Key AI Distinctions & Info

• Look for Learning, Not Just Logic ↓

  Learning vs. Fixed Rules

  AI systems learn patterns from large, varied data. They adapt and can make predictions or decisions on new, unseen data based on these learned patterns.

  Classic automation follows fixed if-then rules you could write on one page. It executes predefined instructions without learning or adapting.

• Three Quick Heuristics ↓

  Quick Checks for AI

  • Does the output keep improving or adapting with more data/use? → likely AI.
  • Could you predict every result with a simple flowchart? → probably not AI.
  • Is there a “training” or “model update” step mentioned? → strong AI clue.
• Common Classroom Examples ↓

  In the Classroom: AI vs. Not AI

  Likely AI:

  • Real-time language translation (e.g., Google Translate)
  • Adaptive math platforms that adjust difficulty based on student performance
  • Voice-activated assistants (e.g., Siri, Alexa)

  Probably Not AI:

  • Random name pickers
  • Static quiz banks with predefined questions and answers
  • Timed PowerPoint clickers (simple automation)
• Typical Marketing “Tells” ↓

  Decoding Marketing Claims

  “Powered by large language models” or “trained on millions of examples” = credible AI claim, suggesting underlying machine learning.

  Vague phrases like “smart engine” or “intelligent workflow” with no details on data or model type = raises suspicion; might be rule-based.

• Data → Model → Output Lens ↓

  The AI Pipeline (Link to Module 2)

  Consider these three components:

  • Data: What raw information goes in? (e.g., student essays, website click patterns, images).
  • Model: How does the system turn that data into insights or predictions? (e.g., neural network, decision tree, clustering algorithm). This is the "learning" part.
  • Output: What does the system produce? (e.g., a personalised hint, a grade prediction, a generated image, a recommended video).

  If any link in this chain is missing or purely rule-based (e.g., data directly triggers a fixed output via simple rules), it’s likely automation, not AI.

• Bias & Limitations to Watch For ↓

  Understanding Potential Downsides

  AI systems reflect biases present in their training data. If the data is skewed, the AI's output will be too.

  Deterministic code (classic automation) reflects the programmer’s assumptions and explicit biases.

  Always ask: “Whose data was used? From where? Who might have been left out or underrepresented?”

• Privacy & Transparency Checks ↓

  Data Handling and Openness

  Vendors should be able to clearly explain:

  • How student/user data is stored and secured.
  • Whether and how data is anonymised or aggregated.
  • Whether their AI models run locally on the device or in the cloud (which has implications for data transfer).

  A refusal to clarify these points, or overly vague answers, often signals minimal or no genuine AI under the hood, or poor data practices.

• Edge-Case Litmus Tests ↓

  Practical Tests for AI Claims

  • Offline mode: True AI features, especially complex ones, usually require significant cloud compute power. If a tool works identically and fully offline, its "AI" components might be simpler rule-based systems. (Some on-device AI exists but is typically more limited).
  • No internet updates: If the tool never needs or receives model updates (distinct from software feature updates), it’s suspect. AI models often benefit from retraining or updates with new data.
• Student-Friendly Explanations ↓

  Explaining to Learners

  “AI is like a student who practises with thousands of examples to get good at a task, and can then apply that learning to new examples.”

  “Automation (not AI) is like a recipe that always bakes the exact same cake every time you follow the instructions.”

• Prompt Stem for Reflection ↓

  Critical Thinking Prompt

  For your reflection wall or journal:

  “The next time I hear a product is ‘powered by AI,’ I will ask: What data taught it to do that, and how does it actually learn?”