Navigating AI Ethics: How Phenom Upholds AI Compliance and Legislation

How Phenom Promotes AI Safety and Regulatory Compliance

AI compliance in technology can be split into three categories:

1. Employment-based compliance

2. AI-specific compliance

3. Privacy-based compliance

The most prominent AI compliance laws have been passed in New York and initiated in California, with other states attempting to follow suit. New York City Local Law 144 (NYC 144), which took effect in July 2023, requires that employers perform a bias audit if they are using automated employment decision tools (AEDTs) to guide their hiring decisions. California's AI Bill of Rights proposes to audit both the AEDT and all other parts of the decision-making process that come with hiring a candidate.

At Phenom, we’ve taken these laws into account and made many innovations in our AI compliance, such as the ability to:

• Enable or disable the use of Fit Score to meet regulation at a company’s jurisdiction

• Track and record anyone who has a Fit Score and has viewed a job profile, with an opt-out feature so applicants can decide whether they're okay with having an AI-based hiring tool assess their application

• Audit internal and external applications of Fit Score across employers, job families, individual jobs, or protected classes

To maintain compliance, it’s essential to follow the Uniform Guidelines on Employee Selection Procedures (UGESP). These guidelines have been developed by Industrial Organizational (IO) psychologists and employment lawyers to ensure that hiring is both valid and fair.

Phenom follows these guidelines and continues to promote AI safety and regulatory compliance by leveraging:

• Human-in-the-Loop Control and Data Annotation: Making sure a human is involved in all feedback loops or decisions helps catch any errors that may have fallen through the cracks.

• AI Governance Framework: This framework was developed for the World Economic Forum to “champion responsible global design and release of transparent and inclusive AI systems.” This allows us to monitor the probability and severity of harm for all of our AI products.

• Risk Assessments: These assessments allow us to compare our AI safety with the safety of other HR AI technology on the market today. Our score is relatively low risk and has a low probability of harm compared to others.

AI Safety at Scale

You can harness the power of ethical guardrails to accelerate your hiring process while maintaining the highest standards of fairness and validity in practice.

But how do you define fairness, particularly in the context of AI? And how do you measure it?

We believe that all AI needs an off button, a monitoring tool, and success metrics to make sure you're reaching your highest goals while maintaining safety across the board.

To achieve all of this, we concentrated on the following focal points:

• Valid AI: The extent to which artificial intelligence makes choices that are precise, dependable, and suitable for a specific context or objective.

• Fair AI: The utilization of algorithms and machine learning models in a manner that ensures equal opportunities and mitigates biases related to attributes like race, gender, or age.

When it comes to bias, there are three categories: statistical bias, machine learning bias, and IO psychology bias.

• Statistical bias denotes systematic errors that distort findings or conclusions, arising when data collection processes or analysis methods favor certain outcomes, leading to inaccurate representations of populations or phenomena

• Machine learning bias is the occurrence of systematically prejudiced results from algorithmic processes due to flawed assumptions in the machine learning process

• IO psychology bias encompasses any unfair or unbalanced beliefs, preferences, or prejudices toward specific groups, characteristics, or ideas, influencing decision-making processes or assessments within a workplace context

At Phenom, we measure all three bias definitions extensively. Because of this, we've developed the Phenom Fairness and Validity Framework to monitor, test, and ensure compliance of all our AI-powered features. This framework looks at our selection decisions and employer selection decisions and allows us to measure them by accuracy, bias, and a range of other metrics. It also measures the correlation between Fit Score and the eventual interview decision of the recruiter to ensure the least amount of bias possible.

Related reading: Avoiding Bias and Improving Hiring Outcomes During the Interview Process

AI Analytics

We have evolved our analytics platform to support compliance and AI initiatives such as OFCCP compliance, fit scores, adverse impact analysis, and DE&I initiatives. This helps provide valuable insights and supports users’ legal requirements.

Phenom Talent Analytics can be available to support AI dashboards that help companies understand and evaluate the impact of the initiatives on the applicant pool and, by extension, the workforce itself. The adverse impact analysis will help analyze the impact of any bias within specific fit scores and how that affects applicant pools. This will enable users to improve the overall hiring process.

Related reading: Metrics That Matter: How to Leverage Analytics to Make Data-Driven Decisions

To simplify the adoption and responsible use of AI to achieve impactful outcomes for businesses, Phenom prioritizes ethical hiring at the enterprise scale. We believe that all good AI needs a control panel, a monitoring tool, and success metrics to help you feel confident while meeting your goals.

To take a technical deep dive into the world of AI, stream AI Day on demand right here or read the complete overview.


*The information provided on this website does not, and is not intended to, constitute legal advice. All information, content, and materials available on this site are for general informational purposes only.