AI Needs Steering, Not Just Brakes
Imagine humanity discovers a new form of intelligence.
It can absorb more written knowledge than any individual could read in a lifetime. It can write software, interpret images, analyze diseases, design machines, tutor students, operate digital tools and increasingly act on our behalf.
What would we do with it?
Would we ask how it could make a doctor more capable, an engineer more inventive, a teacher more effective and an entrepreneur more productive? Would we redesign education around a world in which intelligence itself has become abundant? Would we build institutions capable of governing systems that improve faster than regulation traditionally moves?
Or would our first question be how many people it can replace? Or fear the change over the realization of evolution?
That distinction matters because the future of artificial intelligence will not be determined by capability alone. It will be determined by the objectives, incentives and institutions we build around that capability.
AI does not need a world with no brakes. The risks are real, and pretending otherwise is not optimism. But brakes alone cannot tell us where to go.
AI needs steering.
Fear Is Becoming Part of the AI Story
Public anxiety about artificial intelligence is no longer a fringe reaction to an unfamiliar technology. It is becoming part of the technology's operating environment.
In June 2026, 52% of U.S. adults surveyed by Pew Research Center said they were more concerned than excited about the growing use of AI in daily life, compared with 37% in 2021. Seventy-one percent expected AI to lead to fewer jobs in the United States over the next two decades. A separate Pew study across 36 countries found widespread concern about AI's impact on employment. [1][2]
Those concerns should not be dismissed as resistance to progress. People are asking legitimate questions about employment, human agency, misinformation, surveillance, inequality, concentration of power, creative autonomy and what happens when increasingly capable systems are permitted to act.
The wrong response is to tell society simply to trust the technology.
The equally wrong response is to make fear the organizing principle of technological development.
Concern is information. It tells us where institutions, companies and technologists have not yet created sufficient confidence, participation or accountability. The task is to convert that concern into better direction.
The Future of Work Should Be Augmentation by Design
Auxos Global has consistently viewed AI's most valuable role in the workplace as augmentation: expanding what people can accomplish rather than treating human labor primarily as a cost to be removed.
That does not mean no task, role or occupation will disappear. Technology has always reorganized labor, and AI will be no exception. But the evidence does not support reducing the future of work to a simple replacement narrative.
The International Labour Organization's 2025 global analysis found that one in four workers is in an occupation with some exposure to generative AI. Yet because most occupations contain tasks that still require human input, the ILO concluded that transformation of jobs is the more likely overall effect than redundancy. [3]
We can already see what augmentation can look like. In a field study of 5,179 customer-support agents, access to a generative-AI assistant increased productivity by about 14% on average, with substantially larger gains among novice and lower-skilled workers. The researchers found evidence that the system helped diffuse the practices of more experienced workers. [4]
A later field experiment across 66 firms and 7,137 knowledge workers found that workers using an integrated generative-AI tool spent roughly two fewer hours per week on email during the latter half of the experiment and reduced work outside regular hours. [5]
These studies do not settle the future of employment. They demonstrate something more useful: AI can be designed and deployed as a capability multiplier.
The question for leaders should therefore not be, 'How much headcount can AI remove?'
It should be, 'What becomes possible when every capable person has access to capabilities that previously required an entire team?'
Replacement may improve a quarterly cost line. Augmentation can expand the productive frontier of the organization.
A Kill Switch Is Not a Governance Model
The same distinction applies to AI safety.
Public discussion sometimes reduces control to an emergency metaphor: if an AI system becomes dangerous, someone should be able to switch it off. Emergency controls matter. But a kill switch is not a governance architecture any more than an emergency brake is a transportation system.
Modern AI risk management already reflects this reality. The International AI Safety Report 2026 describes a defense-in-depth approach involving multiple layers: acceptable-use policies, access controls, model evaluations, technical safeguards, deployment monitoring, human oversight and post-deployment measures. It also emphasizes that no combination of safeguards is perfectly reliable and that real-world behavior can differ from pre-deployment evaluation. [6]
Governance therefore has to move.
