The use of artificial intelligence (AI) in recruiting has become a hot topic. When leveraged correctly, AI can help hiring teams save a lot of time and make better, more objective decisions.

However, modern large language models (LLMs) like ChatGPT often operate as black boxes, making decisions without transparency. This can introduce bias into the recruiting process.

At Dover, we take a different approach to ensure our product is transparent, explainable, and free from bias.

AI is used in two main places to help assist with candidate scoring or evaluation:

\AI is also used in other workflows in the app such as helping customers write job descriptions and emails or transcribing calls.

In this article, we’ll focus on areas where AI is used to help with candidate evaluation.

Within Dover’s product, AI is never making the ultimate decision on whether a candidate is good or not.

At Dover, we rely on a deterministic, rules-based system that we call Criteria that leverages approximately 20 different data points to sort candidates. These data points include:

This criteria is inputted manually by the user. The user can also leverage our AI-powered job description parsing that will fill out a draft of the Criteria for them.

This rules-based system forms the backbone of our candidate scoring processes, ensuring that each decision is clear and explainable.

Within this system, AI is employed in specific, limited capacities such as determining job titles or filtering chat queries. For instance, AI might confirm that a candidate is a "backend engineer" or note that a candidate does not mention Python.

Transparency is a core value at Dover. For every candidate, we provide "explanation cards" that detail which rules they passed and how. These cards ensure that every step of the sorting process is transparent and can be easily understood by our clients.

The primary use of AI within Dover is in role title or persona determination. For example, AI helps in mapping variations like "backend eng" to the standardized title "backend engineer." This function, however, can be turned off if preferred. The key here is that AI is not making subjective decisions; it is simply standardizing job titles for consistency.

Another area where we use AI is in chat filtering. Here, AI translates natural language queries into the Criteria system described above. This helps streamline the process by converting user inputs into actionable queries that fit within our deterministic framework.

In this feature, the AI is simply mapping natural language and not making any conclusions or subjective evaluations.

We are in the early days of legislation around AI.

One of the first landmark laws in this area is NYC AI 144, which applies to companies hiring in the state of New York.

NYC AI 144 lays out a framework for conducting audits to inspect pass-through rates across gender, race, ethnicity, and other intersectional categories to ensure that Automated Employment Decision Tools (AEDTs) aren’t biased in their evaluations.

In compliance with NYC AI 144, Dover conducts regular audits for all our customers to ensure there is no bias in our system. These audits help us maintain fairness and transparency, reinforcing our commitment to ethical AI practices.

At Dover, our approach to using AI in recruiting is rooted in transparency and explainability. By combining a robust, rules-based system with limited AI functionalities, we ensure that our candidate sorting process is both efficient and fair. Our commitment to regular audits and compliance with regulations further underscores our dedication to ethical AI practices in the recruitment industry.