A story that gets me every time: a computer-vision system in a warehouse cannot reliably distinguish grey wooden pallets from a grey concrete floor. The engineering response was not to improve the cameras, the lighting, or the dynamic range of the model. It was to paint the physical pallets neon green.

From a budget perspective that is rational. From a philosophical perspective it is a small earthquake. We have just modified physical reality to make the simulation work. The map is dictating the territory.

I just published a long-form essay that takes that story as its starting point.

The Baudrillardian Digital Twin

The Baudrillardian Digital Twin: Engineering Against Hyperreality

The essay applies Jean Baudrillard's 1981 Simulacra and Simulation as a diagnostic lens for industrial AI systems, with NVIDIA's Omniverse industrial AI stack (Omniverse, Isaac, Metropolis, cuOPT) as the worked example. The thesis: digital twins drift through Baudrillard's four orders over time. A freshly-commissioned twin starts as a faithful reflection. The most dangerous slippage is between Order 2 (smoothed renderings of sparse data) and Order 3 (the system asserting evidence it does not have). The final stage is the AI Gym: agents trained, evaluated, and tuned entirely inside the simulator while the physical floor becomes a deployment target for decisions already made elsewhere.

The Wider Thread on LinkedIn

The essay above is part of a larger project I have been writing on LinkedIn this spring: an attempt to recover the philosophical foundations that engineers walk past every day. The thread there is shorter-form and runs roughly twice a week.

A few of the recent ones:

• Hallucinations Aren't Bugs: They're Pure Reason Without Experience. Kant's transcendental illusion (1781) as a diagnosis of what LLMs do when they cannot say I don't know. The post that opened the series.
• The Beetle in the Box: What AI Consciousness Actually Means. Wittgenstein on private language and what it tells us about the consciousness-claims industry currently surrounding LLMs.
• Your AI Knows 12,000 Dimensions But Can Only Speak in 2D. The Tao of latent space: the geometry of an embedding space and what we lose at the embedding-to-text bottleneck.
• Why Every AI Engineer Should Read Philosophy. The synthesis post that hit 50,000 impressions and 371 new followers in 48 hours. Maps the practical engineering payoff of philosophical literacy.

More posts in the pipeline on Dempster-Shafer evidence theory (which sits behind the Order 3 defence in the Baudrillardian essay above), the Region Connection Calculus (RCC-8, which sits behind the Order 4 defence), and the structural walls AI will never break regardless of how much compute we throw at them.

Follow Benny Cheung on LinkedIn

The Through-Line

The pattern I keep finding: engineers build forward until they hit a wall. When they look up, a philosopher has been sitting on that wall for centuries. The Baudrillardian essay above is the longest treatment of that pattern I have published so far. The LinkedIn series is the working notebook. Both are arguing the same thing. Technical mastery without conceptual foundations is going to crash into those foundations eventually, and the engineers who can articulate what the code is supposed to mean are the ones who catch the subtle failures that the metrics miss.

Thanks for reading.

Benny