What Is AI, Really?

Let's start with the clearest possible definition: Artificial Intelligence is a computer system designed to perform tasks that typically require human intelligence.

That's broad because AI is broad. It includes systems that play chess, recommend movies, recognize faces in photos, and draft emails. The common thread: these are all tasks that require some form of problem-solving or pattern-recognition that, in humans, we call "intelligence."

But here's the critical part: AI doesn't understand in the way humans understand. AI systems don't think. They don't have consciousness. They don't care about outcomes. They perform operations on data according to mathematical patterns.

What AI Is Not

Before we go deeper, let's clear away some myths:

• AI is not conscious. It doesn't think, decide, or intend. It processes information.
• AI is not all-knowing. It only "knows" what it was trained on. Ask a 2020-trained AI about 2025 events and it won't know.
• AI is not objective. It's trained on human data, which contains human biases. AI inherits those biases.
• AI is not magical. It's mathematics and statistics. Powerful mathematics, but still math.
• AI is not a person. You can't hold it accountable. You remain accountable for how you use it.

Automation vs. Machine Learning: The Critical Difference

This distinction matters more than you might think. Many recruiting tools claim to use "AI," but they're actually using rule-based automation. Understanding the difference helps you ask better questions when evaluating tools.

Rule-Based Automation

This is what happens when you (or a developer) write explicit rules. Example:

"If a candidate has 5+ years of Python experience AND currently works at a Fortune 500 company AND went to a top-20 university, move them to the fast track."

How it works: You define the rules. The system follows them exactly. Every time. No flexibility.

Strengths:

• Predictable—you set the rules, you know the outcome
• Transparent—you can see exactly why a candidate was included or excluded
• Controllable—you can adjust or disable rules instantly

Limitations:

• Rigid—can't adapt to context or nuance
• Requires you to know which rules matter
• Misses signals that don't fit the rules (e.g., a brilliant candidate who went to a state school)

Machine Learning

Instead of you writing rules, the system learns patterns from data. Example:

You feed the system 1,000 resumes of people you hired and 1,000 resumes of people you passed on. The system learns: "Candidates who stayed at companies for 3-5 years tend to accept offers more often." It didn't memorize a rule; it found a pattern.

How it works: You provide training data. The system learns statistical patterns. It applies those patterns to new data to make predictions or decisions.

Strengths:

• Can find patterns humans miss
• Adapts as data changes
• Can handle nuance (not just binary rules)
• Often more accurate than rule-based automation

Limitations:

• A "black box"—you may not understand why it makes decisions
• Only as good as the data it learned from (if your past hiring was biased, the model learns bias)
• Requires large amounts of historical data
• Can find spurious patterns (correlations that aren't causal)

Machine Learning: Deeper Dive

Machine learning is the engine behind most modern AI applications. It's worth understanding how it actually works, because this understanding reveals both its power and its limitations.

The Three Phases of Machine Learning

Phase 1: Training

You give the system a large dataset with known outcomes. Example: 5,000 resumes, each labeled "hired" or "not hired," along with what happened (did they accept an offer? did they succeed in the role?).

The system analyzes patterns: Which characteristics correlate with success? Which correlate with rejection? It builds a mathematical model—not a list of rules, but a statistical representation of those patterns.

Phase 2: Validation

You test the model on data it hasn't seen before. If the model learned real patterns (not just memorized the training data), it should perform well on new data.

Phase 3: Deployment

You use the trained model on new candidates. The model analyzes their resume and predicts: "This candidate has an 78% likelihood of accepting an offer based on patterns from your past hires."

A Critical Limitation: Probability vs. Certainty

That "78% likelihood" is a probability. It's based on patterns across your historical data. But for any individual candidate, the outcome is either 0% or 100%—they accept or they don't.

This is where organizations often misuse machine learning. They treat a 78% prediction as a high-confidence decision. They auto-reject candidates the model rates low. But the model is working with incomplete information about individuals. A candidate rated 45% likely to accept might have circumstances that make them very likely—a recent relocation, a long recruiting process, a specific role fit.

