Can very smilar nonmatch invalidate weaker identification?

[minu]

I am curious about a possibility. If a pair of highly similar (namely, no unexplainable disagreement and many agreement in details) fingerprints which are not from the same finger is found, will this bring doubt to or even invalidate previous fingerprint identifications which are less similar (weaker) than this one, or at least affect the way of reporting latent identification? Similar question can be asked for other forensic evidence, like DNA.

If this happens, is this a more serious problem for fingerprint than for DNA (assume they are not identical twin)? Since DNA match is reported as a random match probability, rather than an absolute conclusion like fingerprint, DNA analysts can say that this is just a rare event. But if this happens for fingerprints, how should fingerprint examiners explain? If fingerprint examiners say that this is just a rare event, defense may argue since highly similar different fingerprints can happen, the results of fingerprint examination should not be reported as absolute identification.

This kind of highly similar non-match may never be found before. But it may happen in future.

Assume a latent is found to be a match to the fingerprint of a very rich person. In order to find non-matched fingerprints with higher similarity than the similarity between this latent and his fingerprint, this rich defense hires a group of experts from several areas (LPE, lawyer, computer scientist, statistician, etc). Assume the defense has access to a large fingerprint database. They use a special fingerprint system to match these fingerprints against each other. This system is different from current AFIS in that it modifies a fingerprint (by smudging dissimilar area) to make it more similar to a different fingerprint. In this way, they can find a set of highly similar non-matched fingerprints (and the way to smudge fingerprints). Then the hired LPE will check these fingerprints to find the best pair which is more similar than the latent match in this case. Then they do some smudging to make one of the selected two fingerprints look like a real latent.

Finally, the defense will present these two pairs of fingerprints (highly similar non-matched fingerprints, and the latent and his fingerprint) to examiners hired by the court who are not involved in this case. Since the examiners hired by the defense think non-matched fingerprints are more similar, it is likely that the examiners hired by the court will have the same opinion as well.

If this happens, will this invalidate the identification in this case or make the original LPE (hired by prosecutor) offer a less absolute opinion? What kind of reaction of LPE hired by prosecutor can be accepted by LPEs hired by defense, and court?

Thanks for your reply in advance!

Regards
Jianjiang Feng
MSU

[kevin]

Assume the defense has access to a large fingerprint database. They use a special fingerprint system to match these fingerprints against each other. This system is different from current AFIS in that it modifies a fingerprint (by smudging dissimilar area) to make it more similar to a different fingerprint.

CTS has been doing this for years (albeit through photography, cropping and smudging)....I don't think this makes any difference.

[sharon cook]

[minu]

If two fingerprints are from different people, there will not be "many points of agreement" and no "unexplainable disagreement." ... I assume you are talking about ridge events, Level 2 minutia, Galton points, etc. You are not taking into consideration everything else that goes into fingerprint comparison and conclusion...it is NOT just "points."

I am not talking about good quality rolled fingerprints. I am saying finding two similar fingerprints from different fingers and making one of them a latent print by reducing quality and area.

If you agree that LPE may offer an inconclusive conclusion in comparing a latent to a known print (which are from different sources), you will agree that different fingerprints can be similar when quality becomes poor and common area becomes small.

If you agree that different fingerprints can be similar when quality becomes poor and common area becomes small, you may agree that some different fingerprints may appear more similar than other different fingerprints.

If you agree that natural quality degradation in latent deposition can make different fingerprints appear more similar, you may agree that some intentional way can be even more effective, such as by cutting or smudging disagreement area, blurring Level 3 details.

If the similarity between very similar different fingerprints can be further improved intentionally, why it cannot reach the decision threshold of identification?

I am just talking about the possibility and its possible effect on latent identification.

Regards
Jianjiang Feng
MSU

[SConner]

What your suggesting amounts to evidence fabrication. As long as the original evidentiary image or latent card has been preserved correctly and there is a properly documented chain of custody any alterations (digital or physical) should be detectable.

How would your scenario of "smudging" details in an image of a latent be any different than your hypothetical rich person obtaining a DNA report and deleting or altering a band to skew potential donor results?

[minu]

What your suggesting amounts to evidence fabrication.

Here, the defense agrees that the latent has many agreement and no unexplainable disagreement to his fingerprint. But, he will try to find two another different fingerprints, which have even more agreement and no unexplainable disagreement between them (if natural degradation is performed). Why do you think that this amounts to evidence fabrication?

