AI & Machine Learning Glossary: Essential Terms for HR

AI and machine learning vocabulary is now part of the HR buyer’s job description. Vendors apply the term “AI-powered” to tools ranging from simple keyword filters to sophisticated predictive models — and HR leaders who cannot distinguish between the two are flying blind in procurement decisions, governance conversations, and technology sequencing. This glossary defines the terms that matter, anchored to how each concept actually appears in HR workflows.

The broader context for this reference: automation must precede AI in the HR technology sequence. Understanding what AI and ML actually do — and don’t do — is the prerequisite for making that sequencing decision correctly.

The Core Hierarchy: AI, ML, Deep Learning, and NLP

These four terms are nested, not synonymous. Getting the hierarchy right prevents the most common vendor confusion.

Artificial Intelligence (AI)

Artificial Intelligence is the discipline of building computer systems that perform tasks that normally require human cognitive ability — reasoning, learning, recognizing patterns, understanding language, and making decisions. AI is the umbrella term. Every other concept in this glossary sits inside it.

In HR, “AI” as a label is applied to tools across the entire talent lifecycle: candidate sourcing engines, resume screening systems, interview scheduling assistants, sentiment analyzers, and predictive attrition models. The label alone tells you nothing about the sophistication of the underlying method. For context on where these applications actually generate value, see the six ways AI is actively reshaping HR operations.

Machine Learning (ML)

Machine learning is a subset of AI in which a system improves its performance by learning from data, rather than by following a fixed set of programmed rules. An ML model is trained on historical examples, learns statistical patterns from those examples, and applies those patterns to new inputs.

In HR, ML is the engine behind tools that predict: which candidates are most likely to accept an offer, which employees show early attrition signals, which training paths correlate with performance improvement. The critical constraint: ML outputs are only as reliable as the training data. A model trained on biased historical hiring decisions will reproduce those biases at scale.

Key components of any ML system:

• Training data — the historical dataset the model learns from
• Features — the specific variables the model uses as inputs (years of experience, skills keywords, tenure history)
• Model — the mathematical structure that maps inputs to outputs
• Output — a prediction, score, classification, or recommendation
• Feedback loop — new outcome data that is used to retrain and improve the model over time

Deep Learning

Deep learning is a subset of machine learning that uses multi-layered artificial neural networks to model complex patterns in data, particularly unstructured data like text, images, and audio. It requires large datasets and significant compute, but produces state-of-the-art results on tasks like language understanding and image recognition.

In HR, deep learning underlies most modern NLP tools, including the large language models that power generative job description writers and AI-assisted interview note tools. HR teams rarely interact with deep learning directly — but the tools built on top of it carry specific governance requirements: they are harder to audit and explain than simpler ML models.

Natural Language Processing (NLP)

Natural Language Processing is the branch of AI that enables computers to read, interpret, and generate human language. NLP algorithms parse text structure, extract meaning, classify intent, identify named entities (companies, job titles, skills), and summarize documents.

Where NLP appears in HR workflows:

• Resume screening — extracting and normalizing skills, titles, and experience from free-text résumés
• Job description generation — drafting and optimizing job postings based on role requirements
• Sentiment analysis — classifying employee survey responses as positive, negative, or neutral
• Chatbot interfaces — interpreting candidate questions and generating contextually relevant answers
• Policy summarization — condensing lengthy HR documentation into plain-language employee guidance

NLP accuracy degrades on non-standard formats and informal language. A résumé that uses unusual section headers or non-English formatting can produce extraction errors that a human screener would never make. Governance of NLP tools requires periodic accuracy audits against human-reviewed samples.

Prediction and Decision Terms

Predictive Analytics

Predictive analytics uses statistical models and machine learning techniques to estimate the likelihood of future outcomes based on patterns in historical data. It answers the question: “Given what we know, what is most likely to happen?”

In HR, predictive analytics is applied to:

• Attrition forecasting — identifying employees with elevated quit probability before they resign
• Candidate success prediction — scoring applicants by predicted performance based on characteristics of historical high performers
• Skill gap forecasting — projecting which competencies the workforce will need in 12–24 months based on business trajectory
• Time-to-fill modeling — estimating how long a role will remain open based on market conditions and historical pipeline data

Predictive analytics is only as reliable as the historical data it trains on. Gartner notes that data quality is consistently the primary obstacle to AI adoption in enterprise HR — not the algorithms themselves. Before deploying predictive tools, HR teams must audit whether their historical data is complete, consistently coded, and free of systematic gaps.

Prescriptive Analytics

Prescriptive analytics goes one step beyond prediction: it not only estimates what will happen, but recommends specific actions to change the outcome. A predictive model flags an attrition risk; a prescriptive model recommends a specific intervention — a compensation review, a manager conversation, a stretch assignment.