Natural Language Processing (NLP)

NLP is the subfield of AI that deals with language—both understanding text and generating it. Most recruiting AI applications rely heavily on NLP.

What NLP Does

• Extracts information from text: Reading a resume and identifying: name, contact info, skills, experience, education
• Classifies text: Reading a job description and categorizing it as "technical," "sales," "operations," etc.
• Matches text: Comparing candidate skills on a resume to required skills in a job description
• Generates text: Writing job descriptions, emails, or interview questions
• Summarizes information: Reading a long resume and producing a short summary

How NLP Works (Simplified)

Traditional NLP relies on rules and dictionaries. Example: "If I see 'Python' on a resume, tag it as a programming language."

Modern NLP uses machine learning. The system learns: "Words that appear near 'Python'—like 'code,' 'library,' 'project'—are likely technical context." It learns this from analyzing millions of texts.

This is more flexible than rules, but it's still pattern-matching. If a resume says "I'm fluent in Python" (a language, not the programming language), the system might get confused.

Large Language Models: How They Actually Work

Now we get to what's probably transforming your recruiting landscape most: Large Language Models (LLMs). These are systems like ChatGPT, Claude, or the generative AI in recruiting platforms.

What Is an LLM?

An LLM is a machine learning model trained on vast amounts of text to predict patterns in language. Specifically, it learns: "Given this sequence of words, what word should come next?"

Imagine showing an AI billions of sentences from the internet, books, job postings, resumes, and emails. You tell it: "Learn what words and phrases commonly follow other words." The system builds a statistical model of language patterns.

When you ask it a question, it generates an answer by predicting, one word at a time, what word should come next. It's doing this millions of times per response, making probabilistic decisions each step.

A Recruiting Example

You prompt an LLM:

"Draft a compelling outreach email to a senior backend engineer with 5 years of experience. We're hiring for a Principal Engineer role at a Series B startup."

The LLM doesn't understand that backend engineering is a specific discipline, or why a Series B company might be attractive, or what a Principal Engineer does.

Instead, it has learned patterns like:

• Words often found near "senior engineer": experience, technical, leadership, impact
• Structure of professional outreach: greeting → value proposition → specific opportunity → call to action
• Patterns from recruiting emails it saw during training

The model predicts: "Hi [Name], I noticed your impressive work at [Company]..." because that sequence commonly follows similar openings.

The output might sound great. It might be useful. But the AI got there through pattern-matching, not through understanding your candidate or opportunity.

Why This Matters: The Hallucination Problem

Because LLMs are predicting based on patterns, not accessing real information, they can confidently generate false information. This is called "hallucination."

Example: You use an AI tool to summarize a candidate's LinkedIn profile. The tool generates: "Candidate has 12 years of marketing experience and currently leads the marketing team at TechCorp."

You trust this summary and move the candidate forward. During the interview, the candidate mentions they've been at TechCorp for 2 years and have 7 years of experience total. The AI hallucinated.

The AI didn't intentionally lie. It was predicting based on patterns it had seen. But the pattern-prediction happened to be wrong.

Putting It All Together: A Real Recruiting Scenario

Let's trace how AI technologies show up in a realistic recruiting workflow:

Scenario: Resume Screening at Scale

You have 500 resumes for a senior accountant role. You use a recruiting platform that advertises "AI-powered resume screening."

What's actually happening:

1. NLP extracts information: The system reads each resume and uses NLP (natural language processing) to identify: skills, years of experience, job titles, education. This is pattern-matching against known categories.
2. Rule-based filtering: The system applies rules you set: "Must have CPA certification" and "Minimum 5 years accounting experience." These are hard filters—automation, not AI.
3. Machine learning ranking: For candidates who pass the rules, a machine learning model ranks them by "fit." The model was trained on 1,000 resumes of your past hires and non-hires. It learned: "Candidates with specific software skills + Big 4 experience + CPA are more likely to succeed here." It scores each candidate accordingly.
4. You review the top 50: The system prioritizes the top 50 candidates by the ML score. But you don't blindly trust the ranking—you read through them, consider your actual role needs, and make your own judgment.

