced a full investigation of the case. Brandon Mayfield was freed on the same day. He later obtained $2 million in compensation. The FBI conducted an internal review, a review with an international panel of experts. In addition the office of the Inspector General carried out a full review. All documents have been publically released. But know that in this case the fingerprint of the now no donor were available. It's always easier to get clearer if there is a better suspect. But Christophe, can you please tell us more on the cause of the error? >> Sure, Franco. Here I will use the report of the Office of Inspector General, OIG. First, let us have a look at the mark. It has been detected on the blue plastic bag, similar to this one. The mark wrongly associated with Brandon Mayfield is the mark called LFP 17. And the mark is imaged in fluorescent mode, here. You can now see the mark in black and white. At the FBI, the mark has been encoded. And searched on the integrated AFIS system with seven minutia. The minutia are the positions where ridges stop and bifurcate. Four ridge endings and three bifurcations where noted by the FBI on the latent, before it is launched on the AFIS to be searched. You may not have heard what an AFIS system is. Let us see in video with my colleague Marco Dedono how it effectively works. We have precisely used the Madrid Mark LFP 17 for this demonstration. We are here in Front of our AFIS system. And I'd like to present to you Marco. Marco De Donno is our AFIS specialist. And AFIS is an acronym for automatic fingerprint recognition systems. And in the context of the case of interest, I will show you how it is used and why and how it had been used in the case at hand. It all started by the detection on the plastic bag of that mark. It's a difficult mark. As you can see, it's very blurred. It's enlarged here. That mark at the time of the initial stages of the investigation hadn't found any compatible person to be at its source. We had no suspect and the mark wasn't sent, and searched on the AFIS system. The mechanism to do the search is to encode this mark and search it automatically on the collection of potential prints of individuals, and these prints take the form of cards like this one where you will have the rolled impressions of all ten fingers, and the flat impression of each finger on the one side. And on the other side, you will have a palm impression of the individual and that constitutes a ten-print card of the individual. Now an automatic fingerprint recognition system will allow you to search the mark like the mark of interest against a very large set of prints constituted in a database. The system we will use for this demonstration is not an operational system. It's a system which is used for research, but it has all the component of an operational system. It is constituted by a database of roughly a million fingerprints. So we will make a search of that mark against a million potential fingerprints, roughly 10, 100,000 individuals to see how you can search a needle in a haystack. So when the expert came to the print in the fourth position on the hit list of 20 candidates, the latent print expert of the FBI took the decision to identify. The 15 features used by the FBI are shown here. One of the first cause of the error is associated with the verification process. When the first examiner's decision was checked by the second and then by the newly chief, the whole new, the initial decision of that colleague. Independence between experts was not guaranteed and that applied also to the court appointed expert who knew that the FBI had previously identified that mark. The second issue is the use of a AFIS database. When you use a large AFIS database, the IAFIS system of the FBI counted every time 47 million 10 print cards. Research algorithm is trying to find the best match according to the initial minutiae mark by the expert. The print from Mayfield is close in terms of its feature to the mark. In fact, the print of Mayfield is a close neighbor in terms of these features to the print of a true donor, Daoud. Note that the print of Daoud was not in the IAFS system, only Mayfield's was. Had Daoud been in the system, he may have been proposed in the hit list and seen by the examiner and Maco will show you in the next video what will be the result of doing the search today with both Mayfield's prints and that with printing the system. The first thing that you need to know is how this mark here ended up on the desk of the FBI to be searched. What happened is that the Spanish authorities looked for the mark in their own AFIS system without success, initially. And because of the importance of the case, it was decided that the mark will be sent to all national bodies affiliated with Interpol through a secure channel of Interpol. So that mark went through the Interpol channel and was distributed to all countries affiliated with Interpol. The FBI being one of the authorities receiving these demands. So the mark arrived on the desk of an FBI examiner and the examiner had the duty to search it against the large IAFIS database, which is the FBI database. The FBI database is absolutely huge in terms of number of individuals. It counts today roughly 100 million 10 print forms. So we are talking about looking for a mark among 100 million individuals, which will be each of them represented by a form like this encoded in the system. So with Maco will show you how the mark is encoded. Of course, this is not exactly how it was processed at the time. But it will show you how an AFIS system operates in very closed environment as to operation. The first thing which is done is the mark is code in the system like this. The