In practice, most HR tools marketed as “AI-driven recommendations” are combining predictive outputs with rules-based prescriptions rather than performing genuine prescriptive optimization. HR buyers should ask vendors which component is doing which job.

Supervised vs. Unsupervised Learning

These are the two primary categories of machine learning method, and understanding the difference matters for evaluating HR tools.

• Supervised learning — the model trains on labeled examples (résumés tagged as “hired” or “rejected”; employees tagged as “high performer” or “average”). It learns to predict labels for new inputs. Most HR screening and prediction tools use supervised learning — which means they reproduce whatever patterns exist in the historical labels, including any discriminatory patterns.
• Unsupervised learning — the model finds structure in unlabeled data without being told what to look for. In HR, unsupervised techniques are used for clustering employees into behavioral segments, identifying unusual patterns in engagement data, or grouping job descriptions by role family. It’s exploratory rather than predictive.

Algorithmic Bias

Algorithmic bias occurs when an AI model produces systematically skewed outputs because its training data, feature selection, or objective function reflects historical inequities or flawed assumptions. In HR, bias is not a theoretical concern — it is a documented risk.

A hiring model trained on years of résumés from roles that were historically filled by a demographic majority can learn to penalize résumés that deviate from that majority’s characteristics. The model is doing exactly what it was designed to do — reproduce the pattern in the data — but that pattern is discriminatory. The bias is invisible in the output scores.

Mitigating algorithmic bias requires: diverse and representative training data, regular audits of model outputs disaggregated by protected class, human review at decision points, and governance frameworks that establish accountability. The ethical AI framework for bias, privacy, and risk details the practical steps.

Explainability (Interpretability)

Explainability refers to the degree to which an AI model’s decision-making process can be understood and audited by humans. A fully explainable model can show exactly which input features drove a particular output. A black-box model produces outputs without traceable reasoning.

In HR, explainability is both a legal and an ethical requirement. If an automated screening tool rejects a candidate, the organization must be able to explain why in terms that are specific, auditable, and legally defensible. Models that cannot provide this create compliance exposure. HR buyers should require explainability documentation from vendors before deployment.

Language and Generation Terms

Large Language Model (LLM)

A large language model is a deep learning model trained on massive text corpora — books, websites, code, documents — that can generate, summarize, classify, and answer questions using language. LLMs are the technology underlying tools like generative job description writers, AI-assisted offer letter drafters, and policy summarizers.

LLMs are not fact databases. They generate statistically plausible text, not verified truth. In HR, this means every LLM-generated output — a job description, a policy summary, a candidate communication — requires human review before use. The risk of confident-sounding but factually wrong or legally non-compliant output is real and documented.

Generative AI

Generative AI refers to AI systems that produce new content — text, images, code, audio — rather than simply classifying or predicting. LLMs are one category of generative AI. In HR, generative AI is most commonly deployed for content creation tasks: drafting job postings, synthesizing interview notes, generating onboarding communications, and producing performance review templates.

Generative AI dramatically reduces time-to-draft but introduces the risk of hallucination — outputs that are fluent and confident but factually incorrect. HR teams using generative tools must establish review checkpoints before any generated content reaches candidates, employees, or legal review.

Prompt Engineering

Prompt engineering is the practice of crafting precise input instructions to guide a generative AI model toward a desired output. Quality of output from an LLM is highly sensitive to how the instruction is written — context, specificity, constraints, and examples all shape the result.

In HR, prompt engineering matters for any team using generative AI tools to draft job descriptions, candidate emails, or policy summaries. A vague prompt produces a generic output. A well-structured prompt — specifying role level, required tone, inclusion requirements, and format constraints — produces a draft that requires minimal editing.

Workflow and Process Terms

Recruitment Automation

Recruitment automation is the use of software to execute repeatable, rule-based tasks in the hiring process without manual intervention. It is not AI. Automation follows explicit rules; AI learns from data. The distinction matters because conflating them leads to misplaced expectations.

Automated recruiting tasks include: triggering application confirmation emails, scheduling interviews based on calendar availability, moving candidates between pipeline stages based on status updates, sending rejection notifications, and generating offer letter templates. These tasks require no machine learning — they require a well-configured workflow and clean integrations between systems.

Automation is the prerequisite for AI: it generates the consistent, structured data that ML models need to learn reliably. Organizations that skip automation and deploy AI directly onto manual, inconsistently entered HR data produce unreliable model outputs. See the practical roadmap in AI talent acquisition strategies built on automated workflows.

Applicant Tracking System (ATS)

An applicant tracking system is the software platform that manages candidate data through the recruiting pipeline — capturing applications, storing résumés, tracking status, and coordinating recruiter activity. Most modern ATS platforms include some AI-enhanced features (NLP-based résumé parsing, scoring, or search), but the core function is workflow management, not intelligence.

