The Misleading Narratives of AI and Quantum Computing

When an unsolved scientific problem is sold as an engineering project

Every technological revolution attracts visionaries, entrepreneurs, investors and researchers. It also attracts an entire ecosystem of people whose fortunes depend on a particular story being believed. Over time, that story can become so dominant that it is treated not as a hypothesis to be tested but as an established fact. Skeptics are dismissed as pessimists; inconvenient questions are recast as distractions; and the public is encouraged to confuse investment, publicity and technological activity with scientific resolution. Two of the most consequential examples today are artificial intelligence and quantum computing.

Both fields have produced extraordinary achievements. Both have attracted vast sums of money and some of the brightest minds in the world. Yet both are increasingly described through a remarkably convenient narrative: the fundamental science is finished, the invention has been made, and what remains is merely a question of scaling it. In AI, we supposedly need more data, more computing power and more electricity. In quantum computing, we supposedly need better fabrication, more qubits and more sophisticated engineering. But what if the central intellectual breakthroughs have not yet happened?

Consider the prevailing conversation about artificial intelligence. We are told that intelligence is primarily a compute problem. Build larger clusters, train larger models, acquire more data and artificial general intelligence will eventually emerge. The discussion has expanded from algorithms to power grids, nuclear reactors and national energy policy, as though the principal obstacle to creating intelligence were the availability of electricity. It is an appealing proposition because it converts a profound scientific mystery into a capital expenditure plan. If the answer is simply scale, then the winners are those who can raise and spend the most money.

But the original ambition of artificial intelligence was not to manufacture increasingly fluent text generators. It was to invent synthetic intelligence: a system capable of understanding, reasoning, learning, adapting and forming useful abstractions in ways that generalize beyond familiar examples. Contemporary neural networks, including transformer-based large language models, are undeniably useful. They can write software, summarize complex information, translate languages and assist with scientific work. None of that, however, settles the deeper question of whether we have discovered the principles required to construct general intelligence.

A machine can display remarkable competence without establishing that its underlying mechanism possesses the broad, autonomous intelligence originally sought. François Chollet, in his paper “On the Measure of Intelligence,” made an important distinction between performance on learned tasks and the efficiency with which a system acquires new skills in unfamiliar situations. This distinction matters enormously. A system trained on immense quantities of human-generated knowledge may reproduce impressive capabilities, while the question of how to create robust generalization remains open. More training can improve performance; it does not, by itself, prove that the problem of intelligence has been solved.

The issue is not whether large language models are valuable. They are. The issue is whether their commercial success should be mistaken for the completion of a century-long scientific quest. A better language model is not necessarily a theory of intelligence. A larger data center is not necessarily a substitute for a missing conceptual breakthrough. If we prematurely declare AI solved, we risk directing intellectual and financial resources toward refining one successful paradigm while neglecting the possibility that genuine synthetic intelligence requires inventions we have not yet imagined.

Quantum computing has developed a strikingly similar narrative. We frequently hear that the physics has been established, the mathematical foundations are understood, and the only remaining task is engineering. Improve qubit fidelity, suppress noise, build error correction, manufacture larger devices, and useful fault-tolerant quantum computers will follow. Once again, an unresolved scientific challenge is translated into an apparently straightforward industrial roadmap.

Yet the existence of elegant quantum algorithms and functioning experimental processors does not establish that large-scale, reliable, economically useful quantum computation is already within reach. The difficult question is not merely whether we can manipulate a few or even many quantum systems. It is whether we can preserve and control the relevant quantum information through sufficiently deep computations, at acceptable error rates and overheads, in a real physical machine. That is a much more demanding claim than demonstrating a laboratory milestone.

Mathematician Gil Kalai has long challenged the assumption that quantum error correction will necessarily overcome the noise of physical devices. His work examines the possibility that correlated errors and noise sensitivity may impose deeper limitations on scalable fault tolerance. His conjectures are contested and remain unproven; nevertheless, they illustrate precisely why the physics cannot simply be declared finished. If the noise behavior of real quantum systems defeats the assumptions on which scalable computation depends, additional engineering alone may not deliver the promised machine. We may need new physical insights, new computational architectures or inventions not yet available.

This does not erase the substantial progress in quantum control, error correction and experimental demonstrations. Nor does it prove that useful quantum computers are impossible. It means that the distinction between promising evidence and a completed invention must be preserved. A prototype, a benchmark result or a roadmap is not the same thing as a dependable production-grade computer solving valuable problems at scale. The unresolved question is not an embarrassment to be hidden in a footnote; it is the central challenge of the field.

There is a revealing economic symmetry between these two stories. If AI is merely a scaling problem, the obvious beneficiaries are companies selling compute, chips, cloud infrastructure and energy. If quantum computing is merely an engineering problem, the obvious beneficiaries are organizations selling hardware roadmaps, research programs, specialized infrastructure and the promise of future capacity. This does not make their work illegitimate, nor does it mean every optimistic claim is dishonest. It does mean that powerful commercial incentives favor narratives of inevitability over candid acknowledgment of scientific uncertainty.

The danger is not optimism. Science requires optimism, ambition and enormous persistence. The danger is allowing optimism to harden into orthodoxy. When a community begins to treat its preferred technological pathway as the only plausible future, alternative approaches receive less attention and skepticism becomes socially expensive. The resulting consensus can look stronger than the evidence actually warrants. Money and institutional prestige can amplify a narrative; neither can turn an unproven premise into a law of nature.

History repeatedly reminds us that invention cannot be scheduled simply by allocating more resources. Engineering can improve something whose governing principles are sufficiently understood. It cannot guarantee the discovery of principles that remain unknown. Building a thousand larger versions of an incomplete idea does not necessarily produce the missing idea. Sometimes progress comes from scaling; sometimes it comes from abandoning an assumption that everyone else has stopped questioning.

We should therefore resist the temptation to describe either artificial intelligence or quantum computing as a solved problem. AI has delivered astonishingly capable tools, but the quest to invent general synthetic intelligence remains intellectually open. Quantum computing has delivered remarkable scientific experiments and increasingly sophisticated devices, but the route to universally reliable, production-grade quantum computation remains uncertain. In both fields, the greatest opportunity may lie not in accepting the dominant roadmap as inevitable, but in recognizing how much still needs to be invented.

The most consequential breakthroughs of the coming decades may not come from those who build the largest infrastructure around today's assumptions. They may come from those willing to ask whether the assumptions themselves are wrong. That is the difference between scaling a technology and inventing the future.

Selected reading: François Chollet, “On the Measure of Intelligence” (2019), https://arxiv.org/abs/1911.01547; Gil Kalai, “Seven Assertions about Quantum Computing” (2025), https://gilkalai.wordpress.com/2025/02/05/seven-assertions-about-quantum-computing/; Gil Kalai, “The Fully Depolarizing Noise Conjecture for Entangled Physical States” (2026), https://arxiv.org/abs/2609.33508.