TestGorilla

1. The organization

For this data-ethical consultation project I will be looking at the company TestGorilla (https://www.testgorilla.com/). For this data-ethical consultation I am evaluating the website and all public information channels of the company. TestGorilla is a company founded in the Netherlands that develops pre-employment testing software. Using this test software, pre-employment assessments can be used to identify the most suitable candidates for a job position. According to TestGorilla, their software helps their customers: save time, make hiring decisions more objective and avoid costly mis-hires. In many hiring processes, you first apply with a CV and cover letter, when these are of a sufficient level you get invited for an assessment to test your capabilities. The software of TestGorilla actually replaces the CVs with tests that measure aptitude, practical skills and motivation. The idea behind this approach is that the best way to predict job success for candidates is to look at a combination of different skill tests into one assessment.

Companies can use the software of TestGorilla to change their entire hiring process. The first step is that they offer seriously smart screening tests crafted by experts. For specific jobs, the recruiter can select tests and add their own personal questions as well. The assessment can then be sent to applicants who fill in the test. The software ranks the top candidates based on the data of the tests. AI has been used in hiring processes on quite a wide scale now, but this is mostly to screen CV's. TestGorilla uses AI technologies to create a more direct and thus less prone to bias method to test the quality of an applicant.
automated candidate screening

2. The AI technologies Employed

TestGorilla is a software company and uses AI algorithms for a wide range of tasks. AI is used for the three main components of the business process of TestGorilla: developing the tests, creating an assessment by combining several tests, processing the results of the tests.

Starting with the development of the tests. TestGorilla asks subject-matter experts (e.g. Java coding) to create a test from the ground up and launch it on the website. Once the test is launched, participants will start doing the test and thus collecting data on that specific test. Once a test is live, TestGorilla uses AI to test the actual test. After a test is created by an expert and launched on the website, an advanced algorithm provides an analysis that shows where they can improve the tests further. This algorithm is used to identify key indicators of someone's qualifications in a test and uses measures such as "Cronbach's alpha" to help them track and confirm the internal consistency of the questions in a test.

Secondly, they use AI to recommend the recruiter specific tests based on the job title for which people apply. Since they have a large database of tests and to make the process for the hirer easier, this recommendation system can come in handy. This algorithm is based on key words in the job title and trained on previous assessment for similar positions/jobs.

Finally, TestGorilla also uses AI to process the data of the applicants and to give insights into their capabilities. These report consists of four main components. Firstly, a summary of the test results, including extra information about what scores etc, to protect the integrity of the tests the answers to questions are not provided. The scores of the test are created using AI. Secondly, they consist of anti-cheating alerts. Finally it consists of the written out answers of the unique questions and a space for the overall rating, to be filled in by the hirer itself.

3. Ethical concerns

TestGorilla uses AI technologies for a wide range of tasks, in every step of their business process. Based on their website and public information channels, they are very aware of the ethical implications of using such technologies for hiring employees. TestGorilla argues that their method of evaluating candidates is more objective and less throne to bias. Although they pay a lot of attention to the fairness of the product, I have identified three potential points of concern related to the use of AI technologies.

Potential Bias

AI has been used as a tool to assist in the hiring process for quite some time now and this has raised some ethical criticism. Since these algorithms are often trained on hiring decisions in the past, biases in these decisions are adapted and sometimes even by expanded by the AI. An example of this is Amazon algorithm that screened CVs, the machine-learning specialists discovered a serious problem: the machine did not like women. Since in the patters in the past, way more men applied and thus way more were hired, the Amazon system that was trained on this data taught itself that male candidates were preferable. Reducing hiring bias could have very substantial benefits such as improving team performance and increasing profits. Here I identify between two broad categories of bias in hiring. Firstly, hiring bias is the result of specific preferences unequally divided over applicants. Secondly, process bias which occurs when the hiring process bolsters existing biases in society.

According to their website, minimizing bias and hiring Discrimination in the application process is one of the key values of TestGorilla. They already spend a lot of effort to pursue this. Firstly, by following the laws enforced by the Equal Employment Opportunity Commission (EEOC) and by using the UGESP framework of the US Department of Labor's skills analyses, to verify that every test is valid.

