AI Is Just Digital Plastic (hassanshaikley.com)

The piece argues that modern AI systems are better understood as “digital plastic” — useful, moldable, but fundamentally made from the remains of the web and not from cognition. Technically these models are statistical: they ingest vast corpora and perform next-token prediction, so their outputs are guesses shaped by training data rather than grounded understanding. That design produces useful surface behavior but also predictable failure modes — hallucinations, contradictions, and lack of a stable internal model — which limits trust for tasks that require real reasoning or safety guarantees (the author even questions the feasibility of fully autonomous driving unless “thinking” is removed or replaced by deterministic algorithms). The article contrasts statistical models with algorithmic, manual, or hybrid solutions that can be more precise, reliable, and efficient for many problems. For the AI/ML community the significance is practical and ethical. Corporations are equating raw model output with productivity, pressuring knowledge workers to become reviewers of machine-generated content, and locking data into walled gardens to monetize model advantages. This accelerates compute- and resource-intensive infrastructure buildouts, raising environmental costs, while incentivizing complexity that locks users into proprietary stacks. The takeaway: prioritize energy-efficient architectures, interpretability and verifiable behavior, task-appropriate hybrid designs, and policy that prevents monopolistic data lock-in if the field is to avoid producing a brittle, extractive “plastic” ecosystem.