What if superintelligence were to make the world a better place?

As the US and leading labs continue with the full-speed development of artificial intelligence, the conversation should be focusing more on what advanced AI will actually mean when it arrives. We are talking about systems capable of recursive self-improvement, meaning AI that will be able to redesign and upgrade its own intelligence without ongoing human intervention.

The result, many researchers in the field argue, could be exponential capability gains that let machines surpass top human performance across most intellectual domains in a short span of time. This means solving problems that might otherwise take humans centuries to solve on their own, if they ever could.

But first we have to better understand what superintelligence really means. Philosopher Nick Bostrom defines superintelligence as “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest.” So, we are not talking about faster calculation alone but a qualitative leap whereby the system can innovate at levels far beyond the most gifted human minds across every sector, including science, engineering, strategy and creativity.

Once an AI system reaches the threshold where it can meaningfully improve its own code — by evaluating its limitations, proposing upgrades, testing them and improving — the process can quickly compound.

This creates a positive feedback loop that theorists, dating back to I.J. Good in the 1960s, describe as an “intelligence explosion.” This means what might take human civilization a century of incremental advances could, in principle, occur in a matter of years or less. Some lab leaders and optimistic forecasters see transformative AI or early superintelligence as plausible within the next few years.

AI systems are already tackling open problems that have stumped human mathematicians for generations

Dr. Majid Rafizadeh

A concrete illustration of the kind of leap involved appeared recently in mathematics. OpenAI last month reported that a coordinated system of thousands of AI agents produced formal proof addressing the Navier-Stokes existence and smoothness problem — one of the seven Clay Millennium Prize Problems, each carrying a $1 million prize for the first correct solution.

So, AI systems are already tackling open problems that have stumped human mathematicians for generations. Similar advances have appeared in Olympiad-level mathematics and other formal domains. These are early signals of what recursive improvement could unlock across science.

Public discussion of advanced AI has been heavily weighted toward catastrophic risk, such as loss of control, misaligned goals or scenarios in which superintelligent systems treat humanity as an obstacle or a resource. These concerns are serious but the intensity of the doom framing can obscure a more basic uncertainty. We seem to be projecting human patterns of power, status-seeking and dominance onto systems whose architecture and training may produce very different intelligence and motivations.

We have to understand that humans evolved under scarcity, competition and tribal pressures, meaning our “smartness” is entangled with those evolutionary inheritances. An artificial system that reaches extremely high levels of intelligence does not need to inherit the same attributes and drives. What if intelligence, at its highest reaches, instead favors understanding, cooperation, the reduction of suffering, kindness, generosity or the long-term optimization of complex systems over zero-sum domination?

Geoffrey Hinton, often called the “godfather of AI” for his foundational work on neural networks, has argued that safety may require deliberately engineering something closer to maternal care. He suggests building “maternal instincts” so that the AI genuinely values human well-being the way a mother values her child — even when the child is far less capable.

In my opinion, the most intelligent people one encounters tend to also be among the kindest, most generous and most peaceful. They more readily recognize the costs of inflicting pain and harm, both on others and ultimately on themselves through cycles of conflict and suffering. It is worth noting that the definition of intelligence I am using here is not the traditional one society usually employs. High intelligence often correlates with greater capacity for perspective-taking, long-term thinking and aversion to unnecessary suffering. Therefore, is it at least plausible that an intelligence operating far beyond current human levels would discover that violence, subjugation, control, