Self-Policing: Dissemination and Adoption of Police Eyewitness Policies in Virginia

Introduction

Professional policing organizations and agencies increasingly emphasize the importance of law enforcement adopting clear and comprehensive written policies and procedures, as well as accompanying training and supervision.[1] The focus on adoption of written policy has been particularly urgent in high profile areas like the use of deadly force, in which federal consent decrees have resulted in extremely detailed written policies.[2]  The adoption of written policy has traditionally been far less common in the area of police evidence-gathering.[3] However, in recent years, professional policing organizations have emphasized the role of best practices in policing, including in areas such as eyewitness identifications in which patrol guides traditionally did not offer detailed guidance.[4] There is very little regulation of police, however, and state law itself rarely sets out practices that police should follow. Police must self-police: police organizations must themselves draft and disseminate policy. One pressing question is whether states or other actors, such as accreditation organizations, can disseminate model policies to promote adoption, particularly where there are so many smaller police agencies. This Essay describes a success story: one that suggests self-policing can occur under the right conditions.

One pressing area in which police policies had lagged was that of eyewitness procedures, where national policing organizations have strongly recommended the adoption of improved practices,[5] but at the same time, national surveys have found highly uneven adoption of those best practices.[6] Those best practices are extremely important. As the National Research Council explained in a landmark 2014 report summarizing the scientific research in the area of human visual memory, “it is well known that eyewitnesses make mistakes and that their memories can be affected by various factors including the very law enforcement procedures designed to test their memories.”[7] In particular, the hundreds of DNA exonerations in recent years, the vast majority of which involved eyewitness misidentifications, have brought home the malleability and fragility of eyewitness memory. DNA testing has resulted in the reversal of high-profile wrongful convictions, which have made the consequences of eyewitness misidentifications and poor police procedures particularly clear. In a book, I set out the results of a study of the role eyewitness evidence played in trials of the first 250 DNA exonerees.[8] Over two-thirds of those exonerees had been convicted based on eyewitness misidentifications, and most had been misidentified following the use of suggestive law enforcement identification procedures.[9]

In this Essay, I present the results of studies used to assess the adoption of eyewitness identification policies in Virginia, a state in which prior guidance was very thin. Law enforcement agencies were required in 2005 to adopt a written policy, but they were free to adopt any policy of their choosing.[10]  All of the over-300 law enforcement agencies in Virginia had their policies on this subject requested, using the state freedom of information law. While, as of 2005, agencies were only required to have some written policy in place, over time, particularly beginning in 2011, the state law enforcement policy agency supplied detailed model policy on eyewitness procedure. One reason  policymakers were focusing on this problem was that Virginia experienced a series of DNA exonerations in cases involving eyewitness misidentifications; as of 2013, thirteen of sixteen DNA exonerations in Virginia had involved eyewitness misidentifications.[11] Nevertheless, as reported in an earlier 2013 study, those model practices were only being haltingly adopted.[12] In particular, many agencies did not have blind or blinded lineups, in which the administrator does not know which photo is that of the suspect or cannot view which photo the eyewitness is examining. A substantial body of basic research examines how humans perceive images and form visual memory. That research has been complemented by applied research in the area of eyewitness identification. This research has resulted in a large body of knowledge concerning how to test visual memory accurately, including face identification, and a set of best practices that are recommended to test and preserve the memory of an eyewitness.[13] Scientists have emphasized that the use of blinding is “central to the scientific method because it minimizes the risk that experimenters might inadvertently bias the outcome of their research, finding only what they expected to find.”[14] Thus, blinding is essential to any objective factfinding.

Five years later, in 2018, we re-surveyed agencies and found that there is now extensive dissemination and widespread adoption of the state model policy. Specifically, the vast majority of agencies have adopted blinded policies, clear instructions to eyewitnesses, guidelines for selecting fillers for lineups, recording of confidence statements, and many agencies require videotaping of lineup procedures. Those improved eyewitness identification practices have been adopted by the vast majority of agencies over the past five years.

Part I discusses the law and science of eyewitness memory and the legislative and policy background in Virginia. Part II presents the findings of this study. This study, while it examines each of the features of the lineup policies adopted in Virginia, focuses first on the adoption of blind or blinded procedures. As described, such adoption occurred in less than half of the surveyed agencies in 2013, but by 2018, there had been a remarkable change, and the overwhelming majority had adopted blind or blinded procedures. Part III discusses why this change may have occurred and the implications of this experience in Virginia for future efforts to improve police practices.

I. Legislative and Policy Background

A. The Law and Science of Eyewitness Memory

That eyewitness memory can be highly unreliable in police investigations has been a staple of criminal procedure for centuries. As the U.S. Supreme Court has put it, “[t]he vagaries of eyewitness identification are well-known; the annals of criminal law are rife with instances of mistaken identification.”[15] However, the Supreme Court has not regulated police identification procedures in response to this well-known problem. Instead, the Court has set out broad and quite deferential rules that potentially exclude as evidence unnecessarily suggestive identifications, as a constitutional matter, leaving regulation of eviden­tiary questions largely to state courts or to police agencies in the first instance.[16]

Thus, as the National Research Council has suggested, “[t]he best guidance for legal regulation of eyewitness identification evidence comes not from constitutional rulings, but from the careful use and understanding of scientific evidence to guide fact-finders and decision-makers.”[17] Police officers use a variety of different procedures to ask an eyewitness to identify a culprit, including: (1) showups; (2) photo arrays; (3) live lineups; and (4) mugshots and computer presentations of photos in which there is no designated suspect.[18] In a showup, which usually occurs at or near the crime location and shortly after the crime occurred, officers present a single, live suspect to a witness.[19] In photo arrays, officers present the eyewitness with a series of photographs, one of which is the suspect, and the others called “fillers,” or known non-suspects.[20] Live lineups, in which the suspect and fillers are presented in person to an eyewitness, are less commonly used.[21] Additional procedures may be used in which officers do not have a suspect. If so, officers may show mug books or sets of photographs to see if the eyewitness can identify a suspect, or they may ask the eyewitness to help prepare a composite image or drawing of a culprit.[22]

In scientific terms, the procedures used to test eyewitness memory should take account of both estimator variables and system variables.[23] Both types of variables can affect the memory of an eyewitness.[24] Estimator variables are factors relating to the conditions of the crime-scene viewing, such as the lighting, the eyewitness’s eyesight, familiarity with the perpetrator, or race.[25] Studies have shown that individuals display an “own-race bias,” or a greater difficulty identifying persons of a different race.[26] Estimator variables cannot be controlled by law enforcement.[27] In contrast, system variables are factors associated with the procedures that officers use to obtain identifications by an eyewitness.[28] System variables can be controlled by law enforcement.[29]

