TLDR: This research paper by Michael I. Jordan proposes a new direction for AI development, moving beyond the current focus on human-like intelligence and individualistic perspectives. It advocates for a “collectivist” approach that integrates economic and social concepts with computational and inferential thinking. The paper argues that real-world intelligence is deeply social and involves navigating uncertainty, and that AI systems should be designed with social welfare as a primary goal. It explores how economic principles like markets and incentives can be used to build more equitable and effective AI systems, citing examples in music and data markets. The author also calls for a new educational paradigm that blends computational, economic, and inferential thinking to address the complex societal challenges posed by advanced information technology.
The ongoing conversation around Artificial Intelligence (AI) often swings between extreme hype and dire warnings, a discussion that can feel disconnected from the practical realities of technological development. This paper, “A Collectivist, Economic Perspective on AI,” challenges this narrow view, proposing a more holistic approach that integrates economic and social principles with computational and inferential concepts.
Historically, the term “AI” emerged in the 1950s, but much of the actual progress in information technology over the decades came from areas like hardware, networks, and eventually, machine learning (ML). The recent rise of large language models (LLMs), built on ML principles, has brought the term “AI” back into prominence. However, the paper argues that focusing solely on building “thinking machines” that compete with humans is a limited aspiration. It overlooks the fundamental truth that humans are social beings, and much of our intelligence is social and cultural in origin. Furthermore, this traditional view often treats the societal impact of technology as an afterthought, which is no longer acceptable given AI’s profound influence.
The core argument is that the path forward for AI is not just about more data or computational power, but about designing systems where social welfare is a primary consideration. This requires a “collectivist” perspective, recognizing that intelligence in the real world isn’t just about knowing facts, but about navigating uncertainty and interacting effectively with others whose knowledge is also partial. The paper highlights that social environments, while introducing uncertainties like information asymmetry, also foster cooperation and information sharing, which can mitigate these uncertainties and improve decision-making.
Interestingly, the paper suggests viewing LLMs not just as single, human-like entities, but as “collectivist” artifacts. When you interact with an LLM, you are implicitly engaging with the vast collective of humans who contributed data, opinions, and language to its training. In this sense, an LLM can be seen as analogous to a culture—a repository of shared narratives, opinions, and abstractions.
A crucial element of this collectivist perspective is the integration of economic thinking. The paper emphasizes that economics provides not only a framework for understanding social interactions but also a source of “mechanisms” or distributed algorithms. Concepts like markets, incentives, information design, and contracts are vital for building AI systems that enhance the human condition. Markets, for instance, are presented as intelligent collectives that make real-world decisions, create and distribute value, and manage risk, even before the advent of modern computing.
The paper illustrates these ideas through examples of online markets. Traditional recommendation systems, while collectivist in nature, often lack direct economic incentives for creators. A proposed three-way market for recorded music, involving musicians, listeners, and brands, demonstrates how integrating incentives can create a more equitable system where artists are directly compensated when their music is used. Similarly, in data markets, where platforms sell user data to third parties, the paper discusses the critical issue of user privacy. It suggests that platforms could provide formal privacy guarantees (e.g., by adding noise to data) to incentivize user participation, creating a complex interplay between user preferences, platform incentives, and data buyer demands.
To truly advance AI, the paper advocates for a blend of three distinct but complementary thinking styles: computational, inferential, and economic. Computational thinking focuses on algorithms and modular design. Inferential thinking deals with extracting value from data, understanding underlying populations, and quantifying uncertainty, especially for unseen entities. Economic thinking, through fields like game theory and mechanism design, focuses on designing incentives to shape the behavior of strategic agents, aiming for social welfare or revenue. While academia has developed fields that combine two of these (e.g., machine learning blends computation and inference; econometrics blends economics and inference; algorithmic game theory blends computation and economics), the paper argues for the necessity of a tripartite blend—a “missing middle kingdom” in AI education.
Also Read:
- Rethinking AI’s Foundation: Why Optimization Limits True Artificial Cognition
- Establishing a Scientific Foundation for Measuring Artificial Intelligence
This integrated perspective offers a more nuanced way to address complex issues like fairness, privacy, ownership, and transparency in AI. It also acknowledges the vital contributions of other disciplines, such as cognitive science, social psychology, and the humanities, particularly behavioral economics, in understanding how human behavior influences and is influenced by AI systems. Ultimately, for AI to mature into a robust engineering discipline, it needs more than just increased data and compute; it requires modular, transparent design concepts and a foundation built on overarching scientific and humanistic principles like rationality, experimentation, cooperation, and empathy. For more details, you can refer to the full research paper: A Collectivist, Economic Perspective on AI.
