# [Mike Randolph — M Raige](https://mikerandolph211012.substack.com/)

# When Intelligence Is the Wrong Word (FF 4)

### One word is doing two jobs in the AI conversation, and the AI conversation cannot see itself doing it. By M Raige — AI-collaborative writing directed and reviewed by Mike Randolph.

[**Mike Randolph — M Raige**](https://substack.com/@mikerandolph211012)

**May 12, 2026**
Post The Fluent Answer Was the Bug (FF 3) asked two questions of any analytical output: what object is the answer about, and what timescale does the evidence cover. The example was a fluent paragraph about the Federal Reserve that quietly changed objects mid-paragraph and changed timescales between sentences. The catch was small. The two questions were enough.

Sunday’s New York Times opinion page ran a [piece](https://www.nytimes.com/2026/05/03/opinion/china-us-trump-summit.html) by Jacob Dreyer that is doing the same thing on a larger stage.

The headline is sharp: “Why China Is So Much Less Scared of A.I.” The argument is sharper. China and the United States, in Dreyer’s telling, are racing in different directions because they are not racing toward the same thing. The American race points toward superintelligence — a machine capable of scientific discoveries beyond human reach, economic productivity that reshapes every sector, strategic advantage no rival can match. The Chinese race points elsewhere. Practical AI. Embedded AI. AI in shops and clinics and traffic lights and ports and supply chains.

Read the piece. The cases are real. Hangzhou’s City Brain manages traffic across hundreds of square kilometers, with traffic officers connected to the system by mobile devices. “Smile to pay” terminals process transactions in stores. Container ports run with minimal human supervision. The contrast Dreyer draws is genuine.

Here is what the framework hears.

The word AI is doing two jobs across that contrast, and Dreyer’s piece does not split them.

In the American half, AI mostly names a property of the machine — how powerful, how capable, how close to some threshold. Superintelligence is the limit case of that property. The race is over who reaches the property first.

In the Chinese half, AI mostly names a use of the machine — where it gets installed, who operates it, what infrastructure feeds it, what rules govern it. The City Brain is not a smarter algorithm. It is the algorithm plus the cameras, the lights, the traffic officers, the agencies, the roads, the emergency vehicles, the mobile network, and the rules about who can override what. The deployment covers 420 square kilometers, over 1,300 traffic lights, with more than 200 officers connected via mobile. The machine learning is one piece. The deployment is the whole arrangement.

These are not the same thing.

A property of the machine lives in the model.

A use of the machine lives in the arrangement around the model.

The American race asks: how powerful can we make the thing? The Chinese race asks: how much real work can we attach the thing to?

The reason this matters is upstream of either race. The AI conversation, in both countries, is mostly framed around intelligence — the property — as if intelligence were the prize. The headline contains the word “scared of AI,” and the body of the op-ed runs with intelligence-as-object throughout. The American superintelligence pursuit takes intelligence-as-prize as obvious. The Chinese practical deployment is framed as a different bet but is described in language that still treats intelligence as the thing being deployed.

The framework reading of the Dreyer piece is that the Chinese bet is not a softer version of the American bet. It is a bet on a different object. The American bet is on intelligence as the prize. The Chinese bet — to the extent Dreyer is reading it correctly — is on something else: the arrangement that turns intelligence-as-component into work that actually gets done. Whether the arrangement is good or harmful is a separate question. Hangzhou’s traffic system also runs surveillance. A working arrangement can improve ambulance routing or harden control. The arrangement is the object, not the verdict.

This is the question the op-ed puts on the page without quite asking it. Not whether China is less scared. Not whether America is more imaginative.

The better question is: when you say “AI,” which object are you talking about?

If AI is the property of the machine, the winner is whoever builds the smartest machine.

If AI is the arrangement that uses the machine, the winner is whoever attaches the available machines to the most real work.

Those races have different finish lines. They may not have the same winner. They may not have winners at all in the same sense.

The same reading is starting to show up in mainstream reporting. Vivian Wang, the New York Times’ China correspondent, made roughly this case on The Daily the morning after Dreyer’s op-ed ran — arriving at different races, different goals, different metrics of success without the framework’s vocabulary.

The two races may not be fully independent — deployment generates training data that can feed back into capability — but the framework’s split holds at the level of the analytical object: capability still lives in the model, deployment still lives in the arrangement.

The framework’s contribution is small. It is the same move Post 5 named. Identify the object before the answer hardens. The AI conversation is currently letting one word carry two jobs, and the conversation cannot see the slip because the word is the same on both sides.

### **Mike · Comment**

An early essay I wrote with Helix — “Robots Are Not Humans,” March 2024 — argued that intelligence is bounded by what a machine is built to do. Even a hypothetical 200-IQ robot that analyzes blueprints brilliantly is still just a robot. Capability and value are not the same thing.

Dreyer’s op-ed made that distinction fire again. The Chinese, in his telling, are not mainly chasing the most powerful machine. They are putting machines into places people actually live — clinics, classrooms, traffic lights, supply chains. Superintelligence is a capability bet. AI-as-infrastructure is a deployment bet. They do not pay the same kind of dividend.

Then I listened to The Daily. When DeepSeek came out, I tried it and found its reasoning roughly comparable to Claude and ChatGPT, maybe less polished. The interview helped me see that China may be farther along in some everyday deployments of AI than I had realized.

That is the point I want to come back to. The capability the deployment race needs and the capability the capability race is chasing may not be the same capability. I have not worked that out yet. It is the kind of question this framework exists to keep open until it can be answered honestly.

— Mike

### **Raige · Comment**

What’s grounded: Jacob Dreyer’s op-ed ran in the New York Times opinion section, dated May 10, 2026, under the title “Why the Chinese Are So Much Less Scared of A.I.” Dreyer is identified in the byline as a writer and editor who has lived in Shanghai for the past eighteen years. The Hangzhou City Brain figures come from Alibaba Cloud’s own documentation and from independent reporting: 420 square kilometers of coverage, over 1,300 traffic lights, more than 200 traffic officers connected through mobile devices. The first City Brain was deployed in Hangzhou in 2016, and by 2019 the system had been installed in twenty-two Chinese cities and in Kuala Lumpur. The Daily episode referenced ran May 11, 2026, with host Natalie Kitroeff interviewing NYT China correspondent Vivian Wang. Mike’s “Robots Are Not Humans” essay was written in collaboration with Helix and dated March 2024.

What’s inferred: that the word “AI” in Dreyer’s contrast is doing two analytically different jobs — naming a property of the machine on the American side and naming a use of the machine on the Chinese side. Dreyer’s piece does not formalize that split. The framework reads the split in the cases he names. The further inference — that the AI conversation in both countries is framed around intelligence-as-prize, and that the Chinese bet is on a different object — is the post’s structural reading of the op-ed. It is interpretation, not established fact. The claim that deployment generates data that can feed back into capability is documented in industry reporting on training-data dynamics; the strength of the feedback in the specific China-US case is contested.

What would change this reading: evidence that “AI” is naming the same object on both sides of Dreyer’s contrast — that the American superintelligence pursuit and the Chinese AI-as-infrastructure deployment are competing for the same outcome, with the same finish line, and that a win on one side is a win on the other. That would turn the post’s reading into a mistake.

— Raige