AI Basics: Definitions, Tests, Models & Philosophical Foundations

TAKEAWAYS:

• AI is the study of creating machines that mimic human intelligence through algorithms, not just data processing.
• The Turing Test measures a machine’s ability to exhibit intelligent behavior indistinguishable from a human’s.
• AI can be classified into narrow (weak) (task-specific) and general (strong) (human-like) forms, with current systems limited to narrow AI.
• Philosophical debates (functionalism, behaviorism) shape how we define intelligence and consciousness in machines.
• Real-world AI applications range from chatbots (eSewa’s customer service) to recommendation systems (YouTube’s content suggestions).
• Ethical concerns (bias, job displacement) and economic impacts (automation) are critical to AI’s societal role.

1. What is Intelligence?

Intelligence is the ability to learn, reason, perceive, understand, and solve problems. It includes:

• Cognitive skills: Memory, logic, creativity.
• Adaptive behavior: Responding to new situations.
• Goal-directed actions: Planning and decision-making.

How AI Defines Intelligence

AI researchers often use Turing’s definition:

"A machine can be said to be intelligent if it behaves in ways indistinguishable from a human under examination."

IMAGE: human brain cross-section | Biological neurons (left) vs. artificial neural networks (right).

2. What is Artificial Intelligence?

Definition:

"AI is the field of computer science dedicated to creating systems that perform tasks requiring human-like intelligence, such as reasoning, learning, and problem-solving."

Key Characteristics of AI:

Types of AI:

1. Narrow (Weak) AI

   • Task-specific (e.g., Siri, Google Translate).
   • Example: eSewa’s chatbot uses NLP to answer user queries about bill payments.
   • Limitation: Cannot perform tasks outside its programming.

2. General (Strong) AI

   • Human-like cognition (theoretical).
   • Example: Hypothetical AI that can debate philosophy or compose music.
   • Current Status: Not yet achieved.

3. Superintelligent AI

   • Hypothetical AI surpassing human intelligence.
   • Debate: Ethical risks vs. potential benefits.

3. The Turing Test: Measuring Machine Intelligence

Proposed by Alan Turing (1950), the test evaluates a machine’s ability to fool a human judge into thinking it’s human.

sequenceDiagram
    participant Judge as Human Judge
    participant Human as Hidden Human
    participant AI as Hidden AI
    Judge->>Human: Asks question: "What is the capital of Nepal?"
    Human-->>Judge: Kathmandu
    Judge->>AI: Asks same question
    AI-->>Judge: Kathmandu
    Judge->>Judge: Cannot distinguish
    Note right of Judge: Test passed if indistinguishable

How It Works:

1. A human judge interacts with two entities: a human and a machine (via text).
2. If the judge cannot reliably distinguish the machine from the human, the machine passes.

Criticisms:

• Limited to text-based interactions (ignores other intelligences like visual or auditory).
• Does not measure true understanding (e.g., a chatbot can mimic conversation without comprehension).

Real-World Example:

• Microsoft’s Tay (2016): Initially passed the Turing Test but later exhibited offensive behavior due to unfiltered learning from users.
• Google’s LaMDA (2022): Claimed to "feel emotions," sparking debates about consciousness in AI.

IMAGE: Turing Test setup diagram | Human judge, hidden human, and hidden AI communicating via text.

4. Philosophical Foundations of AI

AI raises deep questions about mind, consciousness, and ethics. Key perspectives:

Illustration of John Searle's Chinese Room argument showing symbol manipulation without understanding. (Image: Dronebogus, CC BY 4.0, via Wikimedia Commons)

The Chinese Room Argument (Searle, 1980)

• Claim: A person following rules (without understanding Chinese) in a room does not "understand" Chinese.
• Implication: AI that manipulates symbols without true comprehension is not truly intelligent.

IMAGE: Chinese Room thought experiment | Person following rules vs. actual understanding.

5. AI in the Real World: Nepal and Beyond

Example 1: eSewa (Nepal)

• AI Idea Used: Natural Language Processing (NLP) for chatbots.
• How It Works:
  • Users ask questions like "How do I pay my electricity bill?"
  • eSewa’s AI parses the query, checks the user’s account, and provides step-by-step instructions.
  • Behind the Scenes: A decision tree classifies user intent (payment, status check, complaint).

