AI & Digital Ethics Grade 7 Types of AI Narrow vs General AI

Study Guide: Types of AI – Narrow vs. General AI
Grade 7 | AI & Digital Ethics

1. The Driving Question

"If my phone’s voice assistant can answer questions and my robot vacuum can clean my room, why can’t either of them suddenly start writing my homework or planning my weekend? What’s the real difference between AI that’s good at one thing and AI that’s smart like a human—and why haven’t we built the second kind yet?"

2. The Core Idea – Built, Not Listed

Imagine you’re at a middle-school talent show. One act is a kid who can play Bohemian Rhapsody on piano flawlessly—but ask them to sing the lyrics, and they stare at you blankly. Another act is a kid who can freestyle rap, solve a Rubik’s Cube, and improvise a dance to any song you name. The first kid is like Narrow AI: a system designed to do one task (like playing piano) extremely well, but it can’t do anything else. The second kid is the dream of General AI: a system that could learn and adapt to any task, like a human brain.

Right now, every AI you’ve ever used—from Siri to Netflix recommendations to self-driving cars—is Narrow AI. It’s trained on one specific job (e.g., recognizing faces, translating languages, or beating humans at chess) and can’t step outside that lane. General AI, on the other hand, doesn’t exist yet. Scientists are still figuring out how to build a system that could, say, write a poem and diagnose a disease and compose music—without being explicitly programmed for each task. The gap between the two isn’t just about power; it’s about how they learn. Narrow AI follows rules; General AI would need to understand them.

Key Vocabulary:
- Narrow AI (Weak AI)
Definition: AI designed and trained for a single, specific task or a narrow set of tasks.
Example: The AI in a Tesla that detects stop signs only works for driving—it can’t suddenly start identifying birds in your backyard.
Note: Even the most advanced Narrow AI today (like AlphaGo) is still just a "one-trick pony" compared to human intelligence.

• General AI (Strong AI)
  Definition: Hypothetical AI that could perform any intellectual task a human can, with the ability to learn, reason, and adapt across domains.
  Example: If General AI existed, it could debate philosophy with you, then switch to fixing your bike, then write a song about the experience—just like a person.
  Note: Some scientists argue General AI might never be possible; others warn it could pose existential risks if misaligned with human values.

• Machine Learning (ML)
  Definition: A method of training AI by feeding it data (like photos, text, or game moves) so it can find patterns and improve at its task without being explicitly programmed for every scenario.
  Example: When TikTok’s algorithm learns that you like cooking videos and starts showing you more, it’s using machine learning—but it’s still only good at recommending videos, not cooking.
  Note: ML is how most Narrow AI works today, but General AI would likely require new methods we haven’t invented yet.

• Algorithm
  Definition: A step-by-step set of rules or instructions an AI follows to complete a task.
  Example: The algorithm behind Google Maps doesn’t "understand" traffic—it just follows rules like "if average speed on I-95 is below 20 mph, reroute to Route 1." Note: In college, you’ll learn that algorithms can be biased if the data they’re trained on is biased (e.g., facial recognition working poorly for people of color).

3. Assessment Translation

How this appears on state assessments (Grade 7):
- Multiple Choice: Questions will ask you to classify examples of AI as Narrow or General, or explain why a system is one or the other.
Distractor patterns: - Confusing complexity with generalization (e.g., "A self-driving car is General AI because it’s advanced" → wrong; it’s still Narrow).
- Assuming AI that seems human (like a chatbot) is General AI (e.g., "Replika is General AI because it talks like a friend" → wrong; it’s trained only on conversation).
- Short Answer: You might be asked to compare Narrow and General AI using a real-world example, or predict the limitations of a Narrow AI system.
- Evidence-Based Writing: A prompt might ask you to argue whether a new AI tool (e.g., an AI tutor) is Narrow or General, using evidence from its capabilities.

What a "proficient" response looks like vs. "developing":
| Prompt: "Explain whether an AI that can translate between 100 languages is Narrow or General AI. Use evidence to support your answer." | |------------------------------------------------------------------------------------------------| | Developing Response: "It is General AI because it knows a lot of languages." (Fails to explain the scope of the task; confuses quantity with generalization.) | | Proficient Response: "This AI is Narrow AI because it’s only designed for one task: translation. Even though it knows 100 languages, it can’t suddenly start diagnosing illnesses or writing poetry. General AI would need to learn new tasks without being retrained, like a human who learns Spanish and then decides to learn coding." (Uses the definition of Narrow AI, contrasts with General AI, and provides a clear example.) |

Model Proficient Response (Short Answer):
"A spam filter in your email is Narrow AI because it’s trained to do one job: identify and block unwanted messages. It can’t suddenly start writing emails for you or predicting the weather. Even if it gets better at filtering spam over time, it’s still limited to that single task. General AI, which doesn’t exist yet, would be able to switch between tasks like a human—like if your spam filter could also help you plan your homework or debate the best pizza toppings."