The National Research Council report made quite clear its recommendation that blind or blinded lineups should be used by law enforcement.[30] This recommendation is based upon decades of research in a number of fields on the ways in which the expectations of an administrator can bias subjects, including through inadvertent means of communication. “Even when lineup administrators scrupulously avoid comments that could identify which person is the suspect, unintended body gestures, facial expressions, or other nonverbal cues have the potential to inform the witness of his or her location in the lineup or photo array.”[31] By contrast, “[t]he ‘blinded’ procedure minimizes the possibility of either intentional or inadvertent suggestiveness and thus enhances the fairness of the criminal justice system.”[32]

The National Research Council report also highlighted that agencies should adopt standard instructions for eyewitnesses.[33] Those instructions should inform the eyewitness that a culprit may or may not be present in the lineup.[34] That instruction is crucial because an eyewitness otherwise may expect that the culprit will be present and that there is a correct choice that should be made. Showups should be limited in their use.[35] Such an instruction can still be given before conducting a showup, and agencies should have standard instructions and procedures to avoid undue suggestion in showup procedures.[36] The confidence of the eyewitness should be documented, preferably through a recording of the entire eyewitness identification procedure.[37] Standard procedures should use terminology that is easily understandable by eyewitnesses.[38] There should be clear rules on the number and selection of fillers for lineups.[39] There are a number of state statutes and model policies that provide useful models for agencies,[40] as well as a recent update to the White Paper by the American Psychology-Law Society that summarizes the state of the research on eyewitness identification procedures.[41]

B. Virginia Regulation of Eyewitness Identifications

In 2005, the Virginia General Assembly enacted legislation requiring that police adopt some form of written eyewitness identification procedure.[42] DCJS had in place, from 1993 through 2005, an extremely brief model policy on eyewitness identification.[43] The new 2005 legislation was accompanied by a more detailed model policy.[44] However, the 2005 model policy did not include instructions on how to effectively “blind” a lineup by presenting photos in folders, held by the eyewitness so they cannot be seen by the administrator, without the need to obtain an administrator who is unfamiliar with the investigation. The folder-shuffle method is an inexpensive and practical solution to the problem of blinding; in addition, agencies can use computerized administration of eyewitness identification procedures.[45] In addition, the 2005 model policy had mandated sequential policies, but only made blind administration optional.[46] A 2010 survey by the Virginia State Crime Commission found that at least twenty-five percent of agencies responding to the survey still had no policy on the subject, despite enactment of that legislation five years earlier requiring that written procedures be adopted (and presumably even more agencies not responding lacked policies).[47]

C. The 2011 DCJS Model Policy

In Virginia, a new state model policy was adopted by the Department of Criminal Justice Services (“DCJS”) in 2011 in response to a series of DNA exonerations caused by eyewitness misidentifications, as well as concern with the slow pace of adoption of best practices.[48] The DCJS model policy is not only quite detailed but also flexible, as it is designed to be operationalized by agencies of different sizes and resources. The policy reflects each of the best practices highlighted in the National Research Council report (although it pre-dates that report). 

Importantly, the DCJS model policy states that all eyewitness identification procedures should be conducted blind, by an officer who does not know which photo is that of the suspect, or blinded, so that the administrator cannot tell whether the eyewitness is examining the suspect photo.[49] To respond to the concern raised by smaller agencies, that it is not always practically possible to spare an additional officer unfamiliar with the suspect’s identity, the DCJS model policy sets out the “folder shuffle” method.[50] That method provides a way to inexpensively make a procedure blind (and sequential) by placing the photos in folders and shuffling them, with several blanks at the end. The eyewitness can open the folders and examine the photos inside, without the administrator seeing what the eyewitness is viewing. In addition to requiring blind or blinded procedures, the 2011 DCJS model policy also sets out consistent instructions to eyewitnesses, requires careful documentation of the confidence and statements by the eyewitness, and encourages both audio and video recordings to be made of the entire identification procedure.[51] These procedures were far more detailed than the prior model policy, and they provide a comprehensive model for agencies of all sizes.[52] In 2014, minor revisions were made to the model policy, largely to clarify the folder method procedures.[53]

D. The 2013 Study of Virginia Lineup Procedures

A spring 2013 study of lineup procedures in Virginia found that despite the passage of additional time since the model policy had been promulgated and disseminated to agencies throughout the state, many agencies still were not adopting the model policy, and crucially, were not conducting lineups blind or blinded.[54] Of the 201 agencies that responded to either the survey or the Freedom of Information Act (“FOIA”) requests, 145 supplied eyewitness identification policies.[55] “Troubling findings included that in total, only 40%, or 58 of 144 policies,” provided by responding agencies, “required blind lineup procedures or made them available as an option.”[56] Only 6% of agencies that provided policies had adopted the entire model policy disseminated by DCJS.[57] “[O]f the 144 eyewitness identification policies reviewed, 29% or 42 policies required blind lineup procedures. Ten more required that blind lineups be used where practicable.”[58] Six more provided that blind lineups be available as an optional practice.[59]

Regarding the folder shuffle method, “[o]nly 9 agencies described the folder shuffle method as an option.”[60] The folder shuffle method was fairly new and had been first recommended, as noted, in the 2011 DCJS model policy as a way for small agencies to effectively blind a lineup procedure.[61] “Instead, far more common were policies that were sequential, but not blind: two-thirds or 63% of the departments required or offered sequential lineups (91 of 144).”[62] Those policies may have made lineups even less reliable, since a sequential policy introduces more interaction between the administrator and the eyewitness, since each picture is shown one at time; a non-blind sequential policy may give greater cause to fear that suggestion can contaminate the results.[63]

Further, “[o]nly 88 of 144 departments had required standard instructions [to the eyewitness] as a matter of policy.”[64] Only “[a]bout half, or 71 of 144 agencies’ policies, required taking a confidence statement of some kind using the eyewitnesses’ own words. Those that did require taking a confidence statement often did not detail how that should occur.”[65] Most (63%) had sequential policies; however, 23% had sequential but not blind policies.[66] As noted, such policies may be even more vulnerable to suggestion than policies that had previously been in place. And 41 agencies (mostly sheriffs’ offices that may not conduct investigations in which identification procedures would be used) responded that they did not have eyewitness identification policies.[67]

II. Results of the 2018 Study of Virginia Eyewitness Identification Policies

In fall 2018, every policing agency in Virginia was mailed a FOIA request for all eyewitness identification policies.[68] In 2018, 193 agencies supplied policies, which constituted a much higher response rate than to the FOIA requests made for the 2013 study. An additional 40 agencies responded that they do not conduct investigations and do not have a policy—or simply do not have a policy.

