Raffaella Aghemo, Lawyer
Candidate recruitment platforms and a recent class action
Nowadays, algorithmic systems and platforms are increasingly being used to screen and select new staff. We are all now aware that the data on which these systems are trained – being historical, i.e. based on past human behaviour – could trigger mechanisms that perpetuate inequalities, disparities or even the complete exclusion of entire sections of the population. These are known as algorithmic biases, which, when based on poorly configured training or misguided instructions from the companies using these systems, often result in ‘a priori exclusions’. Such incidents can, and often do, trigger class actions, which can have a significant impact on a company’s reputation as well as its financial assets.
Algorithmic CV screening is a recruitment process that uses artificial intelligence and machine learning algorithms to automatically analyse, filter and evaluate received CVs, identifying the most suitable candidates for a specific job role.
The latest in a long series is the class action filed in January this year in the case of Kistler et al. against Eightfold AI Inc., filed with the Superior Court of Contra Costa County, California, in which, for the first time, a new approach is evident, one that does not focus on the biases and partiality of artificial intelligence, but rather on the ‘opacity’ of the platform in collecting data on potential candidates. That is, not on errors in the training data, but on the inadequacy of the entire process. Specifically, the complaint alleges a breach of the Fair Credit Reporting Act and unfair competition, as stated on the cover page of the document.
The case in question was also brought by the non-profit organization Towards Justice, which explicitly accuses Eightfold AI Inc. of violating federal and state privacy laws by collecting, analyzing and evaluating job applicants without their knowledge or consent. Eightfold is a Delaware corporation with its principal place of business in Santa Clara, California. “The plaintiffs, Erin Kistler and Sruti Bhaumik, on their own behalf and on behalf of all others in a similar situation, state as follows:… This case concerns the way in which the defendant, Eightfold AI Inc., uses hidden Artificial Intelligence technology to collect sensitive and often inaccurate information about unsuspecting candidates and to assign them a score from 0 to 5 for potential employers based on their alleged ‘likelihood of success’ in the job. Eightfold’s technology lurks within the feedback provided on the applications of thousands of candidates who may not even be aware of Eightfold’s existence, let alone the fact that Eightfold is collecting personal data—such as social media profiles, location data, internet and device activity, cookies and other tracking data—to create a profile of the candidate’s behaviour, attitudes, intelligence, aptitudes and other characteristics of the candidate that applicants have never included in their application.”
And further: ‘Specifically, this class action stems from Eightfold’s unlawful practice of collecting, compiling and evaluating information on job applicants using opaque machine learning processes and closely guarded algorithms, which produce unverifiable reports that a growing number of employers rely on for employment decisions, such as hiring…… For the job applicants concerned, this process is often largely invisible… Eightfold processes this information through its own large language model (‘LLM’), which, according to Eightfold, incorporates ‘over 1.5 billion global data points’, including ‘more than 1 million job roles, 1 million skills and the profiles of over 1 billion people working in every role, profession [and] sector”, making it the “world’s largest source of talent data, which updates automatically”…“ analysing data” from “public sources such as career sites, job boards and CV databases (LinkedIn, Hoovers, Crunchbase, GitHub, etc.)”, as well as other sources of information such as Eightfold’s internal database containing job descriptions and candidate data.”
What makes the situation even worse is the fact that Eightfold manages a growing client list of over 100 companies, including Microsoft, Morgan Stanley, Starbucks, BNY, PayPal, Chevron and Bayer.
Lack of transparency in the company’s operations
The first claimant, Ms Kistler, holds a degree in computer science from Ohio State University and has around 19 years’ professional experience in product management and project management, including in the fields of AI, data implementation, technology and marketing, aimed at enhancing the user experience across a variety of sectors, including government services, business, media and entertainment. From 2022 to the present, she has applied for several positions for which she is qualified, with PayPal, Microsoft, Netflix and other employers. During the online application process, she did not consent to waiving her privacy and consumer protection rights under state and federal laws, nor did she receive a stand-alone notice providing a clear and conspicuous warning that, for the purposes of assessing her job application, a ‘profile’ would be created based on her personal data.
Like others, she was not invited for an interview and was not made a job offer for any of the positions she had applied for through Eightfold.
The regulatory function of the FCRA
The problem of employers relying on secret and unreliable third-party reports (or ‘dossiers’) when making employment decisions was a key concern that Congress sought to address through the enactment of the Fair Credit Reporting Act, which, at this point, applies not only to consumer credit reports but to all areas in which information on an individual’s habits, morals and life experiences of an individual can have devastating consequences for people’s employment.
To promote proper and due transparency, particularly in these areas, the FCRA requires reporting agencies such as Eightfold to make certain disclosures, obtain certain certifications and ensure that consumers (in this case, job applicants) have a mechanism to review and correct the reports provided to potential employers for the purpose of determining suitability for employment.
Page 9 of the filed document contains an important passage: “With the trend towards the computerization of invoicing and the creation of all manner of computerized databases, the individual runs a serious risk of seeing his life and reputation reduced to impersonal ‘signals’ and punch holes in an impassive, thoughtless machine, which can literally ruin his reputation for no reason, render him unemployed or uninsurable, and deny him the opportunity to obtain a mortgage to buy a home. We are not so much concerned about the possible erroneous rejection of a consumer for a luxury item as we are about the possible destruction of their good name without their knowledge and for no reason. The loss of a credit card can certainly be costly, but, as Shakespeare said, the loss of one’s good name is priceless and truly impoverishes one.”
How the Eightfold platform works
When it comes to recruitment, Eightfold’s “Talent Intelligence Programme” enables employers to create job profiles, gather internal and external data on candidates, and generate web pages where candidates can apply. The assessment tools then evaluate and rank candidates using data collected during the application process, the employer’s internal data, external data, and the model—trained on billions of data points—with the specific aim of generating a report that Eightfold provides to employers for use in recruitment decisions.
The report that the Talent Intelligence Programme produces on candidates includes what Eightfold calls the ‘Match Score’, ‘an artificial intelligence model that predicts the match between a candidate’s profile and a job role, and presents candidates for a specific role in the form of a ranking’.
The data that Eightfold collects and evaluates includes, amongst other inputs,
- the candidate’s recruitment profile and CV,
- additional data on the candidate collected from public sources regarding their professional history (such as blogs, publications, conferences, application history, etc.),
- data relating to comparable employees at other companies,
- predictions and inferences regarding the candidate’s personality and future, and
- data used to train Eightfold’s AI.
In its privacy policy, Eightfold specifies that the: “personal data’ it collects and uses for employment assessment includes ‘[i]nferences drawn from any of the [various sources of information it collects] to create a profile of the [job applicant] that reflects the applicant’s preferences, characteristics, predispositions, behaviour, attitudes, intelligence, skills and aptitudes.’
The images below describe the algorithmic process used by the company, at the end of which the above is translated into a match score, ranging from 0 to 5 in increments of 0.5, which can then be sorted into a ranked list.