It must observe new capabilities, learn from deployment, identify emerging failure modes, update standards, adjust access, educate users and revise rules as the technology changes.
This is why AI needs steering, not just brakes.
A brake responds to danger by reducing movement. Steering continuously interprets the environment and changes direction. As systems become more capable and more deeply embedded in society, governance must become more adaptive, not merely more restrictive.
That requires engineers, domain experts, social scientists, educators, policymakers and the public to operate in a tighter feedback loop than traditional regulation has often required.
Governance Must Learn at the Speed of Deployment
This is also why the emerging international conversation matters.
At the United Nations General Assembly on 26 September, Singapore's Minister for Foreign Affairs, Dr. Vivian Balakrishnan, described AI as an urgent frontier whose capabilities are advancing faster than our appreciation of its risks. He identified risks ranging from loss of control over autonomous systems and malicious use to emerging economic, social and political disruption, and argued that the international community must be prepared to formulate new rules and, where necessary, new institutions. He also raised the possibility of considering a UN Framework Convention on AI Safeguards. [7]
That argument should not be interpreted as a case against technological progress. It is an argument that governance architecture has to evolve with the technology it governs.
The broader UN system is moving in the same direction. The Global Dialogue on AI Governance now brings all 193 Member States together with the private sector, academia, civil society and the technical community, while an Independent International Scientific Panel provides an evidence base on AI's opportunities, risks and impacts. [8][9]
This matters because AI governance cannot be built exclusively by policymakers who do not understand the systems, nor by technologists whose incentives and expertise do not encompass every societal consequence.
Policy needs technical literacy. Technology needs institutional literacy.
The partnership between the two is not a compromise. It is the operating model.
Profit Is a Signal. It Cannot Be the Destination.
Commercial incentives have been one of the most powerful engines of technological progress. Capital allows ideas to become products, infrastructure and industries. Profit provides a signal that something creates economic value.
But a signal is not the same thing as a destination.
When a technology begins influencing education, labor, information, scientific discovery, security and human decision-making, optimizing it exclusively around short-term financial return becomes structurally inadequate.
An AI system can be economically efficient while weakening human capability. It can maximize engagement while degrading information quality. It can reduce labor costs while hollowing out the pathways through which junior employees acquire expertise. It can optimize an operational metric while creating environmental or social costs that sit outside the spreadsheet.
The correct response is not to reject markets or investment. It is to widen the objective function.
Long-term value creation must account for the resilience of the organizations, communities and institutions in which the technology operates.
Restraint, in that context, is not anti-growth.
It is the discipline to refuse an optimization that improves today's metric by damaging tomorrow's system.
From Responsible AI to Sustainable Intelligence
This is where we believe the conversation needs a stronger concept.
The phrase 'Sustainable Intelligence' already exists in academic and institutional discussions, so it would be inaccurate to present it as a newly coined term. But it deserves a clearer definition for the age of increasingly capable AI.
At Auxos, we propose using Sustainable Intelligence to describe a design principle:
Sustainable Intelligence is the development and integration of artificial intelligence in ways that expand human capability while preserving human agency, institutional resilience, social legitimacy and the ecological systems on which long-term progress depends.
That definition deliberately goes beyond responsible AI.
Responsibility asks whether a system is safe, fair, accountable and compliant.
Sustainable Intelligence asks whether the relationship between humans, institutions and increasingly capable machines can remain beneficial over decades.
It asks whether workers become more capable rather than merely cheaper. Whether education evolves rather than atrophies. Whether governance learns rather than freezes. Whether productivity gains strengthen society rather than concentrate capability so narrowly that the system becomes brittle. Whether the energy and infrastructure supporting AI are compatible with long-term environmental constraints.
Most importantly, it asks what intelligence is for.
A Technological Civilization Also Needs Civilizational Memory
That question is older than computing.
