As artificial intelligence (AI) increasingly integrates into various aspects of business and society, understanding both its inner workings and implications has become crucial for HR and talent acquisition leaders. Over the past two years, you’ve probably gotten far more comfortable with technical concepts like “machine learning,” “neural networks,” and “training data” than you could have possibly imagined when you first entered the talent profession.
But with more and more employers embracing AI as a key component of their hiring processes and talent strategies – not to mention sometimes using automated employment decision tools to make hiring decisions – and AI regulation generally lagging behind, it’s not enough to know the basics of what the technology does and how you might incorporate it into your own efforts. You must understand what it means (and what it takes) to introduce and uphold responsible AI within your organization. That comes with yet another lexicon to learn, however.
To help, we’ve curated this cheat sheet to serve as a valuable resource while you navigate the complex landscape of AI ethics, fairness, and accountability. This glossary defines key terms that underpin responsible AI practices in talent acquisition, from “black boxes” and algorithmic bias to the nuanced differences between explainability and interpretability. By familiarizing yourself with these concepts, you can better ensure that your company not only leverages AI effectively but also maintains moral standards and promotes equitable outcomes for all.
Black-box AI
AI systems whose internal workings are not visible to or understood by humans, making it difficult to grasp or explain how specific decisions or outcomes are produced. As MIT News writes, “modern machine-learning models, such as neural networks, are often referred to as ‘black boxes’ because they are so complex that even the researchers who design them can’t fully understand how they make predictions.”
The use of black-box AI in hiring poses significant risks, including the perpetuation of biases leading to unfair practices, compliance and legal challenges due to lack of transparency and accountability, and potential erosion of trust among candidates and internal stakeholders. Without understanding the inner workings of a black-box AI model, it’s difficult (if not impossible) to optimize or correct it, further complicating its use.
Interpretable AI
Intuitive AI models designed to be easily understood by humans, often allowing users to see or at least understand the path to a decision or prediction. It’s important to note that, per scientific journal Entropy, “an interpretable model does not necessarily translate to one that humans are able to understand the internal logic of or its underlying processes,” but it tends to have an obvious connection between its inputs and outputs.
For example, it makes sense that your ATS might predict how long it will take to close a newly opened req; after all, your ATS has robust data on your company’s applicant flow, applicant quality, time in stage, and so on. But if your ATS starts predicting the weather, that would be far less interpretable.
Explainable AI
According to IBM, explainable artificial intelligence is “a set of processes and methods that allows human users to comprehend and trust the results and output” produced by AI. Unlike black-box AI, explainable AI empowers developers and users to access and modify its underlying approaches to decision-making – the variables used, the weight assigned to each variable, and so on. And explainable AI goes beyond interpretable AI by being not just intuitive and seemingly reasonable but completely transparent.
As a TA leader, using explainable AI in your hiring process is vital for a number of reasons. It promotes fairness and trust by allowing stakeholders to understand, validate, adjust, and continuously optimize an algorithm’s decision-making process. This facilitates compliance with legal regulations, minimizing risk while enhancing your employer brand. Overall, explainable AI leads to more informed, equitable, and defensible hiring decisions.
Algorithmic bias
The Center for Critical Race + Digital Studies defines algorithmic bias as “computational discrimination whereby unfair outcomes privilege one arbitrary group of people over another.” For instance, recruitment AI that systematically favors male over female candidates, such as Amazon’s infamous (experimental) proprietary resume screening tool, may present algorithmic bias.
Algorithmic bias often occurs when social biases are represented in a model’s training data; as Harvard Business Review explains, “data are assumed to accurately reflect the social world, but there are significant gaps, with little or no signal coming from particular communities.”
Therefore, as you integrate AI into your hiring process, it’s imperative that you know the robustness and recency of the data set(s) it’s trained on and audit its outcomes to discover discriminatory patterns as early as possible. It’s also best practice to work with AI developers and vendors that prioritize diversity, equity, inclusion, and belonging (DEIB) on their teams – especially given the homogeneity of the tech sector, where most AI is built. Diverse engineering teams bring a variety of perspectives and experiences to the table, better enabling them to mitigate any algorithmic bias in their models.
