M. Faris Al Hakim, S.Pd., M.Cs. - Implementasi Kecerdasan Buatan

Main ideas, concepts, and lessons

Learning goals (outcomes)

Students should be able to:

1. Understand various AI technologies.
2. Understand how AI technologies are implemented.

AI material outline (two main parts)

• Types/branches of AI technologies (and examples of what they can do)
• Implementation of AI (how these technologies work in practice)

---

List of AI technologies discussed (with methodology/instructions)

1) Intelligence Search (Information/solution search using AI planning)

Core concept

• A system’s ability to search for information or solutions more intelligently and effectively.
• Uses search methods such as:
  • Blind search
  • Heuristic search

Problems/examples it can solve

• Sudoku
• 8-puzzle / sorting puzzle (arranging numbered boxes)
• Tic-tac-toe
• Travel/route problem (A to B): find the most effective path
• Maze problem: find the most effective way out
• Traveling Salesman Problem: determine the most effective order of visiting points

Example illustrated: 8-puzzle (box shifting)

• A goal state exists where all boxes are in the correct positions.
• A “green box” can be moved:
  • Up → shifts the box upward; the described effect is that it makes the wrong-position boxes change (example given: 3 boxes become incorrect relative to the goal).
  • Right → shifts the box rightward; the described effect is that it changes which boxes are incorrect (example given: 1 box remains incorrect).

Instruction-like idea (decision process)

• Choose the move that brings the current configuration closer to the expected goal state (in the example, shifting right is chosen).

---

2) Expert System

Core concept

• A computer system designed to imitate the knowledge of a human expert in a specific field.
• Purpose: provide intelligent solutions/recommendations based on expert reasoning.

Use case

• Helpful when human experts are unavailable, e.g., remote areas that lack specialists.

How it works (example workflow)

• The system receives symptoms entered by a patient.
• It then infers the most likely diagnosis by matching symptom patterns to possible diseases.

Example symptoms mentioned

• Fever + headache + hoarse voice → likely pharyngitis, not cancer
  • (example explanation: cancer symptoms are not present)

• Cough + fever + neck pain → possible cancer

  • (example contrasts which symptoms correspond to pharyngitis vs cancer)

• The system continues with additional inference to narrow to a likely disease based on the symptom input.

---

3) Machine Learning (ML)

Core concept

• A branch of AI where computers learn from data without being explicitly programmed for every rule.
• Learning happens through:
  • learning from examples
  • learning from experience
  • learning from analogies
• The learning may build:
  • statistical models
  • or algorithms (with statistics/ML foundations)

Main types of ML presented (4 branches)

1. Classification / Supervised Learning

   • Goal: group new data into pre-existing categories/classes.
   • If class labels already exist, new items are assigned to the closest category.

2. Clustering / Unsupervised Learning

   • Goal: group data based on similarity when classes are not predefined.
   • The number of clusters can be set (e.g., cluster into 2 or 3 groups).

3. Reinforcement Learning

   • Core idea: an agent interacts with an environment to achieve goals.
   • Learning mechanism:
     • reward for correct actions/trials
     • punishment/penalty for incorrect actions/trials
   • Analogy given: learning to stand/walk by learning “holding the wall” properly.

4. Deep Learning

   • Presented as the last branch (trending approach mentioned later via CNN in computer vision).

Example illustration: learning to recognize cats vs dogs

• Input: data examples (cats and dogs) with learned characteristics.
• Model building: algorithm learns patterns/characteristics.
• Prediction:
  • when new data is given, the model predicts the category (e.g., dog/cat)

Challenge noted

• ML requires a lot of data; it cannot learn reliably from only one example.

Possible errors

• Predictions can be swapped (cat predicted as dog, and vice versa).

---

4) Computer Vision

Core concept

• AI branch focused on enabling computers to understand and analyze visual data:
  • images
  • videos

Methods mentioned

• Local Binary Pattern (LBP)
• CNN
• Other methods also exist (not limited to these)

Examples of computer vision applications

• Face detection
• Road detection (e.g., detecting road condition for driving/steering)
• Medical image analysis (detecting disease from medical images)

Example process illustrated (pipeline-style steps)

1. Input data (visual input)
2. Pre-processing
3. Feature extraction
   • extract characteristics from the object
4. Classification/decision
   • example outcomes:
     • shape of a circle + wheels (described) → classify as a motorbike
     • wheels (described as 4 wheels) → classify as a car

---

5) Natural Language Processing (NLP)

Core concept

• AI branch focused on processing human natural language.
• Goal: how computers recognize and handle human language.

Methods mentioned

• TF-IDF
• LSTM
• (Plus note that more NLP methods exist)

Examples of NLP implementation

• Sentiment analysis
  • classify comments as positive/negative (example labeled: positive/negative)
• Text classification
  • classify news text into categories such as:
    • sports
    • political
    • educational
• Text summarization
  • summarize multiple paragraphs (example: 5 paragraphs into 1 paragraph)
• Similarity comparison
  • compare similarity between one document and another
• Machine translation
  • referenced as discussed previously in earlier material

---

Speakers / sources featured

• M. Faris Al Hakim, S.Pd., M.Cs. (lecturer/host of the video)