I once ended up at a SIGGRAPH dinner with Ken Perlin, who, in good humor, ranted about the death of parametricism at the hands of machine learning zealots. Let’s frame the fight.
Parametric associative modeling, exemplified by constraint graphs, rule engines, and systems like Grasshopper and GenerativeComponents, is the Swiss watchmaking of digital design tooling. It’s symbolically encoded and mostly human debuggable until complexity spirals beyond control. That’s the Robots team.
On the other side, as Andrej Karpathy recently described, LLMs are “people spirits, stochastic simulations of people where the simulator is an autoregressive transformer.” Inference engines–for text, image, video, and eventually everthing else–driven to consume knowledge. They eat brains, this is the Zombies team.
After prompting ChatGPT-4o on Robots vs Zombies capturing the shift from a priori knowledge embedded in parametric modeling networks to frozen knowledge calcified in model weights:
We are moving from programmatic systems (robots) to trained systems (zombies).
Robots = symbolic, rule-based, human-debuggable logic networks (e.g., parametric models, explicit rule engines, constraint graphs)
Zombies = end-to-end learned, data-distilled function approximators with frozen epistemologies (e.g., transformer weights, diffusion priors, RL-trained agents)
Implications: Knowledge engineering shifts from authorship and transparency to harvesting and distillation
The fight-club metaphor, while rhetorically satisfying, is also reductive. It obscures genuine complexity of the problem space and the practical necessity of hybrids. Contrary to a zero-sum fight, a useful metaphor proposed by creative studio Takram is the pendulum: design oscillates between determinism and emergence, between authorship and automation. For example, somewhere along the frontier between Robot and Zombie we find agent orchestration frameworks such as LangGraph where LLMs call symbolic tools, DAGs govern control flow, and human-LLM interactions loop back through rule-based validation.
Polemically, this contest deserves a formal panel: a staged competition where partisans earn points for Team Robot or Team Zombie. Practically, there will be no single victor; the real frontier is in harnessing the combined strengths of both.