Author: Samuel Scialabba

Proponents of virtual reality (VR) as a medium for evidence in the courtroom have argued that it can bring many benefits to jurors, including enhanced empathy and better factual understanding. However, it is also speculated that VR could increase a juror’s biases or a false sense of accuracy. As VR technology advances, the legal field faces the challenge of balancing innovation with impartiality, paving the way for standards that will determine the future role of VR in trials. By examining VR’s speculative and actual impacts in evidence presentation, we gain insight into how this technology could affect the legal landscape further.

I. What Is VR and How Does It Relate To Evidence?

In its broadest sense, VR is “a simulated three-dimensional (3D) environment that lets users explore and interact with a virtual surrounding in a way that approximates reality, as it’s perceived through the users’ senses. VR technology primarily utilizes headwear that covers your eyes completely, and you can see a three-dimensional immersive world in a 360-degree spherical field of view. While VR technology is gaining popularity, many people don’t use or come across it indaily life. While VR technology has been trendy for recreational use, such as VR video games, VR has been implemented in many professional settings for training, education, healthcare, retail, real estate, and more. The visual, auditory, and even tactile aspects of virtual reality, ranging from vibrations to even full-body haptic suits, allow the immersion to feel more ‘real’ and thus allow these practical applications.

These practical applications have led to speculation and interest in using VR technology in the legal field. One of the primary ideas is that jurors can “experience” scenes of the case rather than physically going there. Jurors have shown a desire to visit crime scenes in homicide cases when the scene itself is relevant to the conviction. VR technology can help overcome the hurdles of photographs or videos by virtually going to the scene, as juries can ‘virtually witness’ the scene and simulated events. The power of VR technology to transportjurors to the scene of the crime can also help make complex cases more understandable.

II. Evidentiary Concerns with VR Evidence

Before addressing VR evidence’s potential benefits and harms, it is necessary to consider its admissibility. Federal Rules of Evidence, such as hearsay and authentication, present unique challenges for the admissibility of VR evidence.

Under the Federal Rules of Evidence, hearsay is an out-of-court statement offered to prove the truth of the matter asserted. For example, if trying to prove person A loves VR, person B testifies that person A told them, “I love VR.” In this example, person B testifying to person A’s statement is hearsay. Hearsay is not admissible unless it meets an exemption or exception in the Federal Rules of Evidence. While the exact use of VR evidence containing out-of-court statements would vary on a case-by-case basis, a secondary purpose of VR presentation of evidence would be admissible not for the truth but to clarify other admissible evidence. For example, if there is an admissible recording of a witness describing a crime scene, VR evidence could help contextualize their testimony and immerse the jurors in the scene. In this case, the purpose wouldn’t be to prove the scene looked exactly as described or appeared in VR but to clarify the witness’s admissible testimony. While this may prevent VR demonstrations from jury deliberation, they can still be shown in the courtroom.

Another unique issue that comes with introducing VR presentation of evidence is authentication. According to the Federal Rules of Evidence, to introduce evidence, a proponent must produce evidence “sufficient to support a finding that the item is what the proponent claims it is.” This presents a unique problem for VR demonstrations because a proponent must show that the VR evidence is authentic. For example, with a photograph, a witness can authenticate it by testifying to taking the photo or confirming it accurately represents its contents. However, because VR is created as a simulation rather than a direct capture, VR wouldn’t be able to be authenticated the same way as a photograph. A proponent would rely onFederal Rule of Evidence 901(b)(9) for authentication. Because this rule alone would not be sufficient, a guideline for admitting VR evidence is that the proponent should “demonstrate that a qualified expert created the VR demonstration using an accurate program and equipment. The proponent should also show that all data used to create the demonstration was accurate and that no unfounded assumptions were made. Lastly, the proponent must present witness testimony to “verify the accuracy of the final product.”

III. Speculated and Actual Benefits of VR Evidence

As VR technology has become cheaper, more mainstream, and more widely used, it is being used in actual cases, making the potential for its wider use more achievable. One of the primary speculated benefits was the immersive nature of VR, allowing jurors to engage deeper with evidence by experiencing crime scenes and potentially re-creating events firsthand. Another speculated benefit is VR’s potential to “appeal to a jury’s emotional and subconscious” responses through its immersive nature.