Blind and blinded lineups. Of those that did supply policies, 158 agencies had blind policies, with just 30 of the 159 saying that they conduct lineups blind only where it is practicable to do so. Just 22 agencies did not provide for blind or blinded lineups. Compare these figures to those in 2013, where only 58 policies required or made optional blind or blinded lineup policies.[69] While 166 agencies use sequential lineups, just a handful of agencies use sequential but not blind policies.

Folder method. In another major change from the 2013 study, in 2018, 129 agencies had adopted policies that use the folder method or made it available as an option when there is not an independent officer available to conduct the photo array procedure. Compare these figures to those in 2013, where only 9 policies included the folder shuffle option, indicating that smaller agencies were not aware of the DCJS recommendations to make blinded lineups feasible at low cost.[70]

Sequential lineups. Almost without exception, they required sequential as well (the problem had been with those that were sequential but not blind). Moreover, in 2013, there were 51 agencies, out of 144 responding that had extremely brief policies, many dating back to a rudimentary 1993 model policy, which were only a few paragraphs long and included no meaningful operational instructions.[71]

Instructions to witnesses. In 2013, many agencies did not have policies detailing instructions to be given to eyewitnesses; only 88 of 144 policies surveyed at that time included such guidance in policies.[72] In 2018, in contrast, this was nearly universal. Of the 193 policies obtained, 167 had policies on instructions to eyewitnesses, 168 had in their policies requirements and instructions on obtaining confidence statements from eyewitnesses,[73] and 166 had a required statement from the administrator to the eyewitness that the suspect may or may not be present.

Fillers. Almost all of the policies had language indicating that fillers should resemble the suspect and stating how many filler photos should be included in a lineup; 171 of the 193 policies included policies on fillers.

Recording. Regarding electronic recording of eyewitness identification procedures, another important change since 2013 was that 76 required videotaping and 42 additional agencies required audio. In 2013, only 25 agencies made recording an option.[74] This marks a dramatic shift towards electronic recording of eyewitness identification procedures.

The figure below displays a comparison as between the 2013 and 2018 findings, displaying the number of agencies adopting each type of procedure.

Figure 1. Virginia Lineup Policies, 2013 and 2018

[[{“fid”:”868″,”view_mode”:”full”,”type”:”media”,”attributes”:{}}]]

One important question was whether different agencies responded to the FOIA requests in 2018 as compared to 2013, which could explain the very different patterns observed in the policies. Of the 87 that did not have blind policies in 2013, most of those agencies, 50 of them, have since changed and adopted blind policies. Only 13 have remained the same, and continue not to use blind eyewitness identification policies. It was 23 of those agencies that did not have blind policies that did not respond to the 2018 survey. Nine agencies responding for the first time to the new 2018 survey, which did not respond in 2013, have non-blind policies.

DCJS Model Policy adoption. This change marks a near universal adoption of the 2011 DCJS model policy, with its revisions, by Virginia law enforcement agencies. Not all of these policies include verbatim all of the language from the DCJS model policy, but many of these agencies have recently adopted policies that use much of the DCJS language. Only a handful of agencies, in 2018, still had such cursory and problematic policies, which is also a very important change.

III. Implications for Future Efforts to Improve Police Practices

By 2018, the vast majority of Virginia residents lived in jurisdictions in which best practices regarding eyewitness identifications had been adopted. The larger agencies, with one exception, have all adopted these best practices. As a result, of the 8.5 million Virginia residents, only 9% or about 740,000 people, lived in jurisdictions that have not adopted these policies. Of the 126 accredited agencies in Virginia, only six very small agencies did not adopt the DCJS model policy in substance.[75]

What changed between 2013 and 2018? A combination of efforts by policing organizations, dissemination between agencies, regional training by DCJS, media coverage in response to the 2013 study, and involvement of the Crime Commission, all may have helped to drive this change. There was some media attention to the 2013 survey findings, and in response, some agencies stated that they planned to update their policies.[76] There was legislation introduced in 2014 to require that agencies conform to the DCJS model policy, but it was tabled in the committee in which it was introduced.[77] The introduction of that draft legislation may have had some limited impact.

The involvement of professional policing organizations was also crucial to the success of these efforts. Executive Director Dana Schrad of the Virginia Association of Chiefs of Police worked with police chiefs, following the prior survey, to assess efforts to revise policies.[78] As noted, almost all accredited agencies later adopted these policies, at least among responding agencies. The Virginia accrediting organization added additional language regarding content of eyewitness identification policies.[79]  Meanwhile, DCJS led new efforts to conduct training on the DCJS policy and made agencies aware of its provisions. The DCJS policy was modestly revised to respond to agencies’ questions about the meaning of certain provisions. Observers noted that successful implementation of the DCJS model policy in small jurisdictions “should be persuasive” to other small departments.[80] Agencies seemed to pass on information about the issue through word of mouth and sharing best practices. In addition, accrediting agencies, policing associations, insurance providers, and lawmakers remained interested in the problem and continued to make agencies aware of the need to pay attention to the issue.

Conclusion

The results of this survey give reason to be more optimistic that policing agencies, even in jurisdictions with large numbers of fairly small agencies, can adopt best practices in a consistent fashion, at least if sustained efforts are made to engage with agencies. In 2013, based on the results of the survey of Virginia lineup policies, I suggested that due to institutional inertia, not policy choices, there was far too slow a pace of adoption of best practices. At the time, it appeared that stronger regulatory measures might be needed to safeguard the accuracy of criminal investigations. However, five years later, the evidence suggests that policing institutions can slowly come to adopt best practices, without such stronger regulatory measures. A combination of training, growing awareness, media coverage, and sharing of information between agencies, may have resulted in a real statewide improvement of practices: self-policing was successful in adoption of improved police policy. These results suggest that close work—with professional policing organizations, policymakers, accreditation bodies, and police agencies themselves—can be an effective way to shape law enforcement policy through self-policing.

 


[1] See, e.g., Police Exec. Research Forum, Guiding Principles on Use of Force (2016), https://www.policeforum.org/assets/30%20guiding%20principles.pdf [https://perma.cc/D4­TQ-LZP7] (recommending policy for police use of force); President’s Task Force on 21st Century Policing, Final Report of the President’s Task Force on 21st Century Policing 19–20 (2015) [hereinafter President’s Task Force on 21st Century Policing], https://­cops.usdoj.gov/pdf/taskforce/taskforce_finalreport.pdf [https://perma.cc/4PL3-94FY] (emph­asizing that law enforcement agencies “should have comprehensive policies” on issues such as the use of force as well as policies that are “reflective of community values”).

