FACIAL RECOGNITION AND IMAGE COMPARISON EVIDENCE: IDENTIFICATION BY INVESTIGATORS, FAMILIARS, EXPERTS, SUPER-RECOGNISERS AND ALGORITHMS.

• Date: 01 August 2021
• Author: Edmond, Gary

CONTENTS I Improving Legal Practice with Scientific Research II Legal Interpretations of Images and Experts A Smith, Morgan, Honeysett and the Limits of s 79 Jurisprudence B Sections 78 and 137 III Unfamiliar Face Matching: Relevant Scientific Research A Factors That Affect Accuracy in Face Comparison 1 Familiarity with the Face 2 Image Quality and Quantity 3 Demographics of the Face and Decision-Makers B Evaluating Performance of the Different 'Groups' 1 Lay Persons (including Investigators, Reviewers and Other Purported Experts) 2 Facial Examiners--(Some) Genuine Expertise in Image Comparison 3 Super-Recognisers--Ability without Knowledge, Training, Study or Experience 4 Algorithms 5 Hybrid Systems (and Meta-Analysts) IV Discussion V Conclusion I IMPROVING LEGAL PRACTICE WITH SCIENTIFIC RESEARCH

This article is about the use of images in criminal proceedings. (1) It considers how courts should engage with the interpretation of images to assist with identification and, inexorably, the role of scientific research in the admission, presentation and evaluation of this evidence. The article provides an overview of scientific research on the use of images for the purpose of identifying persons of interest ('POIs'). This follows a review of the piecemeal Australian jurisprudence pertaining to the admission and interpretation of image evidence. (2) Juxtaposed, these two reviews palpably demonstrate the limits of legal concepts and prevailing practice. It is our contention that recourse to scientific knowledge can help to ensure the admission of opinions that are reliable (and actually expert) and likely to enhance factual accuracy, efficiency and the fairness of criminal justice processes.

Confronted with the rapid expansion in the availability of images, from the turn of the millennium Australian courts imposed epistemologically arbitrary constraints on the use of images to assist with the identification of POIs. (3) These restrictions appear to have been based on judicial anxiety about the value of the opinions but are blind to the actual abilities or accuracy of purported identification experts. Drawing upon mainstream scientific knowledge, we explain what courts should be looking for in order to identify: those who possess actual expertise; the scope and limits of their abilities; and how their opinions ought to be expressed. We also draw upon this knowledge base to consider how we might respond to emerging forms of expertise and expert systems that do not necessarily sit comfortably within conventional legal categories and conceptualisations --such as the opinions of super-recognisers, (4) the outputs of face matching algorithms and even combinations of humans and/or algorithms (ie hybrid systems). (5) This article offers an evidence-based approach to the interpretation of images--that is, it is concerned with scientific knowledge. Therefore, the article is conspicuously sensitive to accuracy and presenting evidence (whether an opinion or the output of an algorithm) in a manner that embodies its value so that it is susceptible of rational evaluation. (6)

In this article we refer to those whose abilities at face comparison have not been formally established--ie demonstrated in controlled conditions where the correct answer is known--as purported experts. We call them 'purported' because we do not know if they are actually expert at the specific task, namely identifying a POI or describing their features. (7) For more than a decade (up to 2014), Australian courts allowed purported experts to identify POIs or features said to be shared between a POI and a defendant. (8) Purported experts were not required to provide information about their abilities, the degree to which facial features are correlated, or how patterns of correlations vary (ie diagnosticity). (9) Evaluation of the opinions of purported experts was left to the exigencies of the trial. Reception was moderated by trial safeguards (eg cross-examination and judicial instructions) that were not always informed by contemporary scientific knowledge. (10) Eventually, following Honeysett v The Queen ('Honeysett'), (some) purported experts were prevented from proffering opinions, though the extent of that proscription remains uncertain. (11) Overall, Australian judges erected an admissibility regime that directs limited attention to the epistemic value of images or the serious risks posed by their interpretation. (12)

