Three Waves of AI Development and Their Core Technologies

The article, written by a product manager from Tencent Cloud’s AI & Big Data product center, provides a systematic overview of artificial intelligence for readers without a product or algorithm background.

Preface – By reflecting on history, the author argues that understanding past AI cycles helps grasp current trends.

AI Development’s Three Major Waves

First Wave – Search & Reasoning (≈ 1950‑1970s)

The era began with Alan Turing’s 1950 paper introducing the Turing Test, a method to judge machine intelligence by indistinguishability in conversation. The Loebner Prize rewards successful tests. A satirical code screenshot (removing the Chinese character “吗” and replacing “?” with “!”) is shown as an example of “AI core code”.

The 1956 Dartmouth Conference, organized by John McCarthy and Marvin Minsky, coined the term “Artificial Intelligence”.

Search and reasoning are illustrated with a maze‑solving problem. The author explains how the problem can be represented as a graph and solved by Depth‑First Search (DFS) or Breadth‑First Search (BFS).

Achievements & Problems of the First Wave

IBM’s Deep Blue defeated world chess champion Garry Kasparov in 1997, showcasing brute‑force search. However, the wave was limited to well‑defined “toy problems” because of insufficient computing power and lack of methods for ill‑structured real‑world tasks.

Second Wave – Expert Systems (≈ 1980‑1990s)

ELIZA (1966) is highlighted as an early conversational program that used keyword matching and simple pattern substitution. The author notes ELIZA’s influence on modern chatbots.

Rule‑based expert systems such as MYCIN (diagnosing blood infections) and the Cyc project are described. MYCIN’s 500 rules achieved a 69% correct diagnosis rate, surpassing non‑specialist doctors.

IBM’s Watson, which won the TV quiz show “Jeopardy!”, exemplifies the commercial success of expert systems, though the approach suffers from high knowledge‑base construction costs and difficulty handling vague, unquantifiable problems.

Third Wave – Machine Learning, Deep Learning, and Computer Vision

The author contrasts the “bird‑flying school” (trying to mimic human intelligence) with the “statistical school” (using data‑driven methods). Machine learning belongs to the statistical side.

Feature extraction is explained using a gender‑classification example: physical attributes (hair length, skin tone, height, weight, chest circumference) are encoded as a numeric vector. The Nearest Neighbor (NN) classifier is introduced, followed by extensions such as k‑NN voting and the Support Vector Machine (SVM) . Other classifiers (Bayes, decision trees, neural networks) are briefly mentioned.

Deep Learning is presented as a solution to automatic feature learning, moving beyond hand‑crafted features. An illustrative diagram of multilayer neural networks is included.

Computer Vision is described as the intersection of image processing and machine learning. Images are explained as three RGB matrices (or a single grayscale matrix). Basic concepts such as pixel intensity, color channels, and common feature descriptors (Gabor, SIFT, LBP) are introduced.

Achievements & Problems of the Third Wave

Machine learning and deep learning have propelled advances in computer vision, speech, and natural language processing. The author notes that most current AI systems are narrow (weak AI) and far from achieving general artificial intelligence.

Conclusion

AI, a concept existing for decades, has been revitalized by recent breakthroughs in computing power and algorithms. The article briefly summarizes the three historical waves and their representative technologies.

Promotional Note

Tencent Cloud’s Face Recognition service, based on Youtu technology, offers detection, analysis, search, verification, and liveness detection for various scenarios. Readers are invited to click “Read Original” to try the service and to join upcoming community events.