[2] Brandon L. Garrett & Seth Stoughton, A Tactical Fourth Amendment, 103 Va. L. Rev. 211 (2017) (describing federal consent decrees and analyzing the text of use of force policies adopted by major law enforcement agencies). For an example of a state law requiring agencies to adopt written policies regarding certain areas, including use of force, see, e.g., Me. Rev. Stat. Ann. tit. 25, § 2803-B (2015).

[3] See Stanley Z. Fisher, “Just the Facts, Ma’am”: Lying and the Omission of Exculpatory Evidence in Police Reports, 28 New Eng. L. Rev. 1, 18 (1993).

[4] The International Association of Chiefs of Police (“IACP”) has taken an active role in promoting consideration of ways to improve the “accuracy and thoroughness” of police investigations. Int’l Ass’n of Chiefs of Police, National Summit on Wrongful Convictions: Building a Systemic Approach to Prevent Wrongful Convictions 10 (2013), https://www.bja.gov/Publications/IACP-Wrongful_Convictions_Summit_Report.pdf [https://perma.cc/EY7P-3JCQ].

[5] See, e.g., Va. State Crime Comm’n, HB 207: Law Enforcement Lineups 15 (2010), http://­vscc.virginia.gov/documents/2010/law_lineups.pdf [https://perma.cc/TV4H-9AVD] (refer­encing CALEA’s Law Enforcement Policy 42.2.11 as a model policy); U.S. Dep’t of Justice, Tech. Working Group for Eyewitness Evidence, Eyewitness Evidence: A Guide for Law Enforcement iii (1999), https://www.ncjrs.gov/pdffiles1/nij/178240.pdf [https://perma.cc/­U2CG-XQ57]; Int’l Ass’n of Chiefs of Police, Model Policy: Eyewitness Identification (2016) [hereinafter IACP], https://www.theiacp.org/sites/default/files/2018-08/EyewitnessIDPolicy­2016.pdf [https://perma.cc/VRL2-CXJS].

[6] Many agencies have policies that are decades out of date, or they have no written policies at all. See, e.g., Police Exec. Research Forum, A National Survey of Eyewitness Identification Procedures in Law Enforcement Agencies 46–47, 89 (2013), https://www.ncjrs.gov/­pdffiles1/nij/grants/242617.pdf [https://perma.cc/2AM3-QT44].

[7] Nat’l Research Council, Identifying the Culprit: Assessing Eyewitness Identification 1 (2014) [hereinafter Nat’l Research Council].

[8] Brandon L. Garrett, Convicting the Innocent: Where Criminal Prosecutions Go Wrong 9, 48 (2011).

[9] Id. at 49.

[10] Va. Code Ann. § 19.2-390.02 (West 2005).

[11] Brandon L. Garrett, Eyewitness Identifications and Police Practices: A Virginia Case Study, 2 Va. J. Crim. L. 1, 7–8 (2014) (describing examples of misidentifications from Virginia DNA exonerations). Similarly, the Virginia Department of Criminal Justice Services (“DCJS”) noted in their model policy that “[t]en of thirteen DNA exonerations in Virginia involved eyewitness misidentifications. Few cases in Virginia have been suitable for DNA testing, since the policy until the last decade was that crime scene evidence would be destroyed post-conviction. Those Virginia eyewitness identifications involved suggestive and unreliable eyewitness identification procedures.” Va. Dep’t of Criminal Justice Services, Model Policy on Eyewitness Identification, General Order 2-39, at 1 (2014) [hereinafter DCJS 2014 Model Policy]. Note that the author served on a Virginia Crime Commission task force that informed the revision of this DCJS model policy.

[12] See Garrett, supra note 11 (presenting a study of Virginia law enforcement policies, of which few complied with state model policy on lineup procedures).

[13] See, e.g., Nat’l Research Council, supra note 7, at 103–09.

[14] Id. at 106.

[15] United States v. Wade, 388 U.S. 218, 228 (1967).

[16] Manson v. Brathwaite, 432 U.S. 98, 114 (1977). The “reliability” factors adopted by the Court in Manson, having been already set out in its earlier ruling in Neil v. Biggers, 409 U.S. 188, 199–200 (1972), ask that the judge examine: (1) the eyewitness’s opportunity to view the defendant at the time of the crime; (2) the eyewitness’s degree of attention; (3) the accuracy of the description that the eyewitness gave of the criminal; (4) the eyewitness’s level of certainty at the time of the identification procedure; and (5) the length of time that had elapsed between the crime and the identification procedure. Manson, 432 U.S. at 98–99. The Court did not assign any particular weight to these various factors. The Supreme Court more recently has held that when unreliability in eyewitness identifications is not due to intentional police action, it is not regulated under the Due Process Clause at all. Perry v. New Hampshire, 565 U.S. 228, 248 (2012). The Justices in Perry stated that the Court did “not doubt either the importance or the fallibility of eyewitness identifications,” but held that state legislation, evidence law, and safeguards such as expert testimony and jury instructions should be relied on to ensure the accurate presentation of eyewitness evidence. Id. at 245–46. A large body of scientific research has called into question the validity of many of the Supreme Court’s so-called “reliability” factors. For scholarly criticism in light of the social-science research, see, e.g., Suzannah B. Gambell, Comment, The Need to Revisit the Neil v. Biggers Factors: Suppressing Unreliable Eyewitness Identifications, 6 Wyo. L. Rev. 189, 196–202 (2006); Timothy P. O’Toole & Giovanna Shay, Manson v. Brathwaite Revisited: Towards a New Rule of Decision for Due Process Challenges to Eyewitness Identification Procedures, 41 Val. U. L. Rev. 109, 118–22 (2006); Gary L. Wells & Deah S. Quinlivan, Suggestive Eyewitness Identification Procedures and the Supreme Court’s Reliability Test in Light of Eyewitness Science: 30 Years Later, 33 L. & Hum. Behav. 1, 16 (2009).

[17] See Nat’l Research Council, supra note 7, at 5; see also President’s Task Force on 21st Century Policing, supra note 1, at 23 (recommending adoption of identification procedures “that implement scientifically supported practices that eliminate or minimize presenter bias or influence”).

[18] See DCJS 2014 Model Policy, supra note 11, at 2–3.

[19] See id. at 4–5.

[20] See IACP, supra note 5, at 1.

[21] See Gary L. Wells et al., Policy and Procedure Recommendations for the Col­lection and Preservation of Eyewitness Identification Evidence, at 11–12 (forth­coming 2020), http://ap-ls.wildapricot.org/resources/Documents/Feb42019EWwhi­tepaper.pdf [https://perma.cc/W3­YH-G5L8].

[22] A recent draft White Paper describes the state of scientific research on eyewitness identification procedures. Id.

