From Euclid to Transformer, an algorithm is a recipe a substrate can follow without judgment. The cleaner the recipe, the more independence it has from any particular cook.
智能Intelligence
Algorithm. Compute. Data. The three commodities of the new century.
Every intelligent system — biological, artificial, social — is some combination of three commodities. Algorithm is its truth: the procedure by which it transforms input into output. Compute is its goodness: the raw substrate-work that makes the procedure happen. Data is its beauty: the lived examples that give the procedure something to be about. The history of AI is a story of which of the three was scarce in each decade. Today, for the first time, all three are simultaneously abundant.
FLOPs, watts, transistors, silicon. Compute is what turns an idea into an event. The 21st century is largely the story of compute getting almost free in some regimes and almost priceless in others.
Data is the world's testimony, frozen. Without it, an algorithm is a recipe with no ingredients. With it — and enough compute — an algorithm starts to model the world it came from.
How this trinity came to be.
- 1956
Dartmouth
AI is named. The dominant axis for forty years will be algorithm; compute and data are scarce.
- 2012
AlexNet
GPUs unlock compute. ImageNet unlocks data. Algorithms that had failed for decades suddenly work. The trinity rebalances.
- 2022+
Foundation models
All three axes scale together for the first time. The result is a phase transition.
How to use this lens today.
- 01Score any AI project on which axis is binding: algorithm gap (rare), compute gap, or data gap (most common).
- 02National AI strategy is exactly a three-axis investment portfolio.
Where this trinity is heading.
- →Synthetic data and self-play start to break the data bottleneck; the binding constraint shifts back to algorithm.
- →Compute scarcity becomes geopolitics. Wars are increasingly about who can manufacture the substrate of cognition.