Real-life implementations of VR evidence have already illustrated some of these benefits. One example is from Marc Lamber and James Goodnow, personal injury attorneys who have implanted VR technology in cases to “transport a jury to an accident scene.” Lamber and Goodnow work with engineers, experts, and production companies to recreate the scene where an injury or death occurred. This has allowed jurors to not only visualize the circumstances, events, and injury but also empathize deeper with the injured person’s suffering and aftermath of the incident. This ability to ‘transport’ the jurors to the scene can be incredibly impactful as it may be hard for jurors to visualize the scene in an isolated courtroom. One study in Australia focused on how VR can affect a jury’s ability to get the ‘correct’ verdict. Researchers, legal professionals, police officers, and forensic scientists simulated a hit-and-run scene in VR and photographs, then split jurors into groups to test the differences. This study found that VR technology required significantly less effort than photographs to construct a coherent narrative, leading jurors to reach the correct verdict 9.5 times more frequently than those who relied on photographs alone. The immersive technology also gave the jurors a better memory of the critical details of the case; the photograph group had difficulty visualizing the events of the case from the photographs alone. Researchers said this study was “unequivocal evidence that interactive technology leads to fairer and more consistent verdicts.”

IV. Speculated and Actual Harms of VR Evidence

While the immersive nature of VR technology has brought speculations about potential benefits for the legal field, concerns have emerged about possible harm or shortcomings of VR technology as evidence. The primary concerns are about potential biases and costs. 

 VR technology might cause jurors to impermissibly judge parties, especially defendants in criminal trials, differently according to underlying biases that they hold. One study found that mock jurors who used VR technology to understand a criminal trial were more likely to judge a black defendant more harshly than the white one. These studies used VR to simulate scenes from trials and, through computer generation, swapped out the races of the defendants and tested the differences in guilty verdicts and sentencing. Salmanowitz’s study found that using an avatar instead of accurate visual representations of the defendants can reduce implicit bias based on race. The avatars were shown by only the handheld controllers visible in the virtual space. The VR technology made no substantial difference in the jury’s decisions. However, a study by Samantha Bielen et al.found that jurors may be biased using VR against non-white defendants, finding non-white defendants were more likely to be found guilty on the same evidence as a white defendant when using the VR. 

The cost of VR also presents a barrier to implementing VR technology in courts. In the Australian study, a researcher noted that using VR as an evidentiary medium is “expensive, especially in remote locations, and in some cases, the site itself has changed, making accurate viewings impossible.”  VR technology is expensive, with even the cheapest consumer-grade headsets costing around $500. Further, digital recreation of the scene starts at $15,000 but can “go up to six figures depending on complexity.”

V. Conclusion

While the balance between the benefits and harms of introducing VR as a medium for evidence may vary greatly case-by-case, overall, the demonstrated advantages in improving a jurors’ factual understanding tend to outweigh the drawbacks. Although speculation is a natural reaction to new technologies, as VR finds real-world application in courtrooms, its tangible benefits and harms have been clarified. This allows revisiting the initial speculation and more effectively addressing this balance and admissibility concerns that accompany VR demonstrations use as evidence. Increased use and advancements in VR technology could amplify these benefits by increasing empathy and accuracy and tampering with the effects of emotional bias.  With this evolution in VR technology, the potential for an immersive yet balanced use of VR in the courtroom grows, offering an even greater ability for jurors to engage with evidence to enhance understanding, minimize bias, and support fairer, more informed verdicts.

I. INTRODUCTION

In May of 2024, the Federal Circuit overruled 40 years of precedent for assessing the obviousness of design patents in LKQ Corp. v. GM Global Technology Operations LLC. Already, commentators and practitioners have a wide array of opinions about the impacts of LKQ. If recent history is any guide, however, declarative statements about the impacts of LKQ are premature, and they create risks to businesses, practitioners, and courts alike. Rather, patent law observers should adopt a wait-and-see approach for the impacts of LKQ on design patent obviousness. 

II. THE LKQ DECISION 

In LKQ, the Federal Circuit addressed the standard for assessing design patent obviousness under 35 U.S.C. § 103. Before this decision, to find a claimed design unpatentable as obvious, the two-part RosenDurling test required a primary reference that was “basically the same” as the claimed design and secondary references “so related” to the primary reference that their features suggested combination with the primary reference. 

