The AI industry in the US is too far behind. Now China owns it all.

# The AI Industry in the US is Too Far Behind. Now China Owns It All. **TL;DR:** A longtime US resident in China shares firsthand observations of AI's practical application in architecture and medicine, contrasting China's integrated, tool-based use with the US's problematic consumer-facing models, high-cost infrastructure, and ethical shortcomings, arguing that true value lies in AI as a tool for experts, not a replacement for human judgment. ## AI in Practice: From Translation to Architecture After living and working in China for 14 years, language remains a daily challenge. Reliance on translation tools became dramatically easier about five years ago with the advent of real-time AI translation devices. Initial tests were impressive: the software flawlessly translated complex American baseball and football idioms, technical chemistry terms, and high-level financial articles from sources like Zero Hedge. The Chinese output was accurate, even if word order differed slightly. This translated directly into business applications. At large Chinese factories exporting prefabricated homes, AI design tools could instantly generate specifications and quotes for various architectural styles—Southwestern, modern, coastal stilt houses—opening potential new markets. However, the limitations quickly surfaced during a test of "Cape Cod" or "craftsman" style designs. After the AI generated images, engineers immediately rejected them. They recognized features like chimneys and dormer windows as unsuitable for the US market, particularly in areas with heavy snow loads. The AI was essentially performing a rapid image search and importing results into CAD software, lacking the crucial understanding of local building codes, climate demands, and manufacturing feasibility. ### The Core Limitation: Probability vs. Understanding This highlights a fundamental flaw: AI models operate on probability, guessing what the speaker *might* mean. Nuance is lost. For example: - The phrase "we'll handle this next year" was translated as a definite "I want to solve this next year," missing potential contextual meaning of postponement. - Timeframes like "within the next year" are ambiguous. - Critical financial details, like a house price of "$995,000," were mistranslated as "99.5," omitting the "thousand." These errors underscore the non-negotiable need for human oversight. In meetings, a true bilingual interpreter or an intelligent AI agent that knows when to ask for clarification is essential. ## The Chasm Between US Prediction and Chinese Reality in Healthcare The gap between AI promise and practical use is stark in medicine. In 2016, AI pioneer Geoffrey Hinton (often called the "Godfather of AI") predicted AI would revolutionize radiology so thoroughly that "training radiologists now would be a mistake." This prediction proved largely inaccurate. Globally, there is a severe shortage of radiologists, especially in China and developing nations. While China can manufacture the scanners, it lacks enough professionals to interpret the scans. Here, AI is widely and effectively deployed as a *tool* within hospitals. AI courses are integrated into medical school curricula, and systems are trained on data from top radiological centers to assist doctors, especially in elder care, boosting diagnostic efficiency and accuracy. Crucially, China treats citizen medical data as a national security matter and will not share it with Western large language models. However, Chinese researchers *can* access other countries' patient datasets, which are used to train Western models. In China and allied nations, AI is enabling doctors to replicate top-tier diagnostic quality, democratizing expertise. ### The US Consumer AI Problem The US has a different dynamic. It has top radiologists, high salaries, and the latest AI tools *within elite hospital systems*. The problem is at the consumer level. A JAMA study found that 21 tested LLMs had a failure rate exceeding 80% when handling complex medical cases, and 40% even with provided lab results. A quarter of US adults use chatbots for health advice, largely due to cost barriers (41% cite cost or unwillingness to pay for a doctor). Over 60 million Americans ask questions like "Is my rash serious?" or "Do I have a contagious disease?" and receive incorrect answers about 80% of the time. For the average user, LLMs "hallucinate" information, making them unreliable for self-diagnosis. The problem circles back to the critical need for human verification and expertise. ## Economic Unsustainability and Ethical Quagmire From a practical standpoint, AI often feels like "an incredibly fast search tool." It eases work but hasn't delivered the promised disruptive revolution. A telling observation: apart from purchasing existing CAD/CAM software, no one in the business context is paying extra for AI use. This raises a critical financial question. Since ChatGPT-4's launch, capital expenditures at Amazon, Microsoft, Google, Meta, and Oracle have quadrupled. The largest cost is the energy to run LLMs, for which users are not paying. This means these companies are operating AI divisions at a loss, subsidized by other revenue streams. Oracle faces a particular debt maturity crisis. This model is unsustainable. Companies will be forced to abandon the "grow now, monetize later" strategy that has defined the tech industry. The debt cannot be magically erased—it must be repaid at junk-bond rates, converted to equity, or lead to bankruptcy, unless users start paying significantly more. ### Criticisms from Within Karen Hao, an insider turned critic and MIT graduate, offers a scathing assessment. She argues that industry leaders are often mediocre, lacking the inventive genius of past tech pioneers. They oversee vast capital without fully understanding the AI or the business. Hao concludes that these companies, and the world, might be better off without them, citing severe ethical issues: human rights violations, privacy breaches, and labor exploitation. She notes the industry is "overpromising and under-delivering," falling far short of the vision of Artificial General Intelligence (AGI). ## Conclusion: AI as a Tool, Not a Replacement The core lesson from these real-world applications is the enduring importance of the human element. AI is a transformative tool for industry experts—architects, engineers, doctors—who understand its output's limitations and the real-world context it lacks. It cannot replace the judgment of a seasoned contractor who knows local building codes, nor the diagnostic skill of a radiologist, nor the critical thinking of a financial analyst. The US model, focused on consumer-facing chatbots and massive, unprofitable infrastructure, contrasts sharply with China's pragmatic integration of AI into expert workflows for efficiency gains. As other regions begin to question the current trajectory, the sustainable path forward for AI seems clear: it must prove its value as an indispensable tool for professionals, not as a flawed oracle for the masses. **Source:** [The AI industry in the US is too far behind. Now China owns it all.](https://youtu.be/ny_3PRz6Zeg)