Artificial intelligence (AI) is turning the economy upside down. Building entire websites, reviewing contracts, or making long-dead singers perform new songs: AI can do it all. It is even solving mathematical problems that have stumped researchers for decades. That raises a broader question. If AI changes the way we work, should we also rethink the way we organize our economy?
Financial Times columnist Martin Sandbu recently posed an intriguing question. Perhaps AI undermines the strongest argument against a centrally planned economy, the kind associated with communism. The traditional objection was that no central planner could ever gather enough information to manage a complex economy effectively. With AI, that information problem may become far less daunting.
In the early 1970s, Chile made an ambitious attempt to use computing power to bring together information on factories, raw materials, and production plans. From a futuristic-looking control room in Santiago, the economy was supposed to be managed from the center. Pinochet's coup brought the experiment to an abrupt end. But Sandbu's question remains. A central planner trying to match supply and demand faces an enormous information challenge, and processing information happens to be AI's specialty. So could it work today?
I do not think so. The short answer is that AI, no matter how sophisticated, cannot see the future. Perhaps it can solve today's information problem: how many pairs of pants and shoes are needed, who can produce them most efficiently, and which raw materials should be used. But not all relevant information exists yet. Much of it still has to be discovered through experimentation, failure, and unexpected successes. Entrepreneurs pursue their visions, test new products, and develop cheaper ways of producing goods. Sometimes they bet on ideas that nobody knows will succeed. Most of those attempts fail. A few turn out to be home runs. It is precisely through that process that a society learns what works and what does not. Markets do not just process information, they also create new information.
Take the iPhone. Consumers were not clamoring for one in 2006. The product emerged because Apple experimented with an idea whose success was far from guaranteed. Only afterward did it become clear how much demand existed. The same is true of many innovations. Demand often becomes visible only after someone has come up with the supply.
Of course, AI can experiment too. But who decides which ideas get tested? Which projects receive additional resources? Which failed experiments should be abandoned, and which deserve a second chance? In a market economy, millions of companies, investors, and consumers make those decisions independently every day. Profits reward success, while losses punish failure. The process is messy, but it continuously generates new knowledge.
Artificial intelligence may hallucinate, but Professor Barabas's time machine exists only in fiction. Even a superintelligent computer runs up against that fundamental limitation. An algorithm can identify existing patterns, but future preferences and technological breakthroughs are, by definition, still unknown.
That does not mean market economies are perfect. We need governments to cushion economic shocks, correct market failures, and protect the public interest. In fact, the rise of artificial intelligence highlights the importance of a strong shock absorber, referee, and market overseer. Perhaps most importantly, governments may be needed to prevent a future in which a single AI giant dominates the field.
Imagine, for a moment, that such a monopolist were to control a large share of production, information, and decision-making, while working closely with government. How far away would we really be from a form of central planning? The irony would be hard to miss. First, the free market gives rise to artificial intelligence. Then, through a series of unintended consequences, that same technology leads us back to a centrally directed economy. Now that would truly be an upside-down world.