How would your scenario of "smudging" details in an image of a latent be any different than your hypothetical rich person obtaining a DNA report and deleting or altering a band to skew potential donor results?

The defense is neither altering the latent, nor his fingerprint.

Jianjiang Feng
MSU

[SConner]

Your original scenario of a rich defendant seemed to suggest that they would have access to the latent to search against their own database and once a "highly similar" candidate was found that somehow either the latent or the candidate(s)' prints could be altered to better fit by either "smudging dissimilar areas", cropping, or blurring of the L3 detail.

Alterations such as that would amount to evidence fabrication.

I may have misunderstood the scenario.

[David Fairhurst]

If you agree that different fingerprints can be similar when quality becomes poor and common area becomes small, you may agree that some different fingerprints may appear more similar than other different fingerprints.

If you agree that natural quality degradation in latent deposition can make different fingerprints appear more similar, you may agree that some intentional way can be even more effective, such as by cutting or smudging disagreement area, blurring Level 3 details.

I'm afraid I cannot agree with this.
Degradation of the quality and quantity of detail in the latent does not make it more similar to the exemplar.
It makes it less distinguishable
The commonly used analogy of a blurred photograph of a person's face or of a car fits here too.
It is inability to distinguish the latent from the exemplar that leads to inconclusive evaluations.

Unfortunately this leads me to question your entire premise.
If by 'very similar non-match' you mean an inconclusive result because the latent cannot be distinguished from the exemplar then that cannot be stronger than an identification.
What is a 'weak identification' anyway?
What, conversely, is a 'strong identification'?
Once you have individualised a print, that's it. It's absolute. No varying strength, No different probabilities. 100%.
The same applies if your talking about an exclusion. If you were to find a latent that showed (for the sake of argument only) 8 Galton details that were indistinguishable from the exemplar due to reduced quality and then one difference that enables you to exclude. That's absolute too. "Definitely not his print!"
Take away that one difference by cropping the photo and you're left with an inconclusive evaluation again.
Inprove the quality of the detail (by removing the blurring filters you applied with PhotoShop) and you will reach a point where you are able to ditinguish the differences in third level detail, and the subtle differences in 2LD, and you will once again be able to exclude.

[Gerald Clough]

I am curious about a possibility. If a pair of highly similar (namely, no unexplainable disagreement and many agreement in details) fingerprints which are not from the same finger is found, will this bring doubt to or even invalidate previous fingerprint identifications which are less similar (weaker) than this one, or at least affect the way of reporting latent identification? Similar question can be asked for other forensic evidence, like DNA.

Let's begin with just that. But think about what you're asking to be shown in this challenge. You are, if I read it correctly and can state it more clearly, proposing that they produce another impression that appears to be identifiable to both the case latent AND a different source. That is, in effect, the same as producing another individual to whom the case latent can be identified. That is obviously a contradiction, and it would suggest that the case identification was made from insufficient data. The problem is that if the case identification was made to a generally accepted standard (admittedly undefined), the event you describe would indeed be something of a revelation. While the standard is undefined, that it recognized, and it is set arbitrarily high to prevent error. The fact that what it described above hasn't happened suggests that it is effectively high enough to accomplish this. This suggests, too, that were the actual threshold for identification known, additional identifications might be possible, but the conservative view is taken for safety.

The extension of your hypothetical has them obscuring dissimilarities to produce a latent that could have been found if those areas were not impressed. I understand what you're saying. But, there are two kinds of dissimilarities they could be obscuring. One kind would be a dissimilarity lying outside the impression area of the case latent. That situation reverts to the argument above, in that the identification would not have been made made under the generally accepted standard. The other kind of dissimilarity would be one lying within the impression area of the case latent. There are two possibilities for such a dissimilarity. It either occurs in an area in which the case latent has no such obscured area, or it occurs in an area in which the case latent also has an obscured area. But in both those cases, the original identification was made with sufficient data to be valid, and, again, you would be proposing that the generally accepted standard was insufficiently reliable.

