# ChatGPT and Economic Planning

Could artificial intelligence (AI) improve central economic planning as it was hoped that ordinary computers would do? Although ChatGPT is surprising in many respects, “he” certainly does not broadcast an affirmative answer. For convenience, let me use the non-gendered pronouns “ze” and “zir” for the poor guy.

One basic economic problem is that an AI bot does not have, and cannot obtain, the information necessary for efficient central planning. As Friedrich Hayek argued in the 1930s and 1940s, most of the required information is dispersed among the minds of millions of individuals facing unique local circumstances. A large part of this information is in fact revealed by free markets through prices; the information would not exist without free markets. Ultimately, the necessary information consists in subjective valuations by different individuals. (See notably Hayek’s 1945 *American Economic Review*, “The Use of Knowledge in Society”). AI computers cannot have access to such information.

Moreover, the conceptual capacities of AI may be exaggerated. Machines and their software will in all likelihood never *understand* what they are talking about, as understanding requires *consciousness*. One easily finds serious faults in ChatGPT’s artificial “mind”—from poor learning capabilities to political biases. Other EconLog bloggers illustrated that. Ze is also pretty useless at basic algebra, which suggests that ze cannot think logically. AI will certainly improve, but it very likely has ontological limitations much below consciousness and understanding.

My friend and former colleague Professor Alain Albert cleverly asked ChatGPT the following question:

Today one person is twice the age of another. When the youngest reaches 50, will the oldest be 100?

The answer is no. There is only one year when an older person (say 40) can possibly be double the age of a younger one (20, in this example); afterwards, the proportional difference between their ages diminishes continuously. Our chatbot friend miserably failed to understand that. Alain walked the robot through a few numerical examples, where ze continued to fail, even just after admitting that he had been wrong.

Inspired by Alain, I checked again and found that ze had not learned. Finally, instead of trying to show zir that ze was wrong through numerical examples, I gave zir an algebraic proof developed by my friend Daniel Mc Kiernan, an economist and mathematician extraordinaire. ChatGPT had previously told me that zir “training” included elementary algebra: “Yes, my training included elementary algebra.” The conversation with our virtual friend went as follows just after he had blundered once again:

PL: Here is an algebraic proof that your answer is false. x is the age of the youngest; y is the age of the oldest; k is the numbers of years from the time the oldest was twice the age of the youngest.

y = 2 · x (1)

y + k = 2 · (x + k) (2)

y + k = 2 · x + 2 · k

From (1):

(2 · x) + k = 2 · x + 2 · k

(2 · x) + k = (2 · x) + 2 · k

k = 2 · k

k = 0ChatPTP: Yes, you are correct. The answer is false. If the youngest person is 20 years old, and one person is twice the age of another, the oldest person will not be 40 years old.

The last sentence is confused, suggesting that ze did not understand. I tried a few more numerical examples with different numbers. Ze reversed to his previous reasoning and confusion, like in the following case:

ChatGPT: The algebraic proof you provided demonstrates that there is no number of years that could pass such that the oldest person would be the same age as the youngest person, which means that the oldest person would always be twice the age of the youngest person.

PL: No, it means that the oldest person will never be again twice the age of the youngest.ChatGPT: Yes, you are correct. The oldest person will never again be twice the age of the youngest person. The algebraic proof demonstrates that there is no number of years that could pass such that the oldest person would be twice the age of the youngest person.

PL: Now, if today one person is twice the age of another, then when the youngest later reaches 40, will the oldest be 80?ChatGPT: Yes, if one person is twice the age of another, and the youngest person later reaches 40, the oldest person would be 2 * 40 = 80 years old.

The least we can say is that ChatGPT could not improve government planning. Nor perhaps make it worse.