In this case, the Federal Circuit held that framework to be too rigid under the Patent Act. Instead, the court ruled that the obviousness of a claimed design is to be determined through the application of the familiar Graham four-part test used to assess the obviousness of utility patents. 

A.   EARLY OPINIONS ABOUT LKQ  

In the months since LKQ, opinions about the impacts of the decision have poured in from academics, practitioners, and commentators alike. Some predict a seismic shift, stating that the “far-” and “wide-reaching consequences” of LKQ will likely make design patents harder to obtain and easier to invalidate. Others predict little change at all, stating that the obviousness test “is largely the same as before” and that the expected changes from LKQ are primarily procedural. Still others seem to have landed on a middle ground, expecting “noticeable differences” in the law, with “examiners [having] more freedom to establish that the prior art is properly usable in an obviousness rejection.” 

B. PARALLELS WITH KSR 

LKQ is not the only recent decision dealing with obviousness that evoked immediate and wide-ranging reactions. In 2007, the Supreme Court issued KSR International Co. v. Teleflex Inc., a decision addressing the obviousness standard for patents. Notably, the Court rejected the Federal Circuit’s rigid application of its “teaching, suggestion, or motivation” test for obviousness to a utility patent in that case. 

In the immediate aftermath of that case, commentators and practitioners were “divided on whether the decision of the Supreme Court in KSR [was] (a) a radical departure from the Federal Circuit’s approach, or (b) unlikely to change much.” Even after the Federal Circuit began to issue decisions under KSR, some argued that the case had only a “modest impact” on the Federal Circuit, and others even questioned “whether the Supreme Court achieved anything in KSR other than giving the Federal Circuit a slap on the wrist.”

Experts were also divided on the likely business impacts of KSR in its immediate aftermath. In the summer after the decision came down, two distinguished patent law experts speaking on a panel were asked if KSR would drive up the cost of preparing and prosecuting a patent. One said yes, and the other said no. 

C. CAUTIONARY TALES FROM KSR 

As time went on, however, the impacts of KSR became clear. Empirical studies from years after the decision routinely proved that the impacts of KSR were anything but modest, contradicting “a commonly held belief that KSR did not change the law of obviousness significantly.” Various empirical studies revealed “strong evidence that KSR has indeed altered the outcomes of the Federal Circuit’s obviousness determinations,” “a remarkable shift in the Federal Circuit’s willingness to uphold findings of obvious below,” and that “the benefit of retrospection shows KSR did change the rate of obviousness findings.”

Thus, KSR should serve as a cautionary tale against jumping to conclusions about the impacts of obviousness decisions. In the months following KSR, any declarative statements about its impacts were mere speculation. Even after the Federal Circuit began issuing decisions under KSR, the sample size remained too small to draw conclusions. Only years after the decision could researchers illuminate the impacts of KSR through empirical studies and show which of those early opinions were right and wrong. 

III. THE WISDOM OF A WAIT-AND-SEE APPROACH FOR LKQ

Since the Federal Circuit only issued LKQ in May of 2024, we remain in the window where any declarative statements about its impacts are premature. Indeed, the Federal Circuit acknowledged that “there may be some degree of uncertainty for at least a brief period” in its LKQ opinion. While the urge to jump to conclusions is understandable, a wait-and-see approach offers many advantages. 

First, as KSR demonstrated, early predictions may be inaccurate and may influence practitioners to adopt misguided design patent prosecution strategies. Overstating the impacts of LKQ may lead to overly cautious design patent applications, leaving intellectual property unprotected. A wait-and-see approach will allow prosecution strategies to develop based on reliable trends, reducing the risk of costly errors. 

Second, the Federal Circuit almost certainly has more to say about design patent obviousness than it included in its LKQ opinion. Faulty strategy changes based on an incomplete picture may later need to be undone at great expense. Waiting allows the courts to solidify the impacts of LKQ so that practitioners and businesses can adjust their approaches – if that is necessary – with greater certainty and lower risk. 

Third, overreacting to speculative predictions could cause companies to shift their design-around strategies, leading to unnecessary and wasteful changes in product lines. A wait-and-see approach allows companies to maintain their creative momentum and keep their design strategies consistent until the impacts of LKQ are better understood. 

Fourth, design patents have experienced a boom in recent years. Premature predictions about LKQ risk skewing the perceptions of business leaders and the public about the continued value in pursuing design patent protections. By waiting to confirm the impacts of LKQ, commentators avoid this risk. 

