AI · Topic 1 of 2
AI Description and AI Technologies
Learn what AI is, the Turing test, key AI milestones, symbolism, connectionism and behaviorism, strong and weak AI, and how AI, ML and deep learning nest.
- 5 min
- AI contains machine learning, which contains deep learning; foundation models are large deep learning models.
- Turing test 1950; Dartmouth and the name "AI" 1956.
- Symbolism uses rules and logic, connectionism uses neural networks, behaviorism uses perception, action and feedback.
What it is
Artificial intelligence (AI) is the science and engineering of making machines that perform tasks which normally need human intelligence. Those tasks include perceiving, understanding language, reasoning, learning and making decisions.
AI technologies are the methods and fields used to build such systems. Today most practical AI is built with machine learning, and most modern machine learning uses deep learning.
How it works
Defining intelligence: the Turing test
In 1950 Alan Turing proposed a test in his paper "Computing Machinery and Intelligence". A human judge chats in text with a hidden human and a hidden machine. If the judge cannot reliably tell which is the machine, the machine passes.
The term "artificial intelligence" was coined for the Dartmouth workshop of 1956, organized by John McCarthy and others. It is usually treated as the birth of AI as a field.
Key milestones
| Period | Milestone | What it showed |
|---|---|---|
| 1950 | Turing test proposed | A practical way to discuss machine intelligence |
| 1956 | Dartmouth workshop | AI named as a research field |
| Late 1950s | Rosenblatt's perceptron | A machine that learns weights from examples |
| 1969 | Minsky and Papert, "Perceptrons" | A single perceptron cannot solve XOR; funding dropped |
| 1970s to 1980s | Expert systems | Hand written rules work in narrow domains |
| 1997 | IBM Deep Blue beats Kasparov at chess | Search plus expert knowledge |
| 2012 | AlexNet wins ImageNet | Deep CNNs on GPUs beat earlier methods |
| 2016 | AlphaGo beats Lee Sedol at Go | Deep learning plus reinforcement learning |
| 2017 | Transformer paper published | Attention based models, later the base of LLMs |
| 2022 onward | ChatGPT and other LLMs | General purpose chat and generation |
AI history had two "AI winters", when early promises failed and funding fell. Each revival came from better algorithms, more data and more computing power.
The three schools of AI
| School | Core idea | Typical method |
|---|---|---|
| Symbolism (logicism) | Intelligence is manipulating symbols with logical rules | Expert systems, knowledge graphs |
| Connectionism (bionics) | Intelligence comes from networks of simple connected units, like neurons | Neural networks, deep learning |
| Behaviorism (actionism) | Intelligence comes from perceiving and acting in an environment, with feedback | Control systems, reinforcement learning, robots |
Modern deep learning is the clearest result of connectionism. Reinforcement learning, which learns from rewards, is close to behaviorism.
Strong AI and weak AI
Weak AI (narrow AI) solves specific tasks, such as face recognition or translation, without real understanding. All AI in use today is weak AI.
Strong AI (often linked to artificial general intelligence) would truly think, reason across any domain and have its own mind. It does not yet exist.
The three elements of AI
Modern AI rests on three elements: data, algorithms and computing power. Data provides the examples, algorithms define how to learn from them, and compute makes training large models possible.
Huawei course material often adds scenarios, meaning real business problems, as the place where these elements create value.
Main technical fields
| Field | What it does | Example tasks |
|---|---|---|
| Computer vision (CV) | Understands images and video | Image classification, object detection, segmentation, face recognition |
| Speech processing | Handles spoken audio | Speech recognition (ASR), speech synthesis (TTS), speaker recognition |
| Natural language processing (NLP) | Understands and generates text | Translation, sentiment analysis, question answering, summarization |
Many modern systems are multimodal, combining text, images and audio in one model.
Huawei's AI stack (by role)
| Layer | Huawei name | Role |
|---|---|---|
| Hardware | Ascend AI processors | Chips that accelerate AI training and inference |
| Enabling software | CANN | Heterogeneous compute architecture between Ascend chips and frameworks |
| Framework | MindSpore (also PyTorch on Ascend) | Build and train models |
| Platform | ModelArts | One stop AI development platform |
| Models | Pangu models | Huawei's foundation models |
A worked example
A bank wants to read handwritten check amounts. A symbolism approach would write rules for each stroke shape, which breaks on unusual handwriting.
A connectionism approach trains a neural network on many labeled images. This is weak AI in the CV field, and it needs all three elements: labeled data, a CNN algorithm and GPU or NPU compute.
How the exam asks about this
- "Which school of AI do neural networks belong to?" → Connectionism.
- "Which event is generally regarded as the birth of AI as a field?" → The Dartmouth workshop in 1956.
- "Is face recognition on a phone an example of strong AI?" → No, it is weak (narrow) AI.
What to remember for the exam
- AI contains machine learning, which contains deep learning; foundation models are large deep learning models.
- Turing test 1950; Dartmouth and the name "AI" 1956.
- Symbolism uses rules and logic, connectionism uses neural networks, behaviorism uses perception, action and feedback.
- All AI deployed today is weak AI; strong AI does not yet exist.
- The three elements are data, algorithms and computing power; the main fields are CV, speech and NLP.
Do not equate AI with deep learning. Expert systems and search, such as Deep Blue, are AI but involve no deep learning at all.
Common mistakes
Calling the 1969 perceptron book the end of neural networks. It showed the limit of a single layer perceptron on XOR. Multi layer networks trained by backpropagation later solved this.
Mixing up symbolism and connectionism. Symbolism encodes knowledge as explicit rules. Connectionism learns knowledge as weights inside a network.
Thinking LLMs are strong AI. Large language models are very capable, but they are still trained for prediction tasks. They are considered weak AI.
Check yourself
What does the Turing test measure?
Whether a human judge, chatting by text, can tell a machine from a human. If the judge cannot reliably tell them apart, the machine passes.
Which three elements does modern AI depend on?
Data, algorithms and computing power. Huawei material often adds real scenarios as the place they are applied.
Which AI school is reinforcement learning closest to?
Behaviorism, because it learns by acting in an environment and receiving feedback in the form of rewards.
Topic 1 of 2 in AI
Test yourself on AI
The full Huawei HCIA-AI H13-311 question bank has 600 questions, with timed practice exams and an explanation for every answer.
This study guide is independent and is not affiliated with, endorsed by, or sponsored by Huawei. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.