However, recruiters do often not realise that they are discriminating, which is the result of unconscious bias Since the test of TestGorilla are more objective and form a rating of applicants based on actual skill and not the image portrayed in the head of the recruiter, the use of TestGorilla software could definitely reduce racial, gender or age bias. However, I argue that process bias, which refers to the situation where the hiring process itself props up biases that exist in society. Since many of the test used by TestGorilla test a skill and not per se talent or predisposition for that skill, existing inequalities in society can still be continued in the hiring process. People that are well off can for example visit expensive coding schools to develop their coding skills, while someone with less money cannot. Even when that second person has more talent and could be a better fit for the position, based on the TestGorilla test the first is more likely to be hired. Furthermore, as I explained before, in the report of the applicants, the recruiter is still able to give the final rating for an applicant itself. Therefore, still some hiring bias could exist when using TestGorilla software.
⇒ inequality

Explainability

explainability
into the decision-making process is an important ethical value in all different kinds of AI applications and this is also the case for hiring assessments such as developed by TestGorilla. While the previous point of concern, potential bias or unfairness of such systems, has received a lot of attention in academic research, there has been less work on the role of explainability
in this process. However, I would argue explainability is also very important for AI systems in the hiring process. For both applicants and recruiters it must be possible to gather an explanation for the outcome of a test. As Franchili and Glasser (2022) argue it most be possible to describe, in normal language and accessible formats, the traits the AI assessment tool is designed to test, the method by which those traits will be assessed, and the variables or factors that may affect the assessment or rating.

Looking at this proposition for explainability by Franchili and Glasses and applying it on the practises of TestGorilla, I find some insights. Firstly, for every test TestGorilla clearly explains in normal language what traits it test and what method is used to test it. TestGorilla pays a lot of attention to continuously check that the test are actually valid and test the traits they are supposed to do. So for the first two kinds of explainability TestGorilla does a good job. However, the last type, explainability of the factors that may affect the rating, this is not always very clear. As discussed before, in the test report that is delivered to the recruiter, TestGorilla automatically calculates a score for each test and the recruiter cannot see the answers of the participant. Thus it is not quite clear for both the recruiter and the applicant on what basis this score is exactly calculated. Since this rating plays a critical role in the hiring process and is currently not explainable, this is a factor TestGorilla could improve on. I understand it is not possible to show the correct and wrong answers for all the tests, but at least some identification on what grounds the score is determined could provide the stakeholders a form of explanation .

privacy

Besides the ethical implications of explainability and bias in processing the data, there are also potential concerns about the collecting of data for this task. Data of participants is very personal and it should be treated very carefully, since any form of privacy breach could have severe implications. For example, the data of one bad application should not fall in the hands of other recruiters or be shared publicly because this could negatively influence future hiring opportunities. TestGorilla realises that they gather valuable personal data and that it is vital to protect the privacy of its users.

On their website, TestGorilla provides an extensive description of their privacy policy. I will discuss some things that seemed remarkable to me here. Firstly, TestGorilla is very transparent on what kind of information they collect for every stakeholder (test taker, recruiter, test developer). All the types of software used to collect data on the usage pattern are listed and discussed. Furthermore, they also present an overview of how the data that is collected is used and for what reason it is used. Some types of data are shared with others. What I found remarkable here is that when TestGorilla is sold to another company, the data of users is shared with that company. However, users of the platform also have an extensive set of rights that are in accordance with European Union privacy laws, such as access, consent, restriction and objection. Also the right to be forgotten is included in the policies of TestGorilla.

Besides data on the test and demographics of users, TestGorilla also collects very sensitive data in the form of video webcam data. This data is collected to ensure participants do not cheat or cooperate in any way. Webcam pictures taken as an anti-cheating measure are retained for 6 months. Video recordings of test-takers answering custom questions are retained for 2 years. Given the sensitivity of this data I would argue this time is very long. I would argue that in when the evaluation of a participant is completed, such sensitive data as video and picture data is deleted, since it does not serve a goal anymore anyway.

4. Recommendations

To address the potential concerns highlighted above I have the following recommendations for the use of data and AI by TestGorilla. I think it is safe to say that for TestGorilla respecting ethical values and implications has a very high priority and forms one of the key components of the company. They are aware of most risks that are involved when using AI technologies to help assist in the hiring process. The most discussed concern for algorithms developed to assist in the hiring process is potential bias. By creating a new form of testing that replaces CVs, TestGorilla provides a new hiring tool that is objective and aims to prevent bias. However, as I argued here, TestGorilla still has to be aware that hiring bias could still occur. For the second concern, explainability, I have a more straightforward recommendation. Since it is ethically speaking preferable that users can have an explanation of the rating they receive. It is understandable that TestGorilla cannot provide all the answers but some form of explainability
of the results would be beneficial. Finally, for the privacy concern, it is clear that TestGorilla understands the importance of securing privacy of the sensitive data that it collects. They are very transparent about all their data practices, how these are collected and processed. Only for the video data I would argue that the data is stored for an unnecessary amount of time, given the fact that this type of data is such an invasion of privacy.
right to erasure