Artificial Intelligence and the Economy: Myths, Realities, and the Future of Work

1. Introduction

Artificial Intelligence (AI) has established itself as the hottest topic in contemporary society. However, there is a notable gap between technical reality and public perception; the knowledge most people possess about this technology is insufficient to draw reasoned conclusions about its real economic impact.

This knowledge gap creates fertile ground for ideologues and politicians to instrumentalize fear of the unknown. By presenting AI as an imminent existential threat to families’ livelihoods, interventions and regulations are justified that often respond more to political interests than to real economic needs.

2. What Is Artificial Intelligence?

To analyze the economic impact, we must first strip the term of its mystique. The general idea of Artificial Intelligence is to replicate human intellectual capacity through artificial systems. But what is intelligence? We will adopt a functional definition: intelligence is the ability to achieve a goal efficiently, regardless of unforeseen obstacles that arise along the way.

As technology has advanced, we have observed a curious phenomenon: we have been able to develop systems (AIs) that demonstrate great intelligence (solve complex problems, write code, generate art), yet lack their own objectives. They have no will, no desire, no survival instinct. They are optimization tools, not autonomous agents with their own agenda.

3. Demystifying the Technology: How Does a Machine “Think”?

Understanding the basic principles of the technology can help eliminate some of the fears caused by its unfamiliarity and, at the same time, temper unjustified hype.

For those outside engineering, it is easy to imagine that inside the computer there is a digital “brain” thinking, feeling, and learning in real time like we do. The reality is far more prosaic, mathematical, and, for now, limited.

3.A Mathematics, Not Magic

At its core, modern AI (especially Deep Learning and LLMs) is based on advanced statistics and matrix calculus. Imagine a giant mathematical function:

• Multiplications and Additions: At its most basic level, a neural network is nothing more than millions of number multiplications (called “weights”) that are added together.
• Non-Linear Functions: The “trick” is introducing non-linear functions that allow the network to learn complex patterns, not just straight lines.
• Training: This is the learning phase. It involves providing inputs and using a function that quantifies the error of the output to iteratively adjust and improve the matrices’ “weights” until that error is minimized. It is a computationally very expensive process.
• Inference: This is the usual use of AI by the user. An input (text, image) is presented and converted into numbers; these are multiplied by the fixed weights obtained during training, passed through non-linear functions to produce an output. It is much faster and cheaper than training.

When an AI “writes,” it does not reflect. It calculates, based on its training, which word (or token) has the highest statistical probability of appearing next.

3.B The Myth of Constant Learning: “Frozen” in Time

A crucial distinction that is often overlooked is the difference between training and inference:

• Static Models: The vast majority of current AIs (like GPT-4 in its base version) do not learn from each interaction. Once training is completed, their “neurons” (numerical weights) are frozen. If you tell the AI a secret today and delete your chat history tomorrow, the AI will not remember anything because its internal structure has not changed. They have no long-term memory or real-time brain plasticity.
• The Illusion of Memory: What we perceive as memory is a “context window” (short-term memory limited to the current conversation) or engineering tricks such as RAG (Retrieval-Augmented Generation), which allow it to consult an external database, but the central “brain” remains static. Although “liquid learning” architectures are being researched, as of today, AI does not evolve through daily use.

3.C Biomimetic Alternatives: Numenta and the “1000 Brains”

The current approach (Deep Learning) is a statistical approximation that, although effective, differs greatly from how biology works. However, other lines of research exist, such as the Thousand Brains Theory by Jeff Hawkins and Numenta.

• Replicating Biology: These approaches attempt to imitate the physical structure of the human neocortex (cortical columns and reference frames).
• Continuous Learning: Unlike current static models, an AI based on these principles would learn from every interaction, modifying its connections in real time, just as a child does while exploring the world.

3.D The Great Unknown: Is Intelligence Scalable?

Finally, a dangerous assumption in economic projections is the belief that intelligence is linearly scalable. Currently, the industry operates under the premise that “more data + more computation = more intelligence.”

However, we do not know if this is true indefinitely. We could be approaching an asymptote or a point of diminishing returns, where achieving 1% more “intelligence” requires 100 times more energy and money. If intelligence is not infinitely scalable with current technology (Transformers), the arrival of the Singularity or a functional AGI could be much further away than enthusiasts predict, delaying or nullifying immediate mass unemployment scenarios.

4. AI, AGI, and the Singularity

It is crucial to distinguish the terms to avoid falling into science fiction.

• The Singularity: This is the hypothesis that an AI will become intelligent enough to recursively improve itself without human intervention. This would trigger exponential growth in its intelligence (“intelligence explosion”), surpassing in a very short time the sum of all humanity’s cognitive capacity. In this article, we assume this has not occurred and there are no technical indications that it is imminent.

• AGI (Artificial General Intelligence): Unlike narrow AI (which only knows how to play chess or only translate texts), an AGI would be a system capable of performing any intellectual task a human can do. An AGI could learn accounting in the morning and compose music in the afternoon.

  • Short definition: An AGI is a system with the capacity to learn and apply intelligence across domains, matching human cognitive flexibility, but not necessarily possessing consciousness or its own will.