Fifth, predictions about LKQ could become self-fulfilling prophecies. Widespread speculation could unintentionally influence how courts evaluate obviousness in future cases. A wait-and-see approach allows courts to evaluate obviousness free from the noise of speculative predictions, focusing exclusively on the application of the law to the facts of each case. 

Lastly, practitioners face potential backlash from clients if they offer advice that turns out to be too aggressive or pessimistic. By advocating patience to their clients, practitioners can maintain client trust and offer more measured and thoughtful advice once the implications of LKQ become clear. 

IV. WHEN WILL WE KNOW? 

This all begs the question: when will we understand LKQ so that declarative statements about its impacts are appropriate? Again, we can turn to KSR for guidance. 

More than a year after KSR was handed down, some were still questioning if the decision had any impact at all. The first empirical studies of its impacts seemed to emerge about two to three years after the decision, uniformly finding that it altered the law of obviousness. Therefore, it seems safe to assume that empirical studies will illuminate the impacts of LKQ in 2026. Until then, patent law observers should wait and see. 

V. CONCLUSION

With the recent history of KSR as our guide, patent law observers should adopt a wait-and-see approach for the impacts of the Federal Circuit’s recent decision in LKQ. At this early stage, improper speculation and declarative statements about the impacts of the case creates risks for businesses, practitioners, and courts. Instead, a wait-and-see approach allows reliable trends to guide prosecution strategies and allows design patent momentum to continue. In due time, empirical studies will emerge and make the impacts of LKQ clear to all. 

Trending Uses of Deepfakes

Deepfake technology, leveraging sophisticated artificial intelligence, is rapidly reshaping the entertainment industry by enabling the creation of hyper-realistic video and audio content. This technology can convincingly depict well-known personalities saying or doing things they never actually did, creating entirely new content that did not really occur. The revived Star Wars franchise used deepfake technology in “Rogue One: A Star Wars Story” to reintroduce characters like Moff Tarkin and Princess Leia, skillfully bringing back these roles despite the original actors, including Peter Cushing, having passed away. Similarly, in the music industry, deepfake has also been employed creatively, as illustrated by Paul Shales’ project for The Strokes’ music video “Bad Decisions.”Shales used deepfake to make the band members appear as their younger selves without them physically appearing in the video.

While deepfakes offer promising avenues for innovation, such as rejuvenating actors or reviving deceased ones, it simultaneously poses unprecedented challenges to traditional copyright and privacy norms.

Protections for Deepfakes

Whereas deepfakes generate significant concerns, particularly about protecting individuals against deepfake creations, there is also controversy over whether the creators of deepfake works can secure copyright protection for their creations.

Copyrightability of Deepfake Creations

Current copyright laws fall short in addressing the unique challenges posed by deepfakes. These laws are primarily designed to protect original works of authorship that are fixed in a tangible medium of expression. However, they do not readily apply to the intangible, yet creative and recognizable, expressions that deepfake technology replicates. This gap exposes a crucial need for legal reforms that can address the nuances of AI-generated content and protect the rights of original creators and the public figures depicted.

Under U.S. copyright law, human authorship is an essential requirement for a valid copyright claim. In the 2023 case Thaler v. Perlmutter, plaintiff Stephen Thaler attempted to register a copyright for a visual artwork produced by his “Creativity Machine,” listing the computer system as the author. However, the Copyright Office rejected this claim due to the absence of human authorship, a decision later affirmed by the court. According to the Copyright Act of 1976, a work must have a human “author” to be copyrightable. The court further held that providing copyright protection to works produced exclusively by AI systems, without any human involvement, would contradict the primary objectives of copyright law, which is to promote human creativity—a cornerstone of U.S. copyright law since its beginning. Non-human actors need no incentivization with the promise of exclusive rights, and copyright was therefore not designed to reach them.

However, the court acknowledged ongoing uncertainties surrounding AI authorship and copyright. Judge Howell highlighted that future developments in AI would prompt intricate questions. These include determining the degree of human involvement necessary for someone using an AI system to be recognized as the ‘author’ of the produced work, the level of protection afforded the resultant image, ways to assess the originality of AI-generated works based on non-disclosed pre-existing content, the best application of copyright to foster AI-involved creativity, and other associated concerns.