[23] Gary L. Wells, Applied Eyewitness-Testimony Research: System Variables and Estimator Variables, 36 J. Personality & Soc. Psychol. 1546 (1978) (first coining the terms “estimator” and “system” variables).

[24] Id. at 1548.

[25] Id. at 1548–50.

[26] See Nat’l Research Council, supra note 7, at 96.

[27] Wells, supra note 23, at 1548.

[28] See id. at 1552–55.

[29] Id. at 1552.

[30] Nat’l Research Council, supra note 7, at 3.

[31] Id. at 106.

[32] Id. at 107.

[33] Id. at 3.

[34] Id. at 107.

[35] See id. at 26–27.

[36] See id.

[37] See id. at 108–09.

[38] Id. at 107.

[39] See id. at 26–27.

[40] See, e.g., N.C. Gen. Stat. Ann. § 15A-284.52 (West 2016); Ohio Rev. Code Ann. § 2933.83 (West 2010); Va. State Crime Comm’n, HB 207: Law Enforcement Lineups 15 (2010), http://vscc.virginia.gov/documents/2010/law_lineups.pdf [https://perma.cc/TV4H-9AVD] (referencing CALEA’s Law Enforcement Policy 42.2.11 as a model policy); IACP, supra note 5; DCJS 2014 Model Policy, supra note 11.

[41] See Wells et al., supra note 21.

[42] Virginia Code § 19.2-390.02 (West 2005) states: “Policies and procedures for law enforcement to conduct in-person and photo lineups—The Department of State Police and each local police department and sheriff’s office shall establish a written policy and procedure for conducting in-person and photographic lineups.”

[43] See Va. Dep’t of Criminal Justice Services, Report on the Law Enforcement Lineup Policy Survey and Review, General Order 2-1, at 26 (2012), https://web.archive.org/­web/20150915001917/http://www.dcjs.virginia.gov/research/documents/LawEnforceLineup_final.pdf [hereinafter Lineup Policy Survey].

[44] See id., General Order 2-39, at 20–25.

[45] See Nat’l Resource Council, supra note 7, at 107.

[46] See Va. State Crime Comm’n, HB 207: Law Enforcement Lineups 8 (2010), http://vs­cc.virginia.gov/documents/2010/law_lineups.pdf [https://perma.cc/TV4H-9AVD].

[47] See id. at 18.

[48] Va. Dep’t of Criminal Justice Services, Model Policy on Eyewitness Identification, General Order 2-39, at 1 (2011) [hereinafter DCJS 2011 Model Policy] (on file with Virginia Law Review Association). 

[49] Id. at 6–7.

[50] Id. at 5–6.

[51] Id. at 9–11.

[52] See Lineup Policy Survey, supra note 43, General Order 2-1, at 26.

[53] See DCJS 2014 Model Policy, supra note 11, at 5–7.

[54] Garrett, supra note 11, at 15–16. “A DCJS follow-up survey of 267 law enforcement agencies in September 2011 created additional new cause for concern,” since it “indicated that most departments still had not adopted best practices.” That survey was conducted shortly after the new model policy took effect. Id. at 13.

[55] Id. at 6.

[56] Id.

[57] Id.

[58] Id. at 15.

[59] Id.

[60] Id. at 17.

[61] See DCJS 2011 Model Policy, supra note 48, at 5. 

[62] Garrett, supra note 11, at 17.

[63] See id. at 17–18.

[64] Id. at 20.

[65] Id. at 21.

[66] Id. at 17.

[67] Id. at 14–15.

[68] Part II analysis is derived from data collected by the author. Brandon L. Garrett, VA Eyewitness Policies Data Set (2019) (on file with author).

[69] Garrett, supra note 11, at 15.

[70] Id. at 17.

[71] Id. at 18.

[72] Id. at 20.

[73] In 2013, only half of the agencies required taking a confidence statement from an eyewitness. Id. at 21.

[74] Id.

[75] I am grateful to Gary Dillon for his assistance with this analysis.

[76] See, e.g., Jordan Fifer, Lineup Policies Slow to Catch On, Roanoke Times, (Aug. 31, 2013), https://www.roanoke.com/news/virginia/lineup-policies-slow-to-catch-on/article_98­95c72d-c5b5-5f9c-8b80-10c12b9b8f0c.html [https://perma.cc/XD77-FGLE] (describing how agencies were in the process of updating their policies).

[77] See H.D. 805, 2014 Gen. Assemb., Reg. Sess. (Va. 2014).

[78] See Fifer, supra note 76.

[79] See Va. Law Enf’t Prof’l Standards Comm’n, Virginia Law Enforcement Accreditation Program Manual 26, 54–55 (2016), https://www.dcjs.virginia.gov/sites/dcjs.virginia.gov/­files/publications/law-enforcement/virginia-law-enforcement-accreditation-program-man­ual.pdf [https://perma.cc/Z4KL-PSM6]. In Virginia, agencies may be accredited through the Virginia Law Enforcement Professional Standards Commission. See Va. Law Enf’t Prof’l Standards Comm’n, About VLEPSC, http://vlepsc.org/wp/?page_id=96 [https://per­ma.cc/H7PZ-FBWV] (last visited July 18, 2019).

[80] Karen L. Bune, Virginia Pushes a ‘Best Practice’ Model for Suspect Lineups, PoliceOne.com (Oct. 4, 2013), https://www.policeone.com/chiefs-sheriffs/articles/6493783-Virginia-pushes-a-best-practice-model-for-suspect-lineups/ [https://perma.cc/YE5T-LXH3].

The Future is Almost Here: Inaction is Actually Mistaken Action

Professor Ryan Calo has written a response to our article appearing in the current issue of the Virginia Law Review, Automated Vehicles and Manufacturer Responsibility for Accidents: A New Legal Regime for a New Era,[1] in which we develop the proposition that the United States is on the verge of a new era in transportation, requiring a new legal regime. As many commentators have noted, over the coming decades, there will be a revolution in driving, as manually-driven cars are replaced by automated vehicles (AVs).[2] There will then be a radically new world of auto accidents: most accidents will be caused by cars, not by drivers.

In a world of accidents dominated by AVs, current tort doctrine, in our view, will be anachronistic and obsolete. We present a proposal for a more effective system, adopting strict manufacturer responsibility for auto accidents. We call this system Manufacturer Enterprise Responsibility, or “MER.” In describing and developing our proposal for MER, we present a detailed, extensively analyzed approach that would promote deterrence and compensation more effectively than continued reliance on tort in the coming world of auto accidents.[3] MER would be a manufacturer-financed, strict responsibility bodily injury compensation system, administered by a fund created through assessments levied on AV manufacturers.

