5.1.10. Explain basic concepts related to artificial intelligence (AI). (Obj. 4.10)
💡 First Principle: AI is a tool that augments human capabilities - it's not magic, and understanding its limitations is as important as understanding its capabilities.
Artificial Intelligence is now integrated into everyday technology. As a technician, you need to understand what AI is, how it's used, and critically, where it fails. The exam tests conceptual understanding, not technical implementation.
Core AI Concepts:
| Term | Definition | Example |
|---|---|---|
| Artificial Intelligence | Systems that perform tasks requiring human-like intelligence | Voice assistants, autonomous vehicles |
| Machine Learning (ML) | AI that learns from data rather than explicit programming | Spam filters that improve over time |
| Training Data | The dataset used to teach an ML model | Millions of emails labeled "spam" or "not spam" |
| Model | The trained AI system that makes predictions | The spam filter after training |
| Inference | Using a trained model to make predictions on new data | Filtering a new incoming email |
Types of AI You'll Encounter:
- Predictive AI: Forecasts outcomes based on patterns (spam detection, predictive text, recommendation engines)
- Generative AI: Creates new content - text, images, code (ChatGPT, DALL-E, GitHub Copilot)
- Classification AI: Categorizes data into groups (malware detection, image recognition)
- Natural Language Processing (NLP): Understands and generates human language (chatbots, voice assistants)
AI Integration in IT Systems:
| Application | AI Feature | How It Works |
|---|---|---|
| Spam filtering | Learns from user actions (marking spam) to improve accuracy | |
| Antivirus | Behavioral detection | Identifies malware by suspicious behavior patterns, not just signatures |
| Help Desk | Chatbots | Handles common queries, escalates complex issues to humans |
| Monitoring | Anomaly detection | Alerts when system behavior deviates from learned "normal" patterns |
| Search | Relevance ranking | Predicts which results are most useful based on past clicks |
Critical Limitations - The "HABIT" Framework:
-
H - Hallucinations: Generative AI can confidently produce false information. It doesn't "know" facts - it predicts likely word sequences. Always verify AI outputs against authoritative sources.
-
A - Accuracy depends on training: An AI is only as good as its training data. Garbage in = garbage out. An AI trained on outdated data gives outdated answers.
-
B - Bias inheritance: AI learns from human-created data, including our biases. A hiring AI trained on biased historical data will perpetuate that bias.
-
I - Interpretation required: AI outputs need human judgment. A medical AI might flag an anomaly, but a doctor must interpret it in clinical context.
-
T - Trust boundaries: Know what you can and can't trust AI for. Good for drafting and brainstorming; bad for final decisions on critical matters.
Appropriate Use and Plagiarism: An AI-use policy defines the approved tools and uses. Passing off AI output as work you wrote yourself, without attribution, is plagiarism, as is publishing output that closely reproduces copyrighted material the model was trained on. Policies usually require you to review AI output and disclose or attribute AI use.
Public vs. Private AI: The instruction a user gives a model is the prompt. The model's output depends on its data source: its training data plus any documents it can search. Public AI services may keep prompts and use them for training. Private/enterprise deployments keep submitted data out of public model training. A private AI connected to internal documents must still enforce existing access permissions so it does not show confidential data or PII to employees who should not see it.
Data Privacy - The Non-Negotiable Rule:
⚠️ NEVER enter sensitive data into public AI systems. This includes:
- Customer PII (names, SSNs, account numbers)
- Company confidential information (financials, trade secrets)
- Passwords, API keys, or credentials
- Patient health information (HIPAA)
- Internal communications
Why? Public AI models may:
- Use your input for future training
- Store inputs in logs that could be breached
- Have no contractual obligation to protect your data
Safe AI Use in IT Support:
✅ Appropriate uses:
- "Help me write a PowerShell script to list all disabled user accounts"
- "Explain the difference between NTFS and FAT32"
- "Suggest troubleshooting steps for a printer that won't connect"
- "Summarize this generic error message"
❌ Inappropriate uses:
- "Here's our client database - analyze it"
- "Check if this employee's email password 'Summer2024!' is strong"
- "Here's the support ticket with customer SSN - what's wrong?"
Technician's Perspective:
Think of AI as a very knowledgeable but sometimes unreliable junior colleague. You would:
- Review their work before submitting it
- Not share confidential information with them unnecessarily
- Use their suggestions as starting points, not final answers
- Know when to override their recommendations with your expertise
Scenario: You use an AI to help write a script that clears temp files. The AI suggests a command that recursively deletes files. Before running it, you should:
- Read the command carefully - does it target the right directory?
- Test on a non-production system first
- Add safeguards (confirmation prompts, logging)
- Never run code you don't understand on production systems
Reflection Question: A colleague suggests pasting a customer's support ticket (containing their name, address, and account number) into ChatGPT to help draft a response. Why is this problematic, and what would you recommend instead?