Protections Against Deepfakes

The exploration of copyright issues in the realm of deepfakes is partially driven by the inadequacies of other legal doctrines to fully address the unique challenges posed by this technology. For example, defamation law focuses on false factual allegations and fails to cover deepfakes lacking clear false assertions, like a manipulated video without specific claims. Trademark infringement, with its commercial use requirement, does not protect against non-commercial deepfakes, such as political propaganda. The right of publicity laws mainly protect commercial images rather than personal dignity, leaving non-celebrities and non-human entities like animated characters without recourse. False light requires proving substantial emotional distress from misleading representations, a high legal bar. Moreover, common law fraud demands proof of intentional misrepresentation and tangible harm, which may not always align with the harms caused by deepfakes. 

Given these shortcomings, it is essential to discuss issues in other legal areas, such as copyright issues, to enhance protection against the misuse of deepfake technology. In particular, the following sections will explore unauthorized uses of likeness and voice and the impacts of deepfakes on original works. These discussions are critical because they aim to address gaps left by other legal doctrines, which may not fully capture the challenges posed by deepfakes, thereby providing a broader scope for protection. 

Unauthorized Use of Likeness and Voice

Deepfakes’ capacity to precisely replicate an individual’s likeness and voice may raise intricate legal issues. AI-generated deepfakes, while sometimes satirical or artistic, can also be harmful. For example, Taylor Swift has repeatedly become a target of deepfakes, including instances where Donald Trump’s supporters circulated AI-generated videos that falsely depict her endorsing Trump and participating in election denialism. This represents just one of several occasions where her likeness has been manipulated, underscoring the broader issue of unauthorized deepfake usage.

The Tennessee ELVIS Act updates personal rights protection laws to cover the unauthorized use of an individual’s image or voice, adding liabilities for those who distribute technology used for such infringements. In addition, on January 10, 2024, Reps. María Elvira Salazar and Madeleine Dean introduced the No Artificial Intelligence Fake Replicas And Unauthorized Duplications (No AI FRAUD) Act (H.R. 6943). This bill is designed to create a federal framework to protect individual rights to one’s likeness and voice against AI-generated counterfeits and fabrications. Under this bill, digitally created content using an individual’s likeness or voice would only be permissible if the person is over 18 and has provided written consent through a legal agreement or a valid collective bargaining agreement. The bill specifies that sufficient grounds for seeking relief from unauthorized use include financial or physical harm, severe emotional distress to the content’s subject, or potential public deception or confusion. Violations of these rights could lead individuals to pursue legal action against providers of “personalized cloning services” — including algorithms and software primarily used to produce digital voice replicas or depictions. Plaintiffs could seek $50,000 per violation or actual damages, along with any profits made from the unauthorized use.

Impact on Original Work

The creation of deepfakes can impact the copyright of original works. It is unclear whether deepfakes should be considered derivative works or entirely new creations.

In the U.S., a significant issue is the broad application of the fair use doctrine. Under § 107 of the Digital Millennium Copyright Act of 1998 (DMCA), fair use is determined by a four-factor test assessing (1) the purpose and character of the use, (2) the nature of the copyrighted work, (3) the amount and substantiality of the portion used, and (4) the impact on the work’s market potential. This doctrine includes protection for deepfakes deemed “transformative use,” a concept from the Campbell v. Acuff Rose decision, where the new work significantly alters the original with a new expression, meaning, or message. In such cases, even if a deepfake significantly copies from the original, it may still qualify for fair use protection if it is transformative, not impacting the original’s market value.

However, this broad application of the fair use doctrine and liberal interpretation of transformative use do not work in favor of the original creators. They may potentially protect the deepfake content even with malicious intent, which makes it difficult for original creators to bring claims under § 512 of the DMCA and § 230 of the Communication Decency Act.

Federal and State Deepfake Legislation

Copyright is designed to adapt with the times.” At present, although the United States lack comprehensive federal legislation that specifically bans or regulates deepfakes, there are still several acts that target deepfakes. 

