Differential Randomness And Individualization
_________________________________________________
The reason that the science of fingerprint identification is having difficulty articulating the exact Daubert requirements is not an issue of science, it is an issue of complexity. The opponents of fingerprint identification and many other forensic sciences ask for exact statistics and rates for various processes and methodologies. However, without a specific set of numbers or limits on the information at issue, it becomes impossible to define such statistics accurately and meaningfully. In some cases, processes have definite boundaries. DNA for instance, has known numbers to draw statistical information from. However, in cases where the quantity and qualitative values vary we are unable to assign sufficiently accurate probabilities. These types of overly complex statistical studies must be accomplished in a different manner. In the normal course of statistical mechanics one can reasonably define the parameters of a problem, yet with complex variables, statistical estimates must be made of all the parameters. In turn, the resulting solution is also an estimate. “The model must, by definition, be designed on a subset of features.”[1] The reality is that probability statistics break down simply due to the extreme complexity of the issue which is illustrated by the inability to assign limits or boundaries to the information available for calculation . This is a very common feature of nature in both biological and non-biological structures. The only real limitation is on our lack of understanding of the complexities of nature and of our inability to accept something that is impossible to define in absolute terms. In the mistaken need for absolute definitions, the error is ours.
It is short-sighted to think that complexity and our lack of ability to define such events should prevent such processes from being used in an accurate manner for individualization. In most cases such comparative analysis utilize more information than would DNA, thus making the value of such analysis more accurate even though it is by nature statistically undefinable due to complexity. Identical twins are a prime example of this. Fingerprint identification can make the distinction of individualization between identical twins, even when only a very small section of friction skin in compared. This is an advantage over DNA and also with the previous anthropometry system with its loose tolerances.
We must remember that information is embedded into things at various levels in relative ways. To focus on a single aspect is to devalue the rest of the information. However, once individualization is hypothesized, the rest of the information is generally regulated to a supporting state, rather than used for further detailed comparative use. In simple terms, we don’t need to (also) look at the back of a person to finish our recognition of them. Nor do we need to look at all the areas of friction skin to verify a single fingerprint, or for that matter, review all the friction skin in the world! “There comes a point when there is no perceived need to challenge the principle anymore. The hypothesis is then declared demonstrated and perceived as a certainty. The process itself is highly scientific and should not be viewed as an expression of faith...”[1] The next step in the methodology is verification of that hypothesis. With fingerprint identification, the original comparison is a provisional identification and only becomes complete and valid when the comparison is verified by another qualified examiner.
Accordingly, certain questions should be asked of the process; Are proper protocols and methodology being followed to ensure accurate results? Are the examiners properly trained to analyze the information? It is here that most errors are found within the sciences.
Within the development of friction ridge skin during the early fetal stages, randomness prevails. Dr. Babler’s embryology research has brought to light new information on the development of friction skin. This includes information on both spatial and chronological events involving friction skin development. While the genetic material organizes different cells to accomplish different tasks, the actual growth process of cell division results in randomly generated information throughout the friction skin.
Random: Having no specific pattern or purpose; selected in a way that each member of a set has an equal chance of being chosen. On a larger scale, randomness is in itself statistically unique. This is a key point. “Uniqueness is information that allows for a relative distinction or possibly an individualization.”[2] “It doesn’t even make sense to conduct statistical studies on a pattern which is biologically unique, such as friction ridge skin! Why bring statistical concepts into the debate when in reality, the chance of finding another identical object is by definition zero?”[1] Again, uniqueness is an overly complex subject that cannot be quantitatively contained in a neat package for a simplified discussion by mortal man in the courts. The underlying randomness, hence uniqueness, of such subjects cannot be accurately defined mathematically for probability purposes. This is especially true in variable information models such as fingerprint identification, shoe impressions, facial recognition, etc.
What can be said, is that the concepts are based on examination of both differential information and matching information drawn from random processes at a more fundamental level. A dictionary definition of differential is: The degree or amount to which similar things differ. Showing a difference. In comparative analysis the difference is just as important as that which is the same. In effect, likeness is relative to any differences and visa versa. In fact, it is the infinite differences which make individualization possible. Individualization requires that all else be different. Due to nature’s built-in randomness, thus uniqueness, we understand that matching randomness is not only uncommon, with sufficient randomness it is, for all practical purposes, impossible! Differential randomness, or the degree of difference in items of chance, is what nature is best at. This is the construction of all tangible things. In reality there is no difference between true randomness and differential randomness, the point here is to illustrate our difficulty in comprehending such matters. Again, the link between the proper analysis of the information and understanding the value of that information is quality training and methodology. Proper training and methodology is essential to all aspects of a science. Latent print examiners are no exception.
The science of fingerprint identification itself is sound. Fingerprint identification is validated on a daily basis as millions of comparisons are made by software programs specifically designed to search for matching fingerprints. Here the statistics are very clear. In the many billions of searches thus far, zero fingerprint impressions from different persons have been found to be the same. Accordingly, the differences form a demonstrated basis for individualization. The infrastructure for such comparisons by human examiners relies on sufficient and standardized training as the examiner must consider much more information than does a computer. It is imperative for the examiner to properly evaluate the information.
In a related topic, consider the concept of recognition. Recognition as it relates to identification and individualization is; To acknowledge the validity or reality of. At what point can a particular item be recognized? What does it take to acknowledge the validity or reality of something? At what point does a comparative analysis provide sufficient information for an individualization? The concept of recognition is based on the organization of unique information. It is the uniqueness of the information that allows for recognition. Of course, this is also related to randomness. Yet, just how can a concept such as recognition be defined? It is replete with complexity. Again, we are assigned only to describe such matters within a broad foundation of inelegant statistical mechanics.
No matter how much we would like to define all aspects of nature for the sake of abbreviated legal discussion, complexity continues to keep us at a distance. This is especially true in the variable informational world of forensic science. Accordingly, aspects such as error rates, probability, and testing must first be carefully defined for the specific issue which is to be addressed, only then can progress be made and our understanding expanded. When one does not know which way to talk, he will invariably talk in circles. We must understand that probability theory, does in fact, deal with randomness and estimation. The point is, that we cannot expect exactness in the statistical evaluation of fingerprint identification, nor is exactness necessary for science. “The sciences do not try to explain, they hardly even try to interpret, they mainly make models. By which, with the addition of certain verbal interpretations, describes observed phenomena. The justification of such a mathematical construct is solely and precisely that it is expected to work.” (John Von Neumann) [3] Accordingly, our goal is to improve our fingerprint identification model, not justify it.
Craig A. Coppock
1. Champod, Christophe: Views on statistics and Probabilities in latent fingerprint comparison. The Detail 1-7-02
2. Coppock, Craig: Minimum Information And Fingerprint Identification. The Detail 12-2003
3. Gleick, James (1987) Chaos: Making A New Science. Penguin Books, New York
No comments:
Post a Comment