The AI contributions: NLP for information extraction (useful), rules for filtering (helpful), ML for ranking (okay if used as a guide, not gospel).

The human judgment: You deciding which ranked candidates to interview, recognizing context the AI missed, and taking accountability for your decisions.

Another Scenario: Drafting Job Descriptions

You need to write a job description for a new role. You use ChatGPT or your company's AI tool:

"Write a job description for a Senior Accountant role. We need someone with CPA, 5+ years experience, strong Excel skills, and ability to mentor junior accountants."

The LLM generates a job description. It's using patterns from job descriptions it saw during training. The output includes standard sections: About the Role, Key Responsibilities, Required Qualifications, Preferred Qualifications, Why Join Us.

What's good: The LLM saved you blank-page syndrome. You have a starting point. It's well-structured.

What's not: The LLM doesn't know your company culture, your actual compensation range, your growth trajectory, or why candidates should genuinely want to work for you. It's generating plausible text based on patterns from thousands of job descriptions.

Your role: You take the draft, customize it with your actual company details, tone, and differentiation. The AI generated options; you apply judgment.

Key Takeaway

Frequently Asked Questions

If AI is just pattern-matching, why is everyone so excited about it?

Because pattern-matching at scale is incredibly powerful. AI can process thousands of resumes in seconds, identify correlations in hiring data that humans would miss, and generate drafts that save time. The pattern-matching isn't magic, but the speed and scale are genuinely useful. The key is knowing what it's good at (volume, variations, summarization) and what it's not (judgment, context, accountability).

Can AI be biased if it's just following patterns?

Absolutely. If you train a machine learning model on your historical hiring data, and your historical hiring was biased (e.g., you promoted men at higher rates than women), the model will learn those patterns. The AI isn't intentionally biased; it's faithfully learning what it was trained on. This is why it's critical to understand what data trained a model and how that data was created. We'll dive deep into bias in Chapter 4, but this is foundational: AI can inherit and amplify human bias.

What's the difference between AI, machine learning, and deep learning?

AI is the broad umbrella: any computer system designed to perform tasks requiring intelligence. Machine learning is a subset of AI: systems that learn patterns from data instead of following explicit rules. Deep learning is a subset of machine learning: systems that use neural networks (roughly inspired by how brains work) to learn patterns. LLMs are deep learning systems. For recruiting purposes, you mostly care about machine learning vs. rule-based automation.

If LLMs can hallucinate, should I use them at all in recruiting?

Yes, but carefully. LLMs are excellent for generating drafts (job descriptions, email templates, interview guides), brainstorming frameworks, and summarizing information you already have. They're risky for making claims about candidates (summarizing resumes, predicting fit) without verification. The key is understanding what task you're using it for. Drafting a job description? Great. Auto-screening resumes based on an LLM's understanding? Risky. Verify everything.

How do I know if a recruiting platform is actually using AI or just marketing?

Ask these questions: (1) Is it rule-based automation (you set the rules) or machine learning (learns from data)? (2) What data was it trained on? (3) Can they explain how it works? (4) Can you audit or override its decisions? (5) What happens when it's wrong? If the vendor can't answer these clearly, it's probably marketing hype. Real AI tools can explain what they're doing. Vague claims about "advanced algorithms" or "proprietary AI" often mean they won't or can't explain.

Is AI really the future of recruiting, or is it hype?

It's both. AI is genuinely useful for handling volume, generating options, and finding patterns at scale. Those capabilities are real and valuable. But the hype often suggests AI will replace human recruiters, make perfect hiring decisions, or solve problems it can't actually solve. The future of recruiting isn't "AI" or "humans"—it's humans and AI together, with clear boundaries on what each does best. AI handles volume and structure. Humans provide judgment, context, and accountability.