In Congress, a few proposed bills aim to regulate AI-generated content by requiring specific disclosures. The AI Disclosure Act of 2023 (H.R. 3831) requires any content created by AI to include a notice stating, “Disclaimer: this output has been generated by artificial intelligence.” The AI Labeling Act of 2023 (S. 2691) also demands a similar notice, with additional requirements for the disclaimer to be clear and difficult to alter. The REAL Political Advertisements Act (H.R. 3044 and S. 1596) demands disclaimers for any political ads that are wholly or partly produced by AI. Furthermore, the DEEPFAKES Accountability Act (H.R. 5586) requires that any deepfake video, whether of a political figure or not, must carry a disclaimer. It is designed to defend national security from the risks associated with deepfakes and to offer legal remedies to individuals harmed by such content. The DEFIANCE Act of 2024 aims to enhance the rights to legal recourse for individuals impacted by non-consensual intimate digital forgeries, among other objectives.

On the state level, several states have passed legislation to regulate deepfakes, addressing various aspects of this technology through specific legal measures. For example, Texas SB 751 criminalizes the creation of deceptive videos with the intent to damage political candidates or influence elections. In Florida, SB 1798 targets the protection of minors by prohibiting the digital alteration of images to depict minors in sexual acts. Washington HB 1999 provides both civil and criminal remedies for victims of fabricated sexually explicit images. 

This year, California enacted AB 2839, targeting the distribution of “materially deceptive” AI-generated deepfakes on social media that mimic political candidates and are known by the poster to be false, as the deepfakes could mislead voters. However, a California judge recently decided that the state cannot yet compel individuals to remove such election-related deepfakes, since AB 2839 facially violates the First Amendment. 

These developments highlight the diverse strategies that states are employing to address the challenges presented by deepfake technology. Despite these efforts, the laws remain incomplete and continue to face challenges, such as concerns over First Amendment rights.

Conclusion

As deepfake technology evolves, it challenges copyright laws, prompting a need for robust legal responses. Federal and state legislation is crucial in protecting individual rights and the integrity of original works against unauthorized use and manipulation. As deepfake technology advances, continuous refinement of these laws will be crucial to balance innovation with ethical and legal boundaries, ensuring protection against the potential harms of deepfakes.

Introduction 

Algorithmic bias is AI’s Achilles heel, revealing how machines are only as unbiased as the humans behind them. 

The most prevalent real-world stage for human versus machine bias is the job search process. What started out as newspaper ads and flyers at local coffee shops, is now a completely digital process with click-through ads, interactive chatbots, resume data translation, and computer-screened candidate interviews. 

Artificial intelligence encompasses a wide variety of tools, but in context to HR specifically, common AI tools include Machine Learning algorithms that conduct complex and layered statical analysis modeling human cognition (neural networks), computer vision that classifies and labels content on images or video, and large language models. 

AI-enabled employment tools are powerful gatekeepers that determine the future of natural persons. With over 70% of companies using this technology investing into the promise of efficiency and neutrality, these abilities have recently come into question as these technologies have the potential to discriminate against protected classes. 

Anecdote 

On February 20, 2024, Plaintiff Derek Mobley initiated a class action lawsuit against an AI-enabled HR organization WorkDay, Inc., for engaging in a “pattern and practice” of discrimination based on race, age, and disability in violation of the Civil Rights Act of 1964, Civil Rights Act of 1886, Age Discrimination Act of 1967, and ADA Amendments Act of 2008. WorkDay Inc., according to the complaint, disproportionately disqualifies African-Americans, individuals over the age of 40, and individuals with disabilities securing gainful employment. 

WorkDay provides subscription-based AI HR solutions to medium and large sized firms in a variety of industries. The system screens candidates based on human inputs and algorithms and according to the complaint, WorkDay employs an automated system, in lieu of human judgement, to determine how high volume of applicants should be processed on behalf of their business clients. 

The plaintiff and members of the class have applied for numerous jobs that use WorkDay’s platforms and received several rejections. This process has deterred the plaintiff and members of the class from applying to companies that use WorkDay’s platform.

Legal History of AI Employment Discrimination 

Mobley vs. WorkDay is the first-class action lawsuit against an AI solution company for employment discrimination, but this is not the first time that an AI organization is being sued for employment discrimination. 

In August 2023, the EEOC settled the first-of-its-kind Employment Discrimination lawsuit against a virtual tutoring company that programmed its recruitment software to automatically reject older candidates. The company was required to pay $325,000 and if they were to resume hiring efforts in the US, they are required to call back all applicants during the April-May 2020 period who were rejected based on age to re-apply. 