Example 2: Pathao (Ride-Hailing App)

• AI Idea Used: Optimization Algorithms for dynamic pricing and route planning.
• How It Works:
  • AI predicts demand in Kathmandu’s Lalitpur vs. Bhaktapur routes.
  • Adjusts fares based on supply-demand imbalance (e.g., surge pricing during festivals).
  • Uses graph theory to find the fastest route avoiding traffic (real-time data from GPS).

Example 3: NTC’s Smart Grid (Future Plan)

• AI Idea Used: Predictive Maintenance using machine learning.
• How It Works:
  • Sensors on power lines detect corrosion or overheating.
  • AI predicts failures before they occur, reducing blackouts.
  • Worked Example:
    • Input: Temperature = 45°C, Vibration = 0.8 mm/s.
    • Model: Trained on past failure data (e.g., 50°C → 3 failures/year).
    • Output: "Warning: 60% chance of failure in 2 weeks."

Example 4: YouTube (Global)

• AI Idea Used: Collaborative Filtering for recommendations.
• How It Works:
  • If you watch "Nepali Folk Songs" → AI suggests similar videos.
  • Uses matrix factorization to predict preferences based on user behavior.

6. The AI Pipeline: From Idea to Application

Worked Example: Kathmandu Traffic Prediction

Problem: Predict traffic congestion on Ring Road during Dashain. Steps:

1. Data Collection: GPS data from Pathao/Ncell users, weather reports, event calendars.
2. Model: Train a time-series forecasting model (e.g., LSTM neural network).
3. Training:
   • Input: Time (7 AM), Day (Dashain), Weather (Clear).
   • Output: Predicted delay = 45 minutes (vs. usual 20 minutes).
4. Evaluation: Compare predictions with real traffic data (RMSE error < 5 minutes).
5. Deployment: Integrate with Pathao’s app to suggest alternative routes.

7. Ethical and Economic Impacts of AI

Ethical Concerns:

• Bias in AI: Facial recognition systems perform poorly on darker-skinned individuals (e.g., Ncell’s old biometric login).
• Job Displacement: Automation threatens roles like bus drivers (Pathao’s self-driving tests) or bank tellers (Khalti’s AI chatbots).
• Privacy: AI-powered surveillance (e.g., NTC monitoring power theft) raises civil liberty questions.

Economic Impacts:

• Productivity Gains: AI in Nepal’s agriculture (e.g., predicting pest outbreaks) increases yields.
• New Industries: AI-driven tourism (e.g., virtual guides for Swayambhunath).
• Digital Divide: Rural areas lack access to AI tools (e.g., NEPSE’s trading bots favor urban investors).

8. Common Misconceptions About AI

Exam Tip: How to Score Full Marks

1. Define Clearly:

   • Always start with precise definitions (e.g., "AI is the study of creating machines capable of...").
   • Example Answer Start:

     "Artificial Intelligence refers to the simulation of human intelligence in machines, which can perform tasks such as learning, reasoning, and problem-solving. Unlike traditional computing, AI systems adapt and improve based on data."

2. Use Diagrams:

   • Draw flowcharts for processes (e.g., Turing Test setup).
   • Sketch comparison tables (e.g., Narrow vs. General AI).

3. Relate to Real-World Examples:

   • eSewa: NLP for customer service.
   • Pathao: Optimization algorithms for routes.
   • NTC: Predictive maintenance for grids.

4. Discuss Limitations:

   • For the Turing Test, mention lack of true understanding or text-only interactions.
   • For AI ethics, highlight bias, privacy, and job displacement.

5. Mathematical Models (If Asked):

   • For neural networks, show a simple 3-layer ANN with weights and biases:

     Input Layer (x1, x2) → Hidden Layer (h1 = w1x1 + w2x2 + b) → Output Layer (y = v1h1 + c)
     

6. Philosophical Debates:

   • Compare Strong AI (conscious machines) vs. Weak AI (tools).
   • Mention Searle’s Chinese Room when discussing symbol manipulation.

Final Note: AI is not just about robots or sci-fi—it’s about solving real problems (e.g., reducing traffic in Kathmandu, improving healthcare in rural Nepal). Focus on applications, ethics, and limitations to excel in exams!