Fairness-aware algorithm
An algorithm designed to make decisions that ensure equitable treatment across different demographic (particularly protected) groups. As Sorelle Friedler, former White House Assistant Director for Data and Democracy, details in this study, fairness-aware algorithms incorporate techniques (e.g., reweighting data samples) to detect, mitigate, and prevent biases that could lead to unfair outcomes, thereby reducing discrimination and producing more balanced outcomes.
Since algorithmic bias is a legitimate concern as you introduce AI into your recruitment tech stack, prioritizing fairness-aware models can help protect your company from exclusionary practices, compliance risk, reputational damage, and sub-optimal hiring and business performance. But remember, a computer still may not view “fairness” the way your organization does, so regularly auditing your AI remains non-negotiable.
Assistive intelligence
AI systems – ranging from simple tools like spell checkers to more complex applications such as virtual personal assistants or “copilots” – meant to augment human capabilities by providing support and enhancing performance in various tasks. Unlike fully autonomous AI, assistive intelligence, or assistive AI, works “in conjunction with people to help them accomplish their tasks better,” offering recommendations or direct help without replacing human judgment.
Assistive AI tools can help recruiters and hiring managers by providing data-driven insights throughout the hiring process, suggesting improvements for job posts, or automating administrative tasks like interview scheduling. This not only increases efficiency but also allows hiring teams to focus on more value-added activities that require the intuition, empathy, and interpersonal skills a machine can’t provide.
While so much of the hype around AI centers on its potential to fully remove people from processes, this Stanford University article argues that an AI-assisted world is what we actually desire: one that empowers us with “tools — things we can learn to use — instead of oracles, which give us the right answers but withhold an explanation.” As a talent acquisition leader pioneering the future of AI in your function, you may want to frame your thinking around how you can augment your team members’ abilities with assistive intelligence versus how you can automate their work entirely.
Human-in-the-loop (HITL)
An approach where human judgment is involved in the training, tuning, and testing of AI systems. The literal human in the loop can be a researcher or engineer guiding the development of a model (here at Datapeople, all AI-based guidance is validated by our expert team of data scientists before reaching production) or a user whose feedback on a model’s output (think of the “thumbs down” button on a ChatGPT response) becomes new input for optimizing the model over time.
Assistive intelligence and human-in-the-loop AI are two related but distinct concepts. In essence, while assistive intelligence provides tools that aid human performance, HITL AI integrates humans more deeply into the functioning and refinement of the system, ensuring continuous human oversight.
Automated Employment Decision Tool (AEDT)
AI-powered technology that “substantially assists or replaces discretionary decision making” in hiring or promotions. While many AI tools support the recruitment process without passing judgment on individual candidates, AEDTs are designed to largely determine which talent is shortlisted, interviewed, and/or hired. Common examples include screening software that filters applications based on keywords and interview platforms that analyze candidate responses and behaviors.
The direct impact of automated employment decision tools on employment outcomes has drawn the interest of governing bodies, resulting in AEDTs being subject to stricter regulations – such as New York City’s Local Law 144 – than other AI used in hiring (such as writing assistants like Datapeople’s Smart Editor, chatbots, and interview scheduling assistants). Therefore, understanding the differences between AEDTs and other AI recruitment tools is crucial for remaining compliant with legal standards while fostering a fair, swift, and effective hiring process.
Interested in responsible, explainable AI for a fair, efficient, compliant hiring process?
Embracing responsible AI is not just about adopting shiny new tech; it’s about fostering a culture of fairness, transparency, and accountability – all while staying on the right side of an evolving legal landscape. And with a focus on assistive intelligence, you can enable and upskill your teams with the right tools and knowledge to achieve a more equitable, efficient, and high-performance future.
If you’re ready to elevate your hiring process with secure, compliant, explainable assistive AI, we humbly invite you to explore what Datapeople has to offer by requesting a demo today!