Prior to this settlement, the EEOC issued guidance to employers about their use of Artificial Intelligence tools that extends existing employer selection guidelines to AI-assisted selections. From this guidance, employers, not third-party vendors, ultimately bear the risk of unintended adverse discrimination from such tools.

How Do HR AI Solutions Introduce Bias?

There are several steps in the job search process and AI is integrated throughout this process. Steps include: The initial search, narrowing candidates, and screening.  

Initial search

The job search process starts with targeted ads reaching the right people. Algorithms in hiring can steer job ads towards specific candidates and help assess their competencies using new and novel data. HR professionals found the tool helpful in drafting precise language and designing the ad with position elements, content and requirements. But these platforms can inadvertently reinforce gender and racial stereotypes by delivering the ad to candidates that meet certain job stereotypes.

For instance, ads delivered on Facebook for stereotypically male jobs are overwhelmingly targeted at male users even though the advertising was intended to reach a gender neutral audience. Essentially, at this step of the job search process, algorithms can prevent capable candidates from even seeing the job posting in the first place that further creates a barrier to employment. 

Narrowing Candidates

After candidates that have viewed and applied for the job through an ad or other source, the next step that AI streamlines is the candidate narrowing process. At this step, the system narrows candidates by reviewing resumes that best match the historical hiring data from the company or its training data. Applicants found the resume to application form data transfers helpful and accurate in this step of the process. But applicants were concerned that the model could miss necessary information.

From the company’s perspective, hiring practices from the client company are still incorporated into the hiring criteria in the licensed model. While the algorithm is helpful in parsing vast amounts of resumes and streaming this laborious process for professionals, the algorithm can replicate and amplify existing biases in the company data.

For example, a manager’s past decisions may lead to anchoring bias. If some bias like gender, education, race and age existed in the past and they are present in the employer’s current high performing employees that the company uses as a benchmark, those biases can be incorporated into the outcomes at this stage of the employment search process. 

Screening

Some organizations subscribe to AI tools that have a computer vision-powered virtual interview process that analyzes the candidates’ expressions to determine whether they fit the “ideal candidate” profile, while other tools like behavior/skills games are used to screen candidates prior to an in-person interview. 

Computer vision models that analyze candidate expressions to assess candidacy are found to perpetuate preexisting biases against people of color. For instance, a study that evaluates such tools, found the taxonomies of social and behavioral components creates and sustains similar biased observations that one human would make on an another because the model with those labels and taxonomies is trained with power hierarchies. In this sense, computer vision AI hiring tools are not neutral because they reflect the humans that train and rely on them. 

Similarly, skill games are another popular tool used to screen candidates. However, there are some relationships AI cannot perceive in its analysis. For instance, candidates that are not adept with online games perform poorly on those games, not because they lack the skills, but they lack an understanding of the games features. Algorithms, while trained on vast data to assess candidate ability, still fall short when it comes assessing general human capabilities like the relationship between online game experience and employment skills tests. 

Throughout each step of the employment search process, AI tools fall short in accurately capturing candidate potential capabilities.

Discrimination Theories and AI

Given that the potential for bias is embedded throughout the employment search process, legal scholars speculate courts are more likely to scrutinize discriminatory outcomes under the disparate impact theory of discrimination.

As a recap, under Title VII there are two theories of discrimination, disparate treatment, and disparate impact. Disparate treatment means the person is treated different “because of” their status as a protected class (i.e., race, sex). For example, if a manager were to intentionally use a bias algorithm to screen out candidates of a certain race, then this behavior would be considered disparate treatment. Note, this scenario is for illustrative purposes only. 

Disparate impact applies to facially neutral processes that have a discriminatory effect. The discriminatory effect aspect of this theory of discrimination can be complex because the plaintiff must identify the employer practice that has a disparate impact on a protected group. The employer can then defend that the practice by showing it is “job related” and consistent with “business necessity.” However, the plaintiff can still show that there was an alternative selection process and the business failed to adopt it. Based on this disparate impact theory, it is possible that when AI selection tools disproportionately screen women and/or racial minorities from the applicant pool, disparate theory could apply. 

Existing Methods to Mitigate Bias 

Algorithmic bias in AI tools has serious implications for members of protected classes. 