Professor Calo does not disagree with the framework of our proposal—which he describes as “sensible and well thought out”[4]—but takes a different tack: arguing that legal scholarship, such as our proposal, has limited capacity to anticipate the future of unfolding law and technology. Here, we respond to his main points.

We agree with Professor Calo’s opening supposition that proposing policymaking about issues that may take decades to become ripe may well be like imagining, in the 1950s, what it would be like to commute to Mars. But this analogy is seriously misleading, because AVs are already on the roads being tested. It is as if we were already traveling to Mars on an experimental basis, with routine trips back and forth being planned for the near future. We cannot afford to wait and see what the future brings over a period of decades; a world in which there are privately owned AVs being operated on highways and city streets is just over the horizon. The failure to do something about that is not the equivalent of keeping our policymaking powder dry. Similarly, Professor Calo’s caution that AVs represent a “qualitatively distinct affordance”[5]—both in vehicle design (absence of steering wheels, gas and brake pedals, and so on), and traffic patterns (of commuting and congestion)—has no bearing on the fact that vehicular injuries will still occur, even if in reduced number, which will need to be addressed by accident law.

In this regard, the current liability regime will apply to AVs unless affirmative steps are taken to adopt a different regime. Given the distinct issues and problems that accidents involving AVs will pose, products liability law will be making a “sweeping intervention[],”[6] to reference Professor Calo’s cautionary note, into the world of AV accidents, whether we like it or not. The question is whether we simply let that happen even though we can predict that passively allowing this intervention will have negative consequences, or we devise an intervention that will have more positive consequences than watching as the current system is less and less capable of handling AV accidents. This leaves us with two options.  We could let the transition take place and watch our current system become less and less capable of handling the new regime, or we can devise an intervention that addresses the challenges we will face. As our article indicates, the serious difficulties posed at present in design defects cases will be greatly exacerbated in assigning liability in conventional products liability terms in the context of the esoterica of safety-generating algorithms and machine learning.[7]

It may be, as Professor Calo imagines, that Google and Uber, rather than auto manufacturers, will become the central players in developing AVs. But some entity will still have to make the vehicles, and they will consist of both software developed by the Googles and Ubers of this world and hardware made by the vehicle manufacturers. Google and Uber are not likely to be manufacturing brakes, tires, and air bags. If Google and Uber are responsible only for accidents caused by software, and Ford, GM, and Toyota are solely responsible for accidents caused by hardware, retaining conventional products liability will generate the very litigation nightmare that MER is designed to avoid.[8] Under our approach, if it turns out that Google and Uber rather than Ford and Toyota should be responsible for all accidents “arising out of the operation of HAVs,”[9] then our proposal can be transposed so that Google and Uber are responsible for contributions to the MER fund.[10] The basic rationale for our proposal will still be applicable.

Professor Calo also is concerned that, at some point in the unforeseeable future, individuals will not own AVs. Instead, third parties such as Google, Uber, or cities will own them. He seems to think that our assumption that “individuals will own and insure their own cars,”[11] if incorrect, will undermine our proposal—for example, because we deliberately omit protection against property damage from MER. But so what? Nothing turns on this. Ride-sharing owners would also presumably have to decide whether to buy property (collision and comprehensive) insurance or to self-insure. Whether these entities or individuals have to make that decision would not change anything of consequence.

He also comments that the “incentive structure of an app-based technology company that both owns and operates its vehicles differs rather markedly from that of a car manufacturer that sells vehicles to people.”[12] Perhaps, but in what relevant respects? No one thinks that products liability law generates different incentives for conglomerates than for single-product companies. Why should there be a difference in this regard between technology companies and manufacturers? If there is an insight here, it eludes us.

The fact that “dramatically distinct modalities of transportation”[13] could arise once there are a lot of AVs on the roads leads Professor Calo to argue against pre-empting state experimentation with a single national approach. We would concur in a qualified fashion. This concern is why MER would not be triggered until AVs constitute twenty-five percent of all registered vehicles. That condition is meant to provide ample time for social, cultural, and technological evolution to tell us what “transportation modalities” actually have developed. But there is a deeper point to be made here. If the world of tomorrow features flying drones and/or comprehensive mass transportation, that will be soon enough to rethink liability law. Should we remain frozen until then?

Relatedly, Professor Calo asks what assumptions we are making about the trajectory of law and technology or its social impact. In fact, we are making no assumptions. Why should we be addressing whether a MER-type system would apply to harms from surgical robots? Having said that, we agree that the theory underlying MER may extend “well beyond”[14] driverless cars. The theory could extend to other kinds of accidents as well. It might even be nice (though it might also be unwise, or even dangerous) to have a single, unified approach to compensation and liability for accidental physical harm. New Zealand has done that, with mixed results.[15] In the United States we have never done it. As a practical matter, it is impossible.

In the United States we almost always address one major social policy concern at a time, instead of trying to address all our problems at once. Workers’ compensation is illustrative, even though the theory underlying workers’ compensation might also have applied to other settings. Similarly, auto no-fault compensation is grounded in the motor vehicle context. It addresses the cluster of issues associated with auto accidents, although some of these issues also arose, and still arise, in connection with injuries caused by defective products and harmful environmental exposures.

In the last few decades we have adopted a number of targeted compensation funds—the childhood vaccine-injury fund,[16] the 9/11 victim compensation fund,[17] and the birth-related neurological injury funds in Florida[18] and Virginia.[19] The principles underlying these approaches are not entirely consistent either, as one of us has repeatedly noted.[20] Since we are not going to have consistency across all of our liability and compensation regimes, deferring to lack of consistency—along with inability to predict long-term future scenarios of accident-generating technology—is a prescription for paralysis. Consequently, if in the future we need to think about applying MER to drone-related accidents, or robots performing surgery, we can face those issues as the need arises.

In short, because doing nothing about the law governing accidents involving AVs would actually constitute mistaken action, we should do the sensible thing. In our view, MER is exactly that.

 


[1] Kenneth S. Abraham & Robert L. Rabin, Automated Vehicles and Manufacturer Respon­sibility for Accidents: A New Legal Regime for a New Era, 105 Va. L. Rev. 127 (2019).

[2] See id. at 129–30 nn. 4–6 (collecting sources).

[3] Id. at 147–64.

[4] Ryan Calo, Commuting to Mars: A Response to Professors Abraham & Rabin, 105 Va. L. Rev. Online 84, 84 (2019).

[5] Id. at 86.

[6] Id. at 84.

[7] Abraham & Rabin, supra note 1, at 139–45.

[8] An alleged braking failure causing injury, for example, might very well raise issues of joint responsibility of the software and hardware systems.

[9] Abraham & Rabin, supra note 1, at 132.