So you are proposing that a case actually be found that proves the existence of two portions of skin that cannot be distinguished according to the professional standard practice when all observations are accurately made. And keep in mind that, while such a discovery might significantly affect that case, it can be characterized as being no significant discovery in terms of invalidating the general practice. It is logically true that any array of a large but finite number of accidental features always presents some very small possibility of duplication, and if one could construct a mechanism to generate fingerprints, it would, if allowed to run without limit, eventually produce a duplicate pair. It is not different from DNA, except that, for DNA, there is a known statistical probability of duplication. DNA conclusions are, in fact, qualified by such a probability. The probability for fingerprints is unknown, largely because it is vastly more difficult to code fingerprint features and their complex relationships in such a way that account for things like distortions. But, the standard of practice is set sufficiently high that the conclusion of identification is made to a very high degree of statistical certainty, much higher than most other expert conclusions offered.

Because you propose obscuring dissimilarities, you may have wondered if there is not a fatal flaw in fingerprint identification that is revealed when you ask, "But, since you say a clear contradiction precludes identification, how do you know there is not a dissimilarity that would reveal a different source, if only that portion of skin had been clearly impressed?" This is, again, presuming that the standard for identification is generally flawed, and vast experience and a good deal of searching for that situation suggests that, if what can be seen is sufficient for identification, the possibility that an alternate source with the dissimilarity will be found is incredibly small. The likelihood of finding what you propose would remain infinitesimally low, even if someone actually found one. Finding one in no way affects the likelihood.

No matter how much money was provided, only a scientific charlatan would pretend to Defendant Moneybags that there was any reason to think the search would be successful. They might, in fact, merely reinforce the implications of experience and study that the probability of error was fabulously low. About like hiring an expert to say that, with a different human population, the DNA identification was only one in 150,000,000, not one in 175,000,000. Hardly a win for the defense. And consider that I know of no case in which fingerprints or DNA were presented with no other evidence at all. In other words, no showing that the defendant could have committed the crime. We are not concerned that, should the human race go on indefinitely, someone else might have the same DNA in the limited portion used in criminal forensics, and we're not much more concerned even if some other living person is a match. Even when we convict one person solely upon the fact of a fingerprint identification and the fact that it was indeed reasonably possible for them to have made the latent impression, it is a conviction with vastly more confidence than many, many others obtained without such extreme high probability evidence.

But, to honor your hypothetical with an answer to how the examiner would respond if such an extremely unlikely thing were to happen as finding two actual impressions, I think the response could be that they have simply found an actual event at the far end of the probability curve and that it is that incredibly rare case, never before seen, in which the two cannot be distinguished. The prosecution would then offer the finder of fact choose between the two possible sources on the basis of what else is presented as evidence.

[minu]

I just want to come up with a story to start the debate about the policy to report latent examination conclusion. I agree wrong identification is extremely low when a high standard is set and the current standard(s) is very high. But the other side of the coin is, under a very high standard, how many valuable latent evidence is classified as inconclusive? Should the same high standard be used for any latent evidence, no matter if there exists other evidences that also provide some clues for the suspect?

My opinion is the current 3-category policy is not perfect. The exclusion is a kind of absolute conclusion. So I just question the other two conclusions: inconclusive and identification.

Degradation of the quality and quantity of detail in the latent does not make it more similar to the exemplar.
It makes it less distinguishable
The commonly used analogy of a blurred photograph of a person's face or of a car fits here too.

Thanks. Distinguishable is a better term than similar. I meant the same thing by using "similar" as you.

It is inability to distinguish the latent from the exemplar that leads to inconclusive evaluations.

Assume that nobody has really seen the suspect pressed the latent. If the latent and the exemplar cannot be distinguished, I feel that a more scientific conclusion is a fail-to-exclude (namely, this suspect may produce this latent) and the probability that a random finger can produce this latent should be calculated. This is the practice of DNA typing. I know it is very difficult to shift fingerprint examination to this style.

Unfortunately this leads me to question your entire premise.
If by 'very similar non-match' you mean an inconclusive result because the latent cannot be distinguished from the exemplar then that cannot be stronger than an identification.

Inconclusive result and identification are the same in the sense of they are both fail-to-exclude. The difference between them is that the random match probability of inconclusive result is higher than that of identification. In other words, inconclusive result means the agreement between prints is less rare than that in identification. In other words, inconclusive result means less details/relationship in agreement than that in identification.

You can use a decision threshold to categorize a fail-to-exclude as inconclusive or identification. But, as long as there is a decision threshold, there must be some inconclusive, which are very close to the decision threshold. Now I want to ask, for such inconclusive very close to the decision threshold, if there exists other evidences suggest the same subject, do you want to change your conclusion to identification? Is LPE permitted to draw conclusion of fingerprint examination by combining with other evidences? If this is not permitted, valuable latent evidence may be wasted.

