Nowadays, AI is primarily associated with LLMs, where the goal is to develop autonomous agents with human-level intelligence (whatever that means). I have always found the goal of creating something with “intelligence” boring for some reason. Maybe it is because the objective is too diffuse—create something at least as smart as humans and hope it figures out <insert your problem here>. Regardless, these LLM-based agents are useful tools and should be used, but these systems are a narrow view of intelligence.

Intelligence is often overlooked when considering it as a property of the collective and not the individual. For example, a market intelligently allocates resources without a central entity due to its unique ability to process and coordinate dispersed knowledge that no single agent could ever possess. Hayek presents an an example of this in The Use of Knowledge in Society.

Consider a new use for tin, or a major source of tin is disrupted, making supply scarce relative to demand causing the price of tin to rise. Producers, seeing the higher price, are incentivized to increase output, and entrepreneurs are motivated to find substitutes for tin. Consumers on the other hand do not necessarily need to know why the price increased. Rather, they adjust their behavior to use tin more sparingly or to seek substitutes. Without any central order being issued, thousands of people adjust their behavior in a way that conserves the scarce resource, guiding it toward its most highly-valued uses.

The methods and systems that can control and fairly distribute finite resources across multiple-agents is intelligence. When you begin to think of this collective or market-view of intelligence you begin to think of the systems that

  • Learn from data to provide better services.
  • Agents can express or reveal their preferences.
  • Algorithms provide the glue between the data and the market.

This distributed view of intelligence is better suited for solving large-scale problems with many iteracting agents, for example, in power grids, transportation, and commerce systems. In my view

$$\text{AI} = \text{Data} + \text{Algorithms} + \text{Markets}$$

When you combine methods from the fields of computer science, statistics, and economics in a new way, you get systems that focus on the allocation of finite resources by understanding, optimizing, and controlling large-scale multi-agent systems. Examples of such fields are Mean Field Game Theory, Mean Field Control, and Statistical Contract Theory.