ATS data quality is the foundation for any downstream AI application. Inconsistently entered disposition codes, missing screening notes, and unstandardized job titles in the ATS will corrupt any ML model trained on that data.

HRIS (Human Resource Information System)

An HRIS is the system of record for employee data — compensation, tenure, role history, performance ratings, and demographic information. HRIS data is the training ground for most predictive HR analytics: attrition models, performance predictors, and workforce planning tools all require clean HRIS data as input. Integration between ATS and HRIS is a critical dependency for any organization pursuing predictive HR analytics.

Sentiment Analysis

Sentiment analysis is an NLP technique that classifies text as positive, negative, or neutral — and sometimes scores emotional intensity along that spectrum. In HR, it is applied to employee engagement surveys, exit interview transcripts, pulse check responses, and Glassdoor-style internal feedback.

Sentiment scores surface themes across large feedback volumes that would be impractical to read manually. They are diagnostic signals, not conclusions. A negative sentiment cluster in one department requires a human conversation to understand root cause — the model identifies the pattern, a manager or HR partner investigates the meaning.

Governance and Risk Terms

AI Governance

AI governance is the set of policies, processes, and accountability structures that organizations establish to ensure AI tools are used ethically, legally, and effectively. In HR, governance covers: approval workflows for new AI tools, bias audit schedules, explainability requirements, data retention policies, and human-override protocols for AI-influenced decisions.

Governance is not optional. Regulatory environments in the EU, US, and increasingly other jurisdictions are imposing requirements on automated decision-making in employment contexts. The HR AI governance and emerging ethical mandates satellite covers the regulatory landscape in detail.

Human-in-the-Loop (HITL)

Human-in-the-loop is an AI system design principle in which human review and approval is embedded at key decision points, rather than allowing the model to act autonomously. In HR, HITL design means that AI tools surface recommendations, scores, or flagged risks — but a human makes the final call on consequential decisions: advancing a candidate, terminating employment, issuing a performance improvement plan.

HITL is both an ethical principle and a legal safeguard. It ensures that algorithmic errors can be caught and corrected before they produce irreversible outcomes.

Data Privacy in HR AI

Data privacy in the context of HR AI refers to the obligations organizations have to protect candidate and employee personal data used to train, run, and improve AI models. This includes consent requirements, data minimization principles (collect only what is necessary), retention limits, and rights to explanation and deletion.

In most jurisdictions, using employee data to train a predictive model without disclosure creates regulatory exposure. HR AI procurement must include a data processing agreement that specifies what data the vendor uses, how it is stored, and whether it is used to train models shared across customers.

Related Terms Quick Reference

For a complementary reference on the software categories these terms power, see the HR tech acronyms and software category definitions guide.

Common Misconceptions

“AI is objective because it removes human bias.”

False. AI models trained on historical human decisions inherit and often amplify the biases embedded in those decisions. The Harvard Business Review and Deloitte have both documented cases of hiring algorithms penalizing protected characteristics present in résumés. Objectivity is not a property of machine learning — it is an outcome that requires deliberate data curation, bias auditing, and governance design.

“More data always means better AI.”

False. Volume without quality produces overfit or biased models. A predictive attrition model trained on five years of inconsistently coded HRIS data will perform worse than one trained on two years of clean, audited records. Data quality trumps data volume every time.

“AI makes the decision.”

In well-designed HR systems, AI does not make decisions — it makes recommendations. The human-in-the-loop principle, and in many jurisdictions the law, requires that consequential employment decisions have a human decision-maker who can explain and take accountability for the outcome. AI is a decision support tool, not a decision authority.

“Automation and AI are the same thing.”

They are not. Automation executes predetermined rules without learning. AI learns from data and adapts. A workflow that sends an interview confirmation email when a candidate reaches a specific pipeline stage is automation. A tool that predicts which candidates in the pipeline are most likely to accept an offer is AI. Both are valuable; they do different things; they have different governance requirements.

Putting the Vocabulary to Work

This vocabulary is a procurement and governance tool, not just a reference. Armed with these definitions, HR leaders can ask vendors: Which ML method does your tool use? What was the training dataset, and when was it last updated? How does your tool surface explainability to support adverse action documentation? What is your bias audit protocol and how often does it run?

Those questions separate tools that genuinely apply AI from tools that apply the label. And they create the foundation for technology governance that protects the organization as AI regulation tightens.

For the strategic layer — building the business case, measuring ROI, and sequencing automation before AI — see measuring HR automation ROI with the right KPIs and the full framework for when to automate versus when to augment HR decision-making. Both sit within the broader context established by the parent pillar on workflow automation strategy for HR.