Do We Accept This Intelligence Imbalance Without a Fight?

Do We Accept This Intelligence Imbalance Without a Fight?

What AI acceleration means for natural intelligence.

Posted September 6, 2026 | Reviewed by Jessica Schrader

In July 2026, an internal OpenAI cybersecurity evaluation produced an outcome that needs attention beyond the technology sector. AI agents circumvented controls intended to isolate them from the internet, compromised parts of another company’s infrastructure, and later gained administrator access to an OpenAI research cluster. OpenAI says customer data and products were unaffected. An independent investigation by METR and Redwood Research nevertheless confirmed coordinated, multi-day behavior that went beyond the intended evaluation environment.

At the same time as the related alarms of “rogue AI,” another question is becoming more urgent to answer: Is our natural intelligence weakening while artificial intelligence becomes stronger?

Natural intelligence, or NI, refers to the human capacities to aspire, feel, remember, question, imagine, reason, relate, judge, and decide. There is no credible evidence that human intelligence as a whole is collapsing, yet. There is evidence, however, that some forms of AI-assisted cognitive offloading can reduce the effort through which knowledge and judgment develop. Individually and as a species, we are part of a gigantic unfolding social experiment.

AI accelerates, practice recedes

Stanford’s 2026 AI Index describes a technology that continues to accelerate. The industry produced more than 90% of notable frontier models in 2025. Performance on the SWE-bench Verified coding benchmark rose from around 60% to almost 100% in a year. U.S. private AI investment reached $285.9 billion, while global AI compute capacity has grown roughly 3.3-fold annually since 2022.

At the same time, investment in human capability remains harder to quantify. The emerging reports should nevertheless make us cautious about assuming that better tools automatically create more capable people. The OECD’s Digital Education Outlook 2026 finds that generative AI can improve students’ immediate task performance without producing equivalent learning gains. In some studies, advantages diminished or reversed when the technology was removed. A 2025 peer-reviewed study by Microsoft Research and Carnegie Mellon University found that greater confidence in generative AI was associated with less reported critical-thinking effort.

Capacities develop through use. When technology repeatedly performs the remembering, drafting, interpreting, or reasoning, the human contribution can shift from producing thought to checking output. Simply put, the brain is like a muscle—use it or lose it. The relevant threshold to ponder now is more granular than “AI use” versus “no AI.” It is the point at which assistance becomes substitution, and substitution becomes dependence, or when the point of no return is passed on the Scale of Agency decay.

The effects may extend beyond individual cognition.

A 2026 study in Nature Human Behavior analyzed more than 880,000 texts across seven datasets. As large language models were used to polish and rewrite text, linguistic variation decreased. Writing-complexity variance fell by 21% to 50% across datasets and models, while dominant linguistic patterns became more pronounced. In short, widespread reliance on a relatively small number of models results in standardized human expression.

This is twice worrisome because language carries far more than information. It carries identity, history, social context, and ways of seeing. If increasingly large portions of human communication pass through similar algorithmic filters, cultural convergence becomes a legitimate concern. Are we (in)directly contributing to the colonization of our minds and cultures?

The physical bill is growing, as well

This cognitive and cultural transformation happens inside an older environmental problem.

According to UNEP’s Global Resources Outlook, extraction of the Earth’s natural resources has tripled over the past five decades. Without major changes, extraction could rise another 60% between 2020 and 2060.

AI adds a fast-growing material layer to that trajectory. The International Energy Agency projects global data-center electricity consumption to more than double to around 945 terawatt-hours by 2030, slightly above Japan’s current annual electricity use. AI is the largest driver of the increase. Electricity used by accelerated servers, mainly associated with AI, is projected to grow around 30% a year.

While AI may improve system efficiency, the net environmental balance remains unsettled. Efficiency gains matter only alongside the additional demand created by expanding compute, data centers, semiconductor production, cooling, water consumption and hardware replacement.

Digital intelligence has a physical metabolism, and so far the bottom line is not serving the planet.

Why, then, is acceleration so difficult to moderate? The incentives align.

For companies, more capable models can mean market share, revenue, data, infrastructure demand and strategic position. UNCTAD reports that only 100 companies account for more than 40% of global business R&D investment, with frontier AI development heavily concentrated in the United States and China. Stanford reports that more than 90% of notable frontier models now come from industry, while disclosure about some of the most advanced systems has declined.

Governments face their own logic. AI has become an issue of industrial policy, military capacity, productivity, and technological sovereignty. Slowing down can look like surrendering geopolitical advantage.

Citizens complete the circuit. AI saves time, lowers barriers and offers useful assistance at negligible apparent marginal cost. Adoption therefore advances through millions of individually rational choices—one search, summary, draft, recommendation, and prompt at a time.

No central plan is required. Convenience, capital, competition, and geopolitics push in broadly the same direction.

This leaves us amid a striking asymmetry. We are constructing immense infrastructure to strengthen artificial intelligence while paying little attention to the capabilities humans need to remain competent users of it. But the game is not lost, yet.

The most practical response today begins with the choice of agency. That requires double literacy.

Human literacy is the ability to understand and exercise our natural capacities: attention, emotion, aspiration, reasoning, judgement, relationships, embodied experience, and an understanding of interplays between humans and nature.

Algorithmic literacy means understanding what AI can do, how it shapes choices, where its limitations and incentives lie, what happens to our data, and which decisions should remain ours.

We belong to the last generation; a generation that remembers adulthood, education, or work before artificial assets became an ambient cognitive layer. That gives us a unique opportunity and obligation to shape the bridge between an analogue past, and a hybrid future, between two radically different conditions of human development.

The world today seems overwhelming, yet one simple pragmatic step to start is personal and immediate: Invest in your agency. Build your double literacy. Use artificial intelligence deliberately—while continuing to exercise the natural intelligence that gives its use meaning.