Why AI Has Only a Seven‑Year History and What AI Infrastructure Means for the Future

Editor’s note: This article is a transcript of academician Wang Jian’s speech titled “AI, AI+ and AI Infrastructure” delivered at the 2024 Bund Conference on September 5. The speech offers fresh personal insights into AI, AI+, and AI infrastructure.

Today’s artificial intelligence is fundamentally different from the AI discussed in the early 1980s.

If you cannot build something better than ChatGPT, two constraints are at play: first, your technology – the foundational model; second, the depth of your understanding of the problem domain.

When we talk about infrastructure, we are witnessing the ultimate form of technology penetration. Historically, any technology that becomes a lasting influence turns into infrastructure.

Data is a core component of AI infrastructure; it is not merely an accessory to models or computation. Only when data, models, and compute are integrated into a complete infrastructure can truly exciting innovations emerge.

AI, AI+ and AI infrastructure together illustrate that technology, mechanisms, and infrastructure are all undergoing simultaneous revolutions.

Wang identifies three keywords: AI, AI+, and AI infrastructure. He notes that AI means different things to different people, but these three aspects are now intertwined.

He recalls that artificial intelligence has a long past but a very short history, questioning what AI actually means today. He references the 1940s Turing paper on “intelligent machines” as an early milestone.

He mentions the Dartmouth conference and the term “Cybermetics,” suggesting that the field’s naming has evolved over time.

Wang highlights the 1980s introduction of neural networks (e.g., the PDP textbook) and notes the limited scale of early models compared to today’s massive systems.

He emphasizes that modern AI, starting around 2017 with Google’s Transformer, differs dramatically from pre‑2017 AI.

He discusses the emergence of OpenAI, the 2022 release of GPT, and the significance of the “token” concept for pricing and infrastructure.

Wang contrasts Google’s strengths and weaknesses, noting that while Google pioneered many innovations, it has not produced a product as impactful as OpenAI’s GPT.

He argues that the real innovation mechanism lies beyond simple 0‑to‑1 breakthroughs; it involves deeper understanding of problems and the integration of data, models, and compute.

He points out that ChatGPT is not just an application but an application platform built on a foundational model (GPT) plus a chat interface.

Wang observes that the rapid scaling of data, models, and compute (often by a factor of 1,000) necessitates new AI infrastructure.

He defines AI infrastructure as the ultimate form of technology penetration, comparable to how cloud computing became the backbone of the digital era.

He presents slides showing AI unicorns backed by top cloud providers (AWS, Azure, etc.) and notes that Alibaba Cloud is notably absent, highlighting competitive differences.

Wang concludes that when AI, AI+, and AI infrastructure are considered together, we see simultaneous revolutions in technology, mechanisms, and infrastructure, creating an unprecedentedly exciting future.

Thank you!

Author: Wang Jian Source: “Smart Hyperparameters” public account