After 2016, AI was no longer science fiction—it became reality.
You unlock your phone with face recognition.
You speak to a speaker, and a voice assistant responds.
You scroll through short videos, and the recommendation algorithm knows what you like.
You drive somewhere, and navigation avoids traffic.
AI has penetrated every aspect of life.
AI Industrialization#
After AlphaGo, AI moved from labs to industry.
Tech giants invested in AI:
- Google: TensorFlow framework, TPU chips, Google Assistant
- Facebook: PyTorch framework, face recognition, content recommendation
- Microsoft: Azure AI, Cortana, invested in OpenAI
- Amazon: AWS AI services, Alexa
- Apple: Core ML, Siri
- Baidu: PaddlePaddle framework, Apollo autonomous driving
- Alibaba: DAMO Academy, City Brain
- Tencent: AI Lab, medical AI
AI became standard for tech companies. Without AI, you’re behind.
AI Application Scenarios#
Face Recognition
In 2017, iPhone X introduced Face ID; face recognition went mainstream.
In China, face recognition is used for:
- Phone unlocking, payment verification
- Security monitoring, criminal tracking
- Attendance checking, access control
- Train station, airport security
But face recognition also raised privacy concerns. San Francisco banned government use of face recognition in 2019.
Voice Assistants
Smart speakers became the new home entry point:
- Amazon Echo (2014 release, popular after 2016)
- Google Home (2016)
- Apple HomePod (2018)
- Xiaomi Xiao Ai (2017)
- Alibaba Tmall Genie (2017)
Voice interaction became a new computing paradigm.
Recommendation Systems
Toutiao, Douyin, and Kuaishou use AI to recommend content.
Users don’t need to search; AI automatically pushes content they’re interested in.
This changed how information is obtained but also brought “information cocoon” problems.
Autonomous Driving
Tesla Autopilot kept evolving.
Waymo launched unmanned taxi service in Phoenix.
China’s Baidu Apollo, Pony.ai, WeRide, and other companies are also testing autonomous driving.
But autonomous driving still faces technical, regulatory, and ethical challenges.
Medical AI
AI started assisting doctor diagnosis:
- Skin cancer recognition
- Eye disease detection
- Lung nodule screening
- Pathological slide analysis
AI can improve diagnostic efficiency and reduce missed and wrong diagnoses.
Financial AI
AI used for:
- Risk assessment: Analyze loan applicant credit
- Fraud detection: Identify abnormal transactions
- Robo-advisors: Automatic investment
- Customer service bots: Answer customer questions
Education AI
AI used for:
- Adaptive learning: Recommend content based on student level
- Essay grading: Automatic scoring and feedback
- Oral evaluation: Assess pronunciation accuracy
AI Chips#
AI needs lots of computation; traditional CPUs aren’t enough.
GPU: NVIDIA’s GPUs became standard hardware for AI training. NVIDIA stock rose from about $30 in 2016 to over $800 in 2024.
TPU: AI-specific chips developed by Google for TensorFlow.
NPU: Neural network processors developed by Huawei, Apple, Qualcomm, and others for mobile AI.
AI chips became a new hotspot in the semiconductor industry.
AI Frameworks#
AI development needs framework support:
TensorFlow: Developed by Google, open-sourced in 2015, one of the most popular AI frameworks.
PyTorch: Developed by Facebook, open-sourced in 2017, researchers’ favorite.
Keras: High-level API, simplifying TensorFlow use.
PaddlePaddle: Developed by Baidu, China’s most popular AI framework.
These frameworks lowered the AI development barrier and promoted AI adoption.
AI Ethics Issues#
AI development also brought ethical issues:
Bias
AI models may inherit bias from training data. For example, hiring AI might discriminate against women because historical data had more men.
Privacy
AI needs lots of data; data collection may violate privacy.
Employment
AI may replace some jobs. Customer service, translation, driving, and other professions face challenges.
Security
AI may be maliciously used. Deepfake can create fake videos.
Responsibility
Who’s responsible for autonomous driving accidents? Who pays for AI medical misdiagnosis?
Countries started developing AI ethics guidelines and regulations.
The AI Investment Boom#
AI became an investment hotspot:
- AI startups sprang up like mushrooms
- Big companies acquired AI companies
- AI talent salaries skyrocketed
But AI also went through a bubble period. Many AI companies couldn’t deploy and eventually closed.
Truly successful AI companies were those that found specific application scenarios.
AI’s Limitations#
AI after 2016 was mainly Narrow AI—excellent at specific tasks but lacking general intelligence.
AI can recognize images but doesn’t understand what they mean.
AI can translate text but doesn’t understand the meaning of language.
AI can play Go but doesn’t know what Go is.
Artificial General Intelligence (AGI)—AI that can think and learn like humans—remains distant.
But AI’s progress speed exceeded many people’s expectations.
Next Step: Large Language Models#
In 2020, OpenAI released GPT-3, a language model with 175 billion parameters.
It could write articles, write code, answer questions, have conversations…
It demonstrated a possibility: Using lots of data and computation, train general language understanding ability.
Is this a path to AGI?
Tomorrow, we’ll discuss the story of large language models.
Today’s Key Concepts#
Narrow AI AI that performs excellently at specific tasks, like face recognition, speech recognition, and recommendation systems. Narrow AI has no general intelligence and can only work on tasks it was trained for.
AI Chips: Chips specifically designed for AI computation. GPUs for training large models, NPUs for AI inference on phones and other devices. NVIDIA is the leader in AI chips.
AI Ethics: Ethical issues brought by AI development, including bias, privacy, employment, security, and responsibility. Countries are developing AI ethics guidelines and regulations.
Discussion Questions#
- AI has penetrated every aspect of life. What do you think has the biggest impact on your life?
- AI may replace some jobs. Which jobs do you think are most easily replaced by AI? Which are hardest to replace?
Tomorrow’s Preview: Large Language Models—the principles behind GPT and how it changed human-computer interaction.