latest generation of algorithms allows you to encode the mark automatically and the green little dotty indications represent the minutiae that has been automatically detected by the algorithm. The minutiae are the endings and bifurcation each little papillary ridge will form on the finger. These are the features, which encoded with selectivity give the capability to search in a large scale database. So the search will be focused on these green little dots which each of them represent the minutiae and the little rod give an indication of a direction of that minutiae, of a direction of the ridge bearing the point. Now during the encoding, there is what the algorithm can do, but there is also a lot of expert input who could remove minutiae like Maco is doing at the moment. Judging that these minutiaes maybe misrepresent the mark of interest. The experts may add minutiae by observing that on this image, there's something that was missed by the algorithm and he will assign a minutia in addition and that's done by one awkward moment and he may decide in this case. In fact, that was important in the Mayfield case. He may decide that some section of the mark, typically the top side is or is not related to the bottom side. If Maco decides that he will search only the bottom part It will remove the minutia from the top, and alternatively, it may run another search by just looking at the minutia at the top, and removing the minutia at the bottom. And here, the mark is so complicated in terms of how it was laid down and the difficulties to read the mark will engage the examiner to have multiple encoding and multiple searches on the system find a potential correspondent. So we will leave it here and run it through the system to be searched. So we're about to launch the mark to be searched against the database. As you can see, these are the sets of minutia that Marco annotated to be searched against the whole data set. The whole data set, because the mark itself does not allow to say whether it's a thumb or it's an index or a median. So all the fingers both from the left and the right hand are highlighted. It means that it will be searched against the whole data base here of a million fingerprints. The search is lounged and we will indicate in the system the number of candidates we would like to have back from the search. By candidates, we mean that the system will look in the data set and rank them according to a level of proximity to this mark. The first candidate will be the candidate in which the algorithm will say, these are the closest, this is the closest print. The second will be the second closest and so on. And we can specify how many of them. So we'll indicate that we'd like to have 20 candidate out of a search. And now, the search can be launched. So when a new job is launched against the database, it will appear on our job list very soon. And here this is the case we are currently searching, the state indicate searching latent. The search time is very quick. It depends, of course, on the system. It depends on the size of a database. But here, it's already finished. Now, the stage we are here is verification stage. It's going through the candidates that have been responded back by the system. So we are now in the verification stage of the search. As you can see, the system is returning the mark. And against in comparison, the potential print of the individual and these individual are ranked in a list as you can see down here from number 1 to 20 down there. And here, we are on the individual number three and the expert duty at this point is to access based on the closeness provided back by the algorithm whether or not that print correspond to that mark and whether they have been left by the same person. And in this case at hand, you can see the minutiae that has been detected by the algorithm on the print. It has found some relationship potentially in this area with this area with the mark. However, and it's more easily viewed for an expert, the center of the print is rather different between this mark which is more close and looks like an arch and this print which is round and look like well. So because of that difference, the expert will say, no, this is not a identification and that will be indicated in the system as a non-ident and he will go through each candidate. And here, we'll move up to candidate number one. Candidate number one has a genera pattern which corresponds much more closely and we may want to make a close up on this, and the expert will go through each minutiae, and then each papillary lines, and check whether they fit, and fall in accordance with the information on the print. And for example, the set off minutiae up here, do we find in close correspondence with the set of minutia over here? Do this minutiae here find correspondence with the initial minutia surged against it here and the same for that one? If the expert is satisfied that there is congruence of the information and there is no significant dissimilarity, then he will decide that this is an identification and that will end the search in the AFIS system. That case will then followup to a second expert for this work to be verified. So I'd like to show you one of the key difficulty in this comparison and how expert interact with the system. Here, this is the search that we carried on all of the minutiae that have been indicated including the top part and the middle part. All have been searched and the first candidate that came back from the search is candidate number one. And now, you know that his name Daoud. That's the candidate that Marco identified previously and that candidate was identified. Now, what happened in the search by the FBI is that