What is a 'weak identification' anyway?
What, conversely, is a 'strong identification'?
Once you have individualised a print, that's it. It's absolute. No varying strength, No different probabilities. 100%.

If the RMP of an identification is higher than another identification, or the agreement is less rare, or fewer details/relationship in agreement, it is weaker.

The same applies if your talking about an exclusion. If you were to find a latent that showed (for the sake of argument only) 8 Galton details that were indistinguishable from the exemplar due to reduced quality and then one difference that enables you to exclude. That's absolute too. "Definitely not his print!"

I agree with you on exclusion. Exclusion is absolute. But identification is not absolute in theory (statistics). The current policy of presenting highly rare agreement as absolute identification is fully understandable, since it is easier for human beings to find consensus on category (sufficient or not sufficient), rather than on a quantitative measure (random match probability).

Take away that one difference by cropping the photo and you're left with an inconclusive evaluation again.

This is true for average case (or most cases). But I am saying the two fingerprints, which are most difficult to distinguish from each other if one of them is intentionally modified in a natural way, in a large fingerprint database (say the FBI's IAFIS).

Inprove the quality of the detail (by removing the blurring filters you applied with PhotoShop) and you will reach a point where you are able to ditinguish the differences in third level detail, and the subtle differences in 2LD, and you will once again be able to exclude.

I use 'blurring' in a general sense. You can think of it as any degradation. Not all degradation is revertible in current technology.

Thanks for your reply!

Jianjiang Feng
MSU

[kevin]

I would say the three category rule of ident, exclusion, and inconclusive are necessary to insure the integrity of the science. Any form of 'sliding scale' in the form of possible's and/or weak identifications leaves a massive loophole for the justice system to run through-for both the defense and prosecution. Each latent examination should stand on its own merit. The logic and science behind a sliding scale or alternative conclusions may be valid, but it doesn't make it a good idea to implement. A large part of the problem with the justice system today in dealing with latent print evidence is discussions of methodology and transparency often become an excercise in semantics. Add to that definitions that are open to wider interpretation and we are doing the defendants, jury, and justice system a disservice, even if additional information that can be gleened from the analysis is lost in the process. To do otherwise would only serve to confuse the issue and we already have lawyers to do that for us....

[Gerald Clough]

I agree with you on exclusion. Exclusion is absolute. But identification is not absolute in theory (statistics). The current policy of presenting highly rare agreement as absolute identification is fully understandable, since it is easier for human beings to find consensus on category (sufficient or not sufficient), rather than on a quantitative measure (random match probability).

This is an area that is the subject of much discussion and study. There are a lot of factors involved. One not inconsiderable one is historical. Another is the the difficulty of developing both reasonable ways to code fingerprint features and likelihood numbers that both take into account what examiners use and thresholds that are practically applicable. And I think there is also considerable friction between the ends and modes of science and the ends and modes of criminal justice. The essence of science is formal process and very high standards, as well as an assumption that all conclusions are tentative. Criminal justice seeks a fair process to the end of making the best decision possible, given what is known and can be done right now. Much is admitted in litigation that is not science and not scientifically validated. Psychology is probably the most common example. No well-stated theory of the physical world, and so, therefore, no validation possible. And that means that far more often than not, expert conclusions are diametrically opposed. Not wonderful, but the answers to vital questions cannot be sought in any other way. At the other end is DNA. Well-defined in physical terms. Highly numerical. Limited, well-defined data set. Elaborately validated, albeit with some conditioning on populations, but sufficient that various results all present extremely low likelihoods.