Modern AI has emerged from mathematics, computer science, engineering, markets and institutions that are extraordinarily young when measured against human civilization. Yet the questions it is forcing us to confront are ancient: What is intelligence? What responsibilities accompany knowledge? What constitutes a good life? What should power serve? When does capability require restraint? What do individuals owe the societies and natural systems around them?
Humanity does not need to pretend that ancient philosophy contains instructions for regulating neural networks. It does not.
But civilizations with long intellectual traditions—including India and China, alongside traditions from across the globe—have spent centuries examining duty, harmony, restraint, consciousness, human flourishing, collective responsibility and the relationship between humanity and nature. Those traditions should not replace scientific evidence or modern institutions. They can broaden the questions those institutions ask.
A civilization creating intelligence should draw upon humanity's deepest thinking about what intelligence is for.
This may become especially important as AI governance becomes global. The values embedded in future systems should not emerge accidentally from the commercial assumptions of a handful of companies, the political priorities of a handful of governments or the cultural assumptions of a handful of technology hubs.
Technical intelligence may scale globally.
Wisdom must be allowed to contribute globally too.
The Singularity Worth Pursuing
For decades, the technological singularity has often been imagined as a threshold: the moment machine intelligence exceeds human intelligence and technological change accelerates beyond our ability to predict it.
Perhaps there is another version worth pursuing.
Not a singularity defined by machines leaving humanity behind, but one in which human and machine intelligence become sufficiently complementary that humanity itself becomes more capable.
AI performs calculations no person could complete.
Humans determine which problems deserve solving.
AI explores millions of possible designs.
Humans decide which futures are worth building.
AI optimizes.
Humans decide what deserves to be optimized.
This is not an argument for permanent human superiority, nor an assumption that human judgment is infallible. History provides ample evidence to the contrary.
It is an argument that capability without purpose is not progress.
If AI can amplify human intelligence, then our responsibility is to ensure that judgment, education, governance and wisdom are amplified alongside it.
We Are Holding the Steering Wheel
The AI debate does not need another binary choice between acceleration and restriction.
We need the capacity to accelerate where AI expands human flourishing, constrain where risks become unacceptable, redirect when incentives produce harmful outcomes and continually update those decisions as evidence changes.
That is a much harder task than either 'move fast' or 'stop.'
It requires companies willing to measure value beyond immediate labor savings. Investors willing to distinguish sustainable scale from extractive optimization. Governments capable of learning alongside technologists. Technologists willing to accept that social legitimacy is part of system performance. Educational institutions preparing people not merely to compete with AI, but to work intelligently with it.
And it requires societies to decide what they want increasingly abundant intelligence to serve.
The greatest risk may not be that artificial intelligence becomes too intelligent.
It may be that humanity becomes extraordinarily good at building intelligence before becoming equally thoughtful about what that intelligence should serve.
We should build the brakes.
But brakes exist because we intend to move.
Now we need to become much better at steering.
References
[1] Pew Research Center, “Young US adults are increasingly wary of AI, concerned it will take jobs,” 18 August 2026. Source
[2] Pew Research Center, “Concerns about AI are especially widespread in wealthier countries,” 17 September 2026. Source
[3] International Labour Organization, “Generative AI and Jobs: A Refined Global Index of Occupational Exposure,” 20 May 2025. Source
[4] Brynjolfsson, Li & Raymond, “Generative AI at Work,” NBER Working Paper 31161; published in QJE, 2025. Source
[5] Dillon et al., “Shifting Work Patterns with Generative AI,” NBER Working Paper 33795, revised November 2025. Source
[6] International AI Safety Report 2026, Risk Management. Source
[7] Singapore Ministry of Foreign Affairs, National Statement by Dr. Vivian Balakrishnan to the 81st UN General Assembly, 26 September 2026. Source
[8] United Nations, Global Dialogue on AI Governance. Source
[9] United Nations, Independent International Scientific Panel on AI, Preliminary Report. Source
[10] Examples of prior use of “Sustainable Intelligence” include 2026 academic and institutional publications; the term is therefore used here as an Auxos-defined framework rather than claimed as a new coinage. Source