However, developers currently employ various tools to de-bias algorithms and improve their accuracy. One method is de-biased word embedding in which neutral associations of a word are supplemented to expand the model’s understanding of the word. For instance, a common stereotype is men are doctors and women are nurses or in algorithmic terms “doctor – man + woman = nurse.” However, with the de-bias word embedding process, the model is then trained to understand “doctor – man + woman = doctor.” 

Another practice currently employed by OpenAI is external Red Teaming. In which external stakeholders interact with the product and assess its weaknesses, possibility for bias, or other adverse consequences and provide feedback to OpenAI to improve and mitigate the onsets of adverse events. 

But there are limitations to these enhancements. To start, bias mitigation is not a one-size-fits-all issue. Bias is specific to its geographic and cultural bounds. For instance, a model in India may need to consider caste-based discrimination. Additionally, precision is required to capture the frame where bias is possible and solely relying on foreseeable bias from the developers’ perspective is limiting. Rather, employing some form of collaborative design that includes relevant stakeholders to contribute to the identification of bias, the identification of not biased is needed.

Lastly, a debiased model is not a panacea. A recent study in which users interacted with a debiased model that used machine learning and deep learning to provide recommendations for college majors, indicated that regardless of the debiased model’s output, users relied on their own biases to choose their majors, often motivated by gender stereotypes associated with those majors. 

Essentially, solutions from the developer side are not enough to resolve algorithmic bias issues. 

Efforts to Regulate AI Employment Discrimination

Federal law does not specifically govern artificial intelligence. However, existing laws including Title VII extend to applications that include AI. At this point, regulation efforts are largely at the state and local government level. 

Additionally, the EEOC’s Employer Guidelines is a start to shifting the onus on to employers to investigate the capabilities and outcomes of the technologies incorporated into their hiring practices.

New York City is the first local government to pass an official law that regulates AI-empowered employment decision tools. The statute requires organizations to inform candidates of the use of AI in their hiring process, and before using the screening device, notify potential candidates. If candidates do not consent to the AI-based process, the organization is required to use an alternative method. 

Like New York’s statute, Connecticut passed a statute specific to state agency’s use of AI and machine learning hiring tools. Connecticut requires an annual review of the tools performance, a status update on whether the tool underwent some form of bias mitigation training in an effort to prevent unlawful discrimination. 

New JerseyCalifornia, and Washington D.C. currently have bills that are intended to prevent discrimination with AI hiring systems. 

Employer Considerations

With the possibility of bias embedded throughout each step of the recruiting process, employers must do their part to gather information about the performance of the AI system they ultimately invest in. 

To start, recruiters and managers alike stressed the need for AI systems to provide some explanation why the applicant is rejected or selected to accurately assess the performance of the model. This need speaks specifically to AI models’ tendency to find proxies or shortcuts in the data to reach the intended outcome on a superficial level. For instance, models might find a candidate by only focusing on candidates who graduated from universities in the Midwest if most of upper management attended such schools. In this sense, employers should look for accuracy reports, ask vendors ways to identify and correct this issue in this hiring pool.

Similarly, models can focus on candidate traits that are unrelated to the job traits and are simply unexplained correlations. For example, one model in the UK linked people that liked “curly fries” on Facebook to have higher levels of intelligence. In this case, employers need to develop processes to analyze whether the output from the model was “job related” or related carrying out the functions of the business. 

Lastly, employers must continue to invest in robust diversity training. Algorithmic bias reflects the bias human behind the computer. While AI tools enhance productivity and alleviate the laborious parts of work, it also increases the pressure on humans to do more cognitive-intensive work. In this sense, managers need robust diversity training to scrutinize outputs from AI models, to investigate whether the model measured what it was supposed to, whether the skills required in the post accurately reflect the expectations and culture of the organization. 

Along with robust managerial training, these AI solutions often incorporate “culture fit” as a criterion. Leaders need to be intentional about precisely defining culture and promoting that defined culture in its hiring practices. 

Conclusion

A machine does not know its output is biased. Humans interact with context—culture dictates norms and expectations, shared social/cultural history informs bias. Humans, whether we like to admit it or not, know when our output is biased. 

To effectively mitigate unintentional bias in AI-driven hiring, stakeholders, ranging from HR professionals to developers and candidates, must understand the technology’s limitations, ensure its job-related decision-making accuracy, and promote transparent, informed use, while also maintaining robust DEI initiatives and awareness of candidates’ rights.