1. What is Artificial Intelligence?
1.1 Agenda
Estimated reading time: ~8 minutes
Learning Outcomes
- Articulate (diễn đạt chính xác) a technically precise definition of Artificial Intelligence
- Distinguish (phân biệt) AI from Traditional Programming with concrete examples
- Identify real-world AI applications relevant to your industry
1.2 Glossary
| Term | Quick Explanation |
|---|---|
| Artificial Intelligence (AI) | Hệ thống máy tính mô phỏng trí tuệ con người — lý luận, học hỏi, ra quyết định. |
| Rule-based System | Chương trình dùng luật if-then do con người viết tay. |
| Inference | Quá trình AI áp dụng kiến thức đã học để đưa ra dự đoán trên dữ liệu mới. |
| Deterministic | Hệ thống luôn cho cùng một output với cùng input. Lập trình truyền thống là deterministic; AI thì probabilistic (theo xác suất). |
2. Problem Statement
Traditional (truyền thống) software hits a hard ceiling (giới hạn cứng) on certain problem classes:
- Combinatorial explosion (bùng nổ tổ hợp): Writing explicit (tường minh) rules for every scenario is intractable (bất kh ả thi). A fraud detection system with thousands of
ifconditions still misses new patterns it has never seen. - Brittleness (tính giòn — dễ vỡ khi điều kiện thay đổi): Rules break when the world changes. A price model hard-coded for 2020 fails after supply chain disruptions (gián đoạn chuỗi cung ứng) in 2022.
- Non-generalizing (không tổng quát hóa được): A spam filter defined by blacklisted (danh sách đen) keywords fails against obfuscated (cố tình viết sai) text like "fr€€ m0ney."
The shift: Instead of asking a developer to enumerate (liệt kê) all rules, AI extracts rules automatically from examples in data.
3. What is Artificial Intelligence?
3.1 Technical Definition
Artificial Intelligence (AI) is the discipline (lĩnh vực nghiên cứu) of designing computational systems that can perform tasks requiring cognitive (nhận thức) functions typically associated with human intelligence — including perception (nhận thức cảm quan), language understanding, reasoning (lý luận), and learning from experience.
3.2 Core AI Capabilities
Each capability maps directly to an AI Workload covered in Workshop 1.3 — AI Workloads.
4. AI vs. Traditional Programming
4.1 The Fundamental Inversion
The location of intelligence (nơi trú ngụ của logic quyết định) moves from the developer's mind into the data.
4.2 Side-by-Side Comparison
| Dimension (chiều so sánh) | Traditional Programming | AI / Machine Learning |
|---|---|---|
| Logic source | Human expert writes rules | Algorithm extracts (trích xuất) rules from data |
| Flexibility | Rigid (cứng nh ắc) — changes require code updates | Adaptive (thích nghi) — improves as new data arrives |
| Transparency | Fully auditable (có thể kiểm tra toàn bộ) | Can be a "black box" |
| Failure mode | Fails on unknown edge cases (tình huống ngoài luật) | Degrades (suy giảm) gracefully but can be biased (thiên lệch) |
4.3 Concrete Examples
| Task | Traditional Approach | ML Approach |
|---|---|---|
| Spam Filter | Blacklist specific keywords | Learn statistical (thống kê) signatures from 100,000 labeled emails |
| Bank Fraud | Flag any transaction over $10,000 | Build a behavioral (hành vi) model per user — flag deviations (sai lệch) from their baseline (hành vi thông thường) |
4.4 When NOT to Use AI
| Use AI when... | Use Traditional Programming when... |
|---|---|
| Problem too complex for enumerated (liệt kê từng trường hợp) rules | Logic is well-defined (xác định rõ ràng) and rarely changes |
| Large datasets available | Need 100% deterministic, auditable (có thể kiểm chứng) output |
| Pattern recognition in unstructured data (dữ liệu phi cấu trúc — ảnh, text, âm thanh) | Regulatory compliance (tuân thủ quy định pháp lý) requires full explainability |
5. Discussion Questions
Q1 — Boundary of AI: A traditional navigation app shows fixed routes. A modern one reroutes in real-time based on predicted (dự đoán) traffic. At what point does the app cross from "software" into "AI"? What specific capability creates that boundary (ranh giới)?
Q2 — The Transparency Trade-off: A bank uses a rule: "Reject if credit score < 650." A competitor uses an ML model with 20% higher accuracy but cannot explain individual (từng cá thể) decisions. From a regulatory (quy định pháp lý) standpoint, which is preferable, and under what conditions would you change that preference?
Q3 — The Data Dependency Problem: If a hiring AI is trained on 10 years of biased (thiên lệch) historical hiring decisions, what does it learn? What does this imply about the claim that "AI is objective because it's just math"?
Made by Anh Tu - Share to be share