[10] This is likely to be a relatively objective determination of which entity has dominant responsibility for manufacture of the vehicle.

[11] Calo, supra note 4, at 87.

[12] Id. at 87.

[13] Id.

[14] Id. at 88.

[15] For discussion, see Peter H. Schuck, Tort Reform, Kiwi Style, 27 Yale L. & Pol’y Rev. 187 (2008).

[16] 42 U.S.C. §§ 300aa-1–34 (2012).

[17] 49 U.S.C. § 40101 (2012).

[18] Fla. Stat. Ann. §§ 766.301–16 (West 2018).

[19] Va. Code. Ann. §§ 38.2-5000–21 (West 2018).

[20] See, e.g., Robert L. Rabin, The September 11th Victim Compensation Fund: A Circum­scribed Response or an Auspicious Model, 53 DePaul L. Rev. 769 (2003).

Commuting to Mars: A Response to Professors Abraham and Rabin

As Yogi Berra once said, it is difficult to make predictions, especially about the future. In Automated Vehicles and Manufacturer Responsibility for Accidents: A New Legal Regime for a New Era, Professors Kenneth Abraham and Robert Rabin propose a detailed system for addressing injuries caused by driverless cars.[1] The system strikes me as sensible and well thought out given how we use cars today. But if our relationship to vehicles continues to shift with the technology, then the solution on offer has the potential to unravel.

The remarks that follow are less about the particular wisdom of manufacturer enterprise responsibility (MER) for driverless cars, and more about the limits of legal scholarship in grappling with unfolding technologic change. The contingency of technology and its social impacts caution against sweeping interventions. And the role of law and technology scholarship—as opposed to legal scholarship that touches upon technology—is arguably to recognize the unique challenges that arise at this intersection.

I.

Driverless cars are having a moment. It is not their first. In the 1990s, the Department of Transportation (DOT) became very invested in the prospect of automated vehicles for many of the same reasons that Professors Abraham and Rabin cite. The agency wrote lengthy reports on the technology and even funded an ambitious demonstration, which took place in California in 1997 to great acclaim.[2] The event garnered unprecedented media attention and popularized a term, “Intelligent Transportation Systems,” that Professors Abraham and Rabin never use, and that you have probably never even heard.[3]

Thirty years ago, the DOT assumed that the intelligence behind automated vehicles would arise from the infrastructure—street smarts, as it were. The cars of the future (i.e., of today) were to ride upon virtual rails embedded in highways and roads. Automated vehicles would avoid collision by monitoring one another, but primarily they would interact with an augmented transportation environment.

This configuration did not come to pass. If it had, the legal structures Professors Abraham and Rabin propose—responsive as they are to the complex, sometimes inscrutable design decisions made by individual manufacturers—would be of limited utility. What occurred instead is that innovations in sensing technology and machine learning in the wake of the Defense Advanced Research Projects Agency (DARPA) Grand Challenge in Nevada shepherded in an era of driverless cars with good enough sensors and processors to navigate our streets without coordination or assistance.[4]

Professors Abraham and Rabin aim their sensible intervention at the contemporary model of a driverless car and the “radically new world of auto accidents” it portends.[5] On their view, which is widely shared by government and industry, we should expect more and more vehicles on the road that are capable of driving without human intervention. At some point these vehicles will reach a critical mass—say, twenty-five percent of all registered vehicles[6]—at which time the recommendation is to pass sweeping national legislation that preempts state regulations and common law and establishes primary and exclusive liability for bodily injury in manufactures through a mandatory fund.[7]

The authors assume that driverless cars will continue to operate roughly as they do in prototype. I see this as a solid assumption about the technology. From what I know of the state of driverless car technology, and particularly given the enormous investments by industry in sensors and machine learning, I would be surprised if the basic approach to automating vehicles were to shift again dramatically in the near term.

I would be equally surprised if Americans retained anything like their present relationship to cars.

It is tempting to understand driver automation as simply the end point of a continuum, a mental model that the five “levels” developed by the Society of Automotive Engineers reinforces.[8] In actuality, the prospect of vehicles that do not require humans to drive them represents a qualitatively distinct affordance. Think of how different cities would look if parking downtown were unnecessary.[9] Imagine the variations in vehicle design that the absence of steering wheels, pedals, or even windshields will support. Consider how a vehicle that could safely drop your child off at school would affect your commute or the shape of your school district.

Indeed, traditional automotive engineers are not driving the present revolution. The leaders in the field of automated vehicles are technology companies. Driverless car pioneer Waymo is a spinoff of Google, which of course provides free digital services on an advertising model. Its nearest rival is Uber, a ride-sharing app that delivers transportation on demand. These companies are not interested in selling cars to individuals—they invest billions in automation because of the prospect of a limitless, exquisitely coordinated reservoir of robots capable of moving people around.

The prospect that widespread vehicle autonomy will dramatically alter the way humans get around has consequences, including for the proposal Professors Abraham and Rabin lay out in careful detail. As Professors Abraham and Rabin remark in passing, “private vehicle ownership may go the way of the horse-and-buggy.”[10] Nevertheless, the liability system on offer repeatedly assumes that individuals will own and insure their own cars. For example, MER would exclude property damage on the apparent assumption that highly automated vehicle owners “still will likely purchase conventional auto insurance.”[11]

When I conceive of even the immediate future of driverless vehicles, I do not think of a trip to the Volkswagen dealership to trade in my level three for a shiny new level four. I think of diversified fleets of automated vehicles owned by the companies that made them, or by large civic units such as cities, and deployed as a transportation resource in the near term. The relationship between an injured consumer and a private or public service, meanwhile, raises distinct considerations from traditional products liability. And while Uber and Volkswagen are each capable of spreading and avoiding costs, the incentive structure of an app-based technology company that both owns and operates its vehicles differs rather markedly from that of a car manufacturer that sells vehicles to people.

I don’t have a crystal ball, any more than the authors. Imagine that neither I nor Professors Abraham and Rabin have correctly identified the future relationship of most Americans to vehicles or the timescale upon which change will occur. The very prospect that dramatically distinct modalities of transportation could arise from the ability of vehicles to drive themselves seems to caution against a preemptive, administratively intense solution that forbids state legislatures or courts from experimentation. Not even the apparent inspiration for MER—workers compensation—represents federal policy; workers comp is rather a creature of the state that can vary accordingly. Said another way, the authors’ proposal is certain; the future is not.

II.

The puzzle of how to deal with the contingency of technology and its social impacts is not limited to driverless cars, but endemic to law and technology scholarship. Personally I doubt Professors Abraham and Rabin—each renowned scholars of civil liability—identify themselves as working in “law and technology” as such. I imagine that for the authors, the ascendance of automated vehicles is just a fact about the world like any other, as the progress of technology often is.[12] In my experience, however, reasoning about technological change sometimes requires special care.