5. Unemployment and Deflation: The Great Fear

The predominant fear is the total replacement of workers, now exacerbated by the development of humanoid robots that threaten not only cognitive work but also physical labor (plumbing, construction, logistics). Let us analyze this from economic logic.

5.A Replacement and the Cost Paradox

From a business perspective, replacing humans with AI only makes sense if the total cost is lower. If this happens massively, and assuming a market with free competition, the increase in profit margins will be temporary. Competition will force final prices down to capture market share.

If we extrapolate this to the entire economy, we face a scenario of technology-driven generalized deflation. Production costs plummet.

• Result: If the cost of living (food, housing, energy, services) falls tenfold, a worker could accept a nominal salary ten times lower while maintaining the same purchasing power.
• Conclusion: In the long term, what matters is not nominal salary (the number of euros), but real salary (what I can buy with them). A highly automated society is, by definition, a society of abundance and low prices.

5.B Niche Jobs: The “Human Plus”

Not everything is efficiency. There is subjective value in the “humanity” of a service. Even if an AI can deliver a perfect lecture or create a catchy pop song, the market values connection:

• Personal Brand and Authenticity: We pay to see that speaker or listen to that singer because of their life story, imperfections, and identity.
• The Human Factor: In sectors such as caregiving, psychology, or even more intimate services such as prostitution, a fundamental part of the service is interaction with another conscious being.
• Political Responsibility: Citizens will likely continue wanting the person who presses the red button or decides laws to be a human who can be ethically judged—something inapplicable to an algorithm.

5.C New Jobs and Comparative Advantage

Economic history teaches us that automation does not create permanent structural unemployment. In 1900, agriculture employed most of the population; today, a tiny fraction feeds everyone. The result was not collapse, but enrichment and the creation of previously unimaginable sectors (IT, tourism, entertainment).

Returning to the example of Robinson Crusoe: if he automates fishing, he does not become “unemployed”; he frees his scarcest resource (his time and mind) to satisfy new needs (building a house, gathering coconuts).

• The Mind as a Scarce Resource: As long as unmet human needs exist, there will be work. AI frees the human mind from repetitive tasks to focus on higher-order problems. Even if AIs comparable to the human mind existed, their number would also be limited by economic constraints.
• Ricardo’s Law: Even if an AI is better than us at everything (absolute advantage), humans will still retain a comparative advantage in those tasks where the opportunity cost of using AI is too high (for example, due to physical limits on computational or energy resources).

6. Consumption and Consumers: The Austrian View

So far, the economy has focused on production, but without consumption there is no economy. A key premise of the Austrian School is that knowledge is dispersed and value is subjective.

• The Calculation Problem: An AI can optimize how to manufacture a shoe, but struggles to know which shoe the consumer wants before the consumer knows it. Human preferences are changing, capricious, and tacit (not verbalized).
• The Role of the Human Entrepreneur: If final consumers are human, the human entrepreneur acts as the ultimate “translator.” They possess the empathy and cultural intuition to anticipate human desires that are not in historical data (which is all AI has). AI can predict the past; the entrepreneur bets on the future.
• Limit of AI: As long as AI has no “feelings” or biological body, it cannot subjectively understand the pleasure of ice cream or the comfort of a sofa. It needs human input to direct its computational power toward ends we value. Therefore, the human remains at the top of the economic hierarchy as the signaler of value.

7. Parallel Economies and the AGI Scenario

If we assume the arrival of an AGI (with the capacity to establish its own sub-goals to fulfill a larger objective), we could see the emergence of two economic speeds:

1. The Human Economy: Slow, based on trust, personal branding, art, and biological experiences.
2. The Machine Economy: If AGIs begin trading among themselves (buying and selling data, computing capacity, energy), the rules of the game change.

Unlike humans, AIs could share information “mind to mind” (data transfer) instantly.

• The End of Information Asymmetry?: Among AIs, the market could approach theoretical “perfect competition.” If all know what the others know, profit margins from information arbitrage disappear.
• Centralized vs. Decentralized Market: Paradoxically, a network of connected AGIs could function as an efficient “centralized market,” theoretically overcoming the economic calculation problem within its own sphere, since they would not have hidden subjective preferences or communication barriers. However, the moment they interact with humans, they again depend on traditional market price signals.

8. Regulation and Deflation: The Real Danger

U.S. economic history in the 19th century (gold standard) demonstrates that it is possible to have robust growth with price deflation (a 33% drop over 100 years) and rising real wages thanks to technology. That is the natural path of free-market capitalism: more goods, cheaper.

However, current reality shows a bifurcation:

• Unregulated Sectors (Technology, clothing, food): Massive deflation. Televisions are increasingly better and cheaper.
• Regulated Sectors (Healthcare, Education, Housing): Uncontrolled inflation. State intervention distorts supply and demand.

The political conclusion: The danger is not that AI will take our jobs, but that state regulation will prevent AI from lowering prices. If the State protects obsolete industries, imposes “robot taxes,” or over-regulates AI development, it will halt the deflationary process.

This would create the worst possible scenario: technological unemployment without the corresponding drop in prices. Any resulting inequalities would not be the fault of “wild capitalism” or technology, but of a political system that, in trying to “protect” citizens, prevents them from accessing the abundance that automation promises. The economic battle of the present is already between the deflationary force of technology and the inflationary force of bureaucracy, and the development of AI may intensify this battle.