This article is intended to encourage courts to reconsider their position(s). Courts, we contend, should be more sceptical consumers of facial recognition and image comparison evidence. Guided by scientific research, courts can and should be more accommodating of some types of opinion evidence, but less accommodating of others. They should possess effective means of distinguishing between different types of direct and indirect witnesses, determining which indirect witnesses have genuine abilities, and gauging the probative value of both opinions and the outputs of face matching algorithms. (13) They should also be more attentive to scientific research on unconscious bias and its deleterious impacts on perception and interpretation. (14) Our analysis begins with a synoptic overview of the main developments in the Australian jurisprudence over the last two decades, including the expanding reliance on common law categories such as the 'ad hoc' expert. (15) In Part III we have assembled and synthesised scientific research on unfamiliar face comparison to provide the foundations for a more principled approach to the use of images. Drawing on these reviews, Part IV raises a series of considerations (and makes a few recommendations) that ought to inform legal reliance on images used as evidence of identity in criminal proceedings.

II LEGAL INTERPRETATIONS OF IMAGES AND EXPERTS

Now ubiquitous, images can fulfil a range of evidentiary functions in investigations and prosecutions. They can be used to identify persons, to track movements and to determine what was done, when and by whom. (16) All of these uses involve interpretations that may vary from the easy and mundane to those which are extremely difficult, contested, and error-prone. The value of images as evidence depends on their use (eg the type of interpretation), the quantity and quality of the images, as well as the abilities of those (whether individuals or algorithms or hybrid systems) interpreting them--see Part III.

In this article we are concerned with the use of images to assist with the identification of POIs--frequently those involved in criminal activity. This can be based on recognition (from memory and some degree of familiarity) or the comparison of a POI in images with reference images (usually of a person whose identity is known). (17) We are primarily concerned with identification by strangers based on comparisons--sometimes described as unfamiliar face matching or forensic image comparison. (18) That is, comparisons performed by those with little to no familiarity with the suspect/defendant (prior to an investigation). (19)

A Smith, Morgan, Honeysett and the Limits of s 79 Jurisprudence

Our review begins with Smith v The Queen ('Smith') (20)--the High Court's first attempt to respond to some of the implications of the dramatic expansion in the availability of images. (21) During the trial, two police officers with some limited exposure to Mundarrah Smith purported to identify him in low-quality images of an armed robbery--see Figure 1. (22) On appeal, the High Court deemed this evidence inadmissible. (23) According to the majority, the police officers were not entitled to identify Smith as the POI in the bank robbery because each offered nothing beyond what a jury could bring to the comparison. (24) During the course of the trial, members of the jury would have more exposure to the defendant than either of the police officers obtained during their previous dealings. (25) For the majority, the opinions (ie interpretations) of the police officers would add nothing to the jury comparisons. (26) They could not 'rationally affect (directly or indirectly) the assessment of the probability of the existence of a fact in issue in the proceeding'. (27) They were, by definition, irrelevant. (28) An exception--the Smith caveat--was available where witnesses held some non-trivial advantage over the trier of fact. (29) This might arise where the appearance of the defendant had changed, or movement (eg gait) was significant and they had been exposed to these in ways that gave them a distinct advantage over the trier of fact. (30) The majority explained:

In other cases, the evidence of identification will be relevant because it goes to an issue about the presence or absence of some identifying feature other than one apparent from observing the accused on trial and the photograph which is said to depict the accused. Thus, if it is suggested that the appearance of the accused, at trial, differs in some significant way from the accused's appearance at the time of the offence, evidence from someone who knew how the accused looked at the time of the offence, that the picture depicted the accused as he or she appeared at that time, would not be irrelevant. Or if it is suggested that there is some distinctive feature revealed by the photographs (as, for example, a manner of walking) which would not be apparent to the jury in court, evidence both of that fact and the witness's conclusion of identity would not be irrelevant. (31) Justice Kirby adopted a different course. He was unwilling to unilaterally invoke (ir)relevance in the High Court when it was not relied on by the parties or judges during the trial and appeals. (32) Rather than treat the police officers' impressions as irrelevant, for Kirby J there was no exception to the exclusionary opinion rule (33) that would render them admissible. (34) The opinions were not 'based on what [the officers] saw, heard or otherwise perceived...