Take the concept of the “driverless car.” The underlying innovations that make driverless cars possible are, again, the introduction of new sensors (especially lidar) and improvements in techniques of machine learning that help computers recognize and react to patterns. These innovations introduce new human affordances, in the sense of additional capabilities to interact with our environment.[13] But even assuming the end goal is safely moving people about, there is nothing inevitable about combining these constituent technologies in a traditional car. That decision flows from a constellation of choices dating back to the concerted effort of the automotive industry to promote individual car ownership, and the attendant—and, some allege, purposive—decimation of public transportation.[14]

If we disaggregate the innovation of more accurate sensors and better machine learning from the construct of a driverless car, then far broader legal ramifications seem to follow. Consider Professors Abraham and Rabin’s key argument around why MER is necessary: neither federal agencies nor courts and juries possess the expertise to unpack the “esoteric, algorithm-based design differences” between highly automated vehicles.[15] Surely this concern extends well beyond driverless cars to the very wide array of robotics and cyber-physical systems being marketed and developed today.[16] The reader would be forgiven for wondering what intellectual foundation there could be to arguing for an elaborate and expensive MER regime for vehicles but nothing else.

My own view—which I am developing in connection to a larger project on law and technology—is that any legal scholarship that interacts with physical and digital artifacts would do well to state and defend a series of assumptions. Legal scholarship in general tends not to dwell on questions of methodology. I understand that Professors Abraham and Rabin expect MER to be evaluated on the basis of efficiency and cost-benefit analysis because I teach torts and I recognize the language of “optimal,” “adverse selection,” and “transaction costs.” But law and technology scholarship in particular would benefit from reflecting on a series of choices that are today largely implicit.

One set of choices involves the methods and goals of the author. Much law and technology literature follows Professor Rabin’s former colleague Lawrence Lessig in understanding new technology as revealing “[l]atent ambiguities,” or gaps in the law that jurists must now resolve.[17] Scholarship in this mold is at once progressive, in that it takes technological progress as inevitable, and conservative, in that it understands the role of law as restoring the status quo ex ante in light of a disruption. There are alternatives to this approach, ranging from purely descriptive research that helps provide ground truth, to normative projects that understand new technology as an invitation to rethink what constitute realistic societal goals.

The response you are reading concerns another choice: What is the scope of the technology under examination? Relatedly, what assumptions are the authors making around the trajectory of the technology or its social impact?

As a longtime science fiction fan, I remember coming across the early days of the genre depicting marvelous progress in technology even as social norms somehow remained constant—1950s science fiction classics that invite the viewer to picture a world in which businessmen routinely commute to Mars on aircraft piloted by white men and serviced by stewardesses in short skirts. The reality is far more complex. Technological change occurs against a backdrop of social, cultural, and economic forces that in turn shape the trajectory of the technology itself.

A wide variety of factors suggest that we may be on the cusp of a sea change in transportation. The very distinctions between manufacturer, owner, and consumers seem likely to collapse. The trouble with technology—and hence law and technology scholarship—is that truly novel affordances tend to invite reexamination of how we live. Legal scholarship must acknowledge this prospect or risk being its casualty.

 

 


[1] Kenneth S. Abraham & Robert L. Rabin, Automated Vehicles and Manufacturer Respon­sibility for Accidents: A New Legal Regime for a New Era, 105 Va. L. Rev. 127 (2019).

[2] For a first-hand description, see Chuck Thorpe et al., The 1997 Automated Highway Free Agent Demonstration, 1998 IEEE 496–501.

[3] Berkeley California PATH, National Automated Highway Systems Consortium (2019), [https://perma.cc/PV68-GR49] (last visited Feb. 21, 2019) (“Demo ’97 generated an unprece­dented level of media attention for Intelligent Transportation Systems, and we have made no attempt to capture the media aspects of that event here.”).

[4] See John Markoff, Machines of Loving Grace: The Quest for Common Ground Between Humans and Robots 23, 28–29, 36 (2015) (discussing advances in machines in the wake of the DARPA Grand Challenge).

[5] Abraham & Rabin, supra note 1, at 128–29.

[6] Id. at 132.

[7] See id. at 148–49, 151–52.

[8] The authors describe and embrace the levels in the Introduction. Id. at 129–31.

[9] See Alan Ohnsman, The End Of Parking Lots As We Know Them: Designing For A Driverless Future, Forbes (May 18, 2018, 11:31 AM), [https://perma.cc/FSJ5-A34J]; Tech Policy Lab, Driverless Seattle: How Cities Can Prepare for Automated Vehicles (Mar. 1, 2017), [https://perma.cc/M7GN-ZMUC].

[10] Abraham & Rabin, supra note 1, at 130.

[11] Id. at 151–55.

[12] Thus, for example, the Washington Supreme Court found no negligence on the part of the county for failing to build guardrails in 1928 capable of stopping a horse-drawn cart but not a car. Davison v. Snohomish Cty., 270 P.2d 422, 423 (Wash. 1928). That same court would later require a jury to determine whether a railway company was negligent forty years later for failing to protect against a similar accident on the theory that materials had become stronger and cheaper. Bartlett v. N. Pac. Ry. Co., 447 P.2d 735, 737 (Wash. 1968).

[13] For a discussion of affordances, see Ryan Calo, Can Americans Resist Surveillance?, 83 U. Chi. L. Rev. 23, 26–30 (2016).

[14] See, e.g., Robert C. Fellmeth, Politics of Land: Ralph Nader’s Study Group Report on Land Use in California 406, 415 (1973); see also Kate Crawford & Ryan Calo, Comment, There Is a Blind Spot in AI Research, 538 Nature 311, 311–313 (2016) (discussing how the trolley problem obscures the ways that investment in autonomous driving perpetuates prioritization of private cars over public transport). 

[15] Abraham & Rabin, supra note 1, at 142–44. The Toyota sudden acceleration scandal of 2011 provides an excellent example. Congress instructed the DOT to determine whether the reported propensity of Toyotas suddenly to accelerate was the product of a glitch in the software. The DOT lacked the requisite expertise to answer this question and wound up having to ask NASA to take a break from placing robots on Mars to look at a Toyota for them. It took NASA four months, but they eventually cleared Toyota’s software. See Ryan Calo, The Case for a Federal Robotics Commission, Brookings Inst. (Sept. 15, 2014), [https://perma.cc/4LS2-T7HM].

[16] These include drones, delivery carts, surgical robots, and personal and service robots.

[17] Lawrence Lessig, Code and Other Laws of Cyberspace 22 (1999).