It is no great disability that fingerprint conclusions are not now subject to the same quality of validation. While an accumulation of experience and decades of proven utility are not valid scientific proofs, they are more than adequate for judging the general reliability of fingerprint examination in the arena in which it is applied. Common sense and reasoning, which would horrify a scientists being presented with a proof based upon experience, are not only accepted in litigation, but are commonly urged. Note that the conclusions of not-validatable expert fields are also presented as "absolute." One psychiatrist may testify that the person suffers schizophrenia. Another declares that the same person actually has a different disorder. One offers an absolute opinion that the person is mentally retarded. The other declares that the person is merely below average and not retarded. Each is confident, but both will readily admit that there is some possibility of error. DNA conclusions are stated exclusively as "not excluded" with a statement of probability of identity. The point is that "absolute" means one thing to the scientist, who generally does not admit to absolute knowledge. An "absolute" conclusion in law means something else, and it is well-understood and accepted. Even the scientist assumes certain observations are "absolute," in order to construct a conclusion that, in the grand scheme, is tentative. Criminal justice does not have the luxury of holding questions open forever, much as the engineer does not have the luxury of putting all projects on hold in hopes of one miraculous day applying ultimate truth. The engineer and the court apply what works. The conclusion of absolute identity is the best and, in its application, the most appropriate statement of conclusion, given the state of knowledge and practice today, which is all you can say about any expert conclusion.

[minu]

Criminal justice does not have the luxury of holding questions open forever, much as the engineer does not have the luxury of putting all projects on hold in hopes of one miraculous day applying ultimate truth. The engineer and the court apply what works.

The conclusion of absolute identity is the best and, in its application, the most appropriate statement of conclusion, given the state of knowledge and practice today, which is all you can say about any expert conclusion.

Really appreciate! Your explanation makes me believe that it is not necessary for fingerprint to follow the style of DNA conclusion to get permitted in courts.

Then I don't understand why there seems to be huge efforts for transparency, standard, documentation, statistics research. I know some effort is necessary to control quality. But it seems the effort is huge. Should every LPE change his/her habit to do very detailed documentation for the purpose of transparency? Or for eliminating any possible erroneous identification? When the error rate is sufficiently low, buying insurance may be more economical. Let insurance company pay Mayfield, instead of FBI. I don't know if insurance companies are interested in this business. But if they have this service, I guess the premium should not be high (at least for lab of good reputation), considering the extremely low likelihood of erroneous identification.

Jianjiang Feng
MSU

[David Fairhurst]

Thanks for the reply minu. I now understand where you are coming from.

The human brain is a fantastic organ.
Give a maths professor a load of data on the initial velocity of a rotating spherical object, the average wind speed, the drag coefficient of the surface of the sphere as it passes through the air, the curvature of the earth beneath it, and its accelleration due to gravity and he might take half an hour to work out where and when that sphere might come to rest.

Now put a nine year old in the middle of a baseball field and hit him a fly-ball. Watch him catch it. He has instinctively done exactly the same mathematics as the maths professor in a fraction of the time and has placed his glove in the right place at the right time to catch the ball. He has also worked the speed and direction he needs to travel to put his glove there and set off running too.

As latent print examiners we are the Major League catchers.

The maths that we instinctively deploy has not yet been fully worked out; research into the data needed to calculate even vaguely accurate RMPs is in its infancy. When we individualise a latent print we are efectively declaring that the RMP is so close to zero that it makes no practical difference. We can't give you an exact, or even an approximate figure because we're doing it instictively rather than on a calculator, but do we really need to give a figure if it's that close to zero that the jury can't comprehend the difference? Isn't that juries do with incredibly small DNA RMPs. They hear "1 in 1 billion" and they'll think "well that's so close to zero that i'll just treat it as zero because it makes no practical difference." In DNA the maths and the data are there to produce the numbers and, rightly, allow the jury to make that decision (some have called it a "leap of faith", others have said "it's no great leap") In latent prints the data and maths are not there. We have to help the jury and make the leap for them. That is the role of an expert witness.

I'd back a professional Baseball team against a load of robots every time, but as a word of caution "Even the best catcher occasionally drops a sitter!"

[Gerald Clough]

I'd back a professional Baseball team against a load of robots every time, but as a word of caution "Even the best catcher occasionally drops a sitter!"

And even if a fully developed scheme that accounted for all that examiners can legitimately observe and analyze were to allow validation that was directly applicable to practical work, that would still be true. The data in fingerprint examination must always be an examiner's interpretation, since we don't have the more or less absolute values of DNA. So, there is always the possibility of frank error or the possibility of argument about the interpretation. They're not ever going to be very frequent, since most of what we work with can be characterized in a pretty straightforward way. But it can happen, and in a some well-known error cases, it should have been argued at very early stages. I view the nature of the prominent error cases to really point up that the errors are really quite easily detected and that the weaknesses in the system are organizational and procedural, rather than issues of technique or general technical competence.

Feng: Be careful what you wish for, and consider if involving insurers in medical decision making has improved patient care.