What Is Knowledge in AI? Exploring Intelligent Systems and Future Opportunities

1. What is knowledge: feature representations that can solve problems. Human language or graphs are just one form of expressing knowledge. Feature representations learned by AI models are also knowledge as long as they solve problems. The ability to acquire knowledge is the core of AI technology.

2. What is an intelligent system: Whether biological or non‑biological, an intelligent system must contain three elements—observation system, intellect system, and action system. The observation system collects data, the intellect system generates knowledge from data, and the action system uses the learned knowledge to achieve goals.

3. The essence of intelligence is the general ability to acquire knowledge and use it to reach objectives.

4. Historical significance of AI: (1) It is the first system beyond humans capable of creating knowledge; (2) Knowledge creation will accelerate, rewriting technology, society, and the economy.

5. Directions for AI development: (1) More natural human‑machine interaction, following past revolutions such as the mouse, touch screens, and smartphones. (2) More powerful back‑ends, driven by AI’s hunger for compute and data. (3) Defining new‑era experiences, building on the first two points, with examples like the iPhone, Windows 95, and browsers. (4) Full digitalization of the physical world; past partial digitalization includes IBM’s desktop, Microsoft’s enterprise information systems, and Apple’s social and transactional digitalization. AI will digitize every object, tool, vehicle, and venue, making them intelligent and digital.

6. Innovation opportunities brought by AI: Historically, humanity invented partial universal capabilities such as livestock, writing, steam engines, trains, computers, and airplanes. AI now represents the most universal capability because it creates knowledge to solve problems. Knowledge is power, so this universal ability will generate unprecedented opportunities.

7. AI‑driven innovation should follow this loop: (1) Identify an imperfect application scenario. (2) Pose the question: what knowledge is needed to improve it? (Opportunity: sensors). (3) Extract that knowledge (Opportunity: software, hardware, algorithms). (4) Apply the learned knowledge to improve the scenario (Opportunity: business models). (5) Collect new data from the transformed scenario and repeat steps 1‑4, forming a closed loop.

8. Potential revolutions from AI: (1) A brand‑new information industry, as AI requires compute‑enabled sensors and high‑dimensional parallel processing beyond the von Neumann architecture, prompting a rewrite of hardware and software. (2) Emergence of new industries—intelligent objects, robots, autonomous vehicles, smart venues—that will transform all existing sectors.

9. Advancing AI innovation: (1) Capture financial, talent, and data capital, with data capital being a novel revolutionary factor. (2) Require a friendly policy environment and supporting infrastructure. (3) Must be market‑driven.

10. AI startup challenges: Data. While past startups could rely on open‑source software, AI ventures face a near‑insurmountable data barrier.