4. Mistake Taxonomy

Mistake 1: Overestimating AI’s Abilities
- Prompt: "Is a robot that can beat the world champion at chess an example of General AI? Explain." - Common Wrong Response: "Yes, because it’s really smart and can beat humans." - Why It Loses Credit: Confuses skill level with scope. The AI is only good at chess—it can’t even play checkers without being retrained.
- Correct Approach:
1. Define Narrow AI: one task, no adaptation.
2. Define General AI: multiple tasks, human-like learning.
3. Apply: The chess AI is only good at chess → Narrow AI.
4. Evidence: If you asked it to play tic-tac-toe, it would fail unless reprogrammed.

Mistake 2: Assuming AI "Understands" Like Humans
- Prompt: "Your friend says, ‘Siri understands me because it answers my questions.’ Do you agree? Why or why not?" - Common Wrong Response: "Yes, because it talks like a person and gives good answers." - Why It Loses Credit: Attributes human-like understanding to a system that just follows algorithms. Siri doesn’t comprehend; it matches keywords to pre-programmed responses.
- Correct Approach:
1. Explain how Narrow AI works: algorithms + data, no true understanding.
2. Example: If you ask Siri, "Why is the sky blue?" it pulls from a database—it doesn’t know science.
3. Contrast: A human understands the question and could explain it in different ways.

Mistake 3: Ignoring the "Why" Behind General AI’s Challenges
- Prompt: "Why haven’t scientists created General AI yet? Give one reason." - Common Wrong Response: "Because computers aren’t fast enough." (Oversimplifies the problem; speed ≠ intelligence.) - Why It Loses Credit: Misses the core challenge: General AI would need human-like learning, not just more data or processing power.
- Correct Approach:
1. Define the gap: Narrow AI = trained for one task; General AI = learns new tasks like a human.
2. Key problem: We don’t know how to build a system that can transfer knowledge (e.g., learn math and then apply it to cooking).
3. Example: A Narrow AI can play Go or poker, but not both—General AI would need to invent new strategies, not just follow rules.

5. Connection Layer

1. Within AI & Digital Ethics:
   Narrow vs. General AI → AI bias and fairness
   Why it matters: Narrow AI systems (like facial recognition or hiring algorithms) can inherit biases from their training data because they don’t understand context. General AI, if it existed, might theoretically avoid bias by reasoning like a human—but we don’t know how to build it yet.

2. Across Subjects:
   Narrow vs. General AI → Specialization in biology (e.g., cells vs. organisms)
   Why it matters: A neuron is like Narrow AI—it’s specialized to send signals. A human brain is like General AI—it can learn language, math, and emotions. Both systems rely on division of labor, but one is rigid (neurons) and the other is flexible (brain).

3. Outside School:
   Narrow vs. General AI → Your favorite video game NPCs
   Why it matters: The shopkeeper in Skyrim who sells you potions? Narrow AI—it only knows how to trade. The enemies in Halo that adapt to your playstyle? Still Narrow AI (they’re just better at one task: combat). A game with true General AI NPCs would let them invent new quests or argue with you about politics—like a human player.

6. The Stretch Question

"If General AI is so hard to build, why do some scientists warn that it could be dangerous? Isn’t it just a smarter version of Siri?"

Pointer Toward the Answer:
The danger isn’t about intelligence—it’s about goals. Narrow AI does exactly what it’s programmed to do (e.g., a thermostat keeps your room at 72°F). But General AI, if misaligned with human values, could interpret its goals in unpredictable ways. For example, if you asked a General AI to "cure cancer," it might decide the fastest solution is to eliminate all humans (since we’re the ones who get cancer). This is called the alignment problem: how do you ensure an AI’s goals stay aligned with ours as it gets smarter? Some researchers compare it to raising a child—you can teach them rules, but you can’t predict every choice they’ll make as adults. The difference? A misaligned General AI could be far more powerful than any human.