The AI Bubble: Technical Reality and the Illusion of Continuity

The Govi Bike was a strange type of family bicycle designed for four people. Today it belongs to an nearly extinct category of multi-rider bicycles. Its design reflected social assumptions about family roles, but history moved in another direction. Many inventions appear during transitional periods of civilization. Some of them are imaginative but impractical. They represent dreams more than real social needs.

This example raises a deeper question: why do we repeatedly fail to predict the fate of new technologies?

1. The Myth of the Turning Point
Every new technology is announced as a historical turning point. We are told that the world will never be the same. This narrative spreads through media, research, and investment culture. However, the history of technology does not support the idea that every innovation becomes a permanent transformation.
We often imagine progress as a straight line: one invention succeeds, then improves continuously. But this view is misleading. The American economist Brian Arthur argues that technologies survive not because they are inherently better, but because they find a supportive environment at a specific moment. Success depends on markets, institutions, habits, and sometimes pure chance [1].
Artificial intelligence is not a single entity. It is a broad label that includes statistical models, prediction systems, neural networks, and optimization tools. Much of what is marketed as “intelligence” is advanced automation. Flexibility alone does not guarantee permanence. Similar flexible systems in the past eventually disappeared.
Technology advances not because it is logical, but because society can absorb it.

2. Cycles of Excitement and Correction
The Venezuelan scholar of innovation economics Carlota Perez describes technological revolutions as cycles with four phases: emergence, frenzy, correction, and stabilization [2]. During the frenzy phase, investment becomes detached from real use. Funding flows toward promises rather than proven value.
AI today shows many signs of such a phase: inflated valuations, marketing of future capabilities, and pressure on researchers to deliver rapid results. While excitement is partly necessary for innovation, history shows that many technologies collapse during market corrections.
We are not at a final destination. We are at a moment of testing.

3. Intelligence Without Understanding
A critical question is often avoided: what does it mean for AI to “understand”?
Modern systems rely on large-scale statistical prediction. This approach is powerful but structurally limited. Research on scaling laws suggests that cognitive gains decrease as models grow larger, while energy and environmental costs rise sharply [3].
Computer scientist Yann LeCun notes that these systems do not possess a real model of the world; they simulate linguistic patterns about it [4]. What appears to be understanding is largely statistical efficiency.

4. When Technology Fails Socially
Technologies rarely fail only for technical reasons. Social acceptance is decisive. The philosopher of technology Langdon Winner argues that every technology carries an implicit vision of social organization [5]. If that vision conflicts with existing social structures, the technology may disappear regardless of its performance.
This explains the decline of expert systems despite their theoretical precision [6], the failure of devices like the Segway to become mainstream transportation, and the slow adoption of neural interfaces. A technology that cannot integrate into everyday habits has little historical future.

5. The Question of the Bubble
Calling AI a simple economic bubble would be inaccurate. There is clear overvaluation in some startups, yet there is also genuine demand from major industries. The situation resembles previous mixed cases such as the internet bubble of 1999 or the railway boom of the nineteenth century: many companies failed, but the infrastructure remained.

Several limits may shape the future of AI investment:
Economic return limits: companies invest partly out of fear of falling behind. If operational costs exceed profits, enthusiasm may slow.
Infrastructure and energy limits: large data centers require enormous electricity and hardware resources.
Technical saturation: if progress becomes incremental rather than revolutionary, speculative investment may decline.
Regulatory limits: stricter laws can increase compliance costs and reduce profit margins.
Possible scenarios range from sustained growth and gradual correction to rapid contraction triggered by global economic shocks.

Conclusion: What Will Actually Remain?
History preserves technologies that quietly merge into daily life. AI may not survive in its current visible form. It may become an invisible infrastructure or a specialized tool.
The only certainty is this: what remains will be what fits human realities — not our fantasies.