they haven't searched all the minutia. They made a choice initially, judging that this lack of features in between suggested that maybe there is two oppositions. They decided to search only the core parts of minutiae against the database ignoring the contribution of these features assuming that these features were coming from a double touch or from another finger. So when we do this on the system and we search only the minutiae as they had been searched by FBI, these are the yellow minutia here, searched against the database. The first candidate that comes back in position one is the print corresponding to Mayfield. Now if we go down in position four, we found the print from Daoud. Now remember that at the time of the misidentification by the FBI, Daoud was not in the database. So the FBI had Mayfield in position four or five and we can go back on Mayfield again and the expert decided, but based on this information. And indeed, not allowing for the contribution of the top part that just based on what was visible in the central part decided that he will identified navy and that was the start of the process which went through then subsequent verification off of Brandon Mayfield. I show you here how close these two prints are as underlined in the ORJ report. All features used for the initial search are shared between the two respectively on the print of Mayfield and on the print of Daoud, then the OIJ indicated that the FBI expert were quick to explain away differences invoking reasons, such as multiple touches, rankage of a plastic that induced distortion. >> Here Christophe, as previously in this course, we see that after the findings, it's quite easy to explain away the differences. So it's important to think of what we expect before we proceed with the comparison. >> And yes, you're right, Franco. And indeed, the expert designated corresponding features between remark and the print, but were not documented in the analysis phase at least as we can judge from the initial coding. The printed self was tainting the judgement on the mark or form of circular reasoning. Finally, the OIG indicated that the FBI relied too heavily on third level details. The shape of the pores, the shape of the ridge edges seen in comparison between the mark and the print. But again, these features where noted only once the print was made available and not indicated before in the analysis phase when observing uneven mark. >> Okay, Christophe, thanks. So we can advice on basic here. Check, when assuming fingerprint comparison. First, ask for the commendation of the feature seen on the mark before any comparison. This analysis phase is critical, because the touch from the print. Otherwise, there is no guarantee that all feature of being seen only during the comparison tainted by the knowledge of the print. Two, ask for a full independent blind verification and the expert should be able to produce separate and the independent notes of their examination. Third, be aware that when used databases are searched, taking advantage of highly efficient matching algorithms, you can expect close non-matching prints to be returned. >> Thank you, Franco. Before we close this case, it is worth asking fingerprint experts for more humble conclusions. When the FBI notified the Spanish national police, the Spanish authorities replied on April 13th that they were not convinced of the identification. They reported back to the FBI a negative outcome, but so convinced of their 100% certainty. The FBI tried to persuade the Spanish authority of their identification during a visit in April in Madrid at no point have they ever doubted of that fact until that was nominated. Adopting a more humble reporting process, acknowledging the possibility of error is the obvious way forward. At the end of day, it is all about probabilities and research have shown that the value of the evidence in terms of likelihood ratio is high with finger print evidence. It is increasing as the number of minutiae accumulating comparison between the mark and the print. And here, I showed you one of the graphs of the latest publication by Neumann and colleague that showed that as the number of minutiae increase here from 3 to 15, the strength of the evidence will move up. But and generally speaking, you will have a Likert ratio to 10 to the power of 7 with 7 minutiae and 10 to the power of 8 with 8 minutiae. But of course, that value will depend on the type of minutiae and their location on the print. The value for a given number of minutiae may change slightly significantly, depending on nature of minutiae involved. After the evidence, the FBI latent prints unit made extensive changes in its procedure to prevent such event from reoccurring. It entails a separation of the analysis phase and the comparison phase with full documentation. The adoption of blind verification procedures in cases and clear dispute resolution procedures. And above all, a huge program of research and development has been put together and that led to very high quality published paper that have been produced by the FBI recently. The OIG kept watching brief on the FBI and reported about their achievements in 2011. >> You may wonder what about the so-called evidence found during the search at this home? Well, the document in Spanish were associated with some homework of her daughter. To include this case, we invite you to listen to the interview of Brandon and Sharia Mayfield. They will give you the perspective of the victim. If another religion, Muslim had a role to play and our blinding from forensic on relevant case information may help to avoid bias in forensic science. Thank you, Christophe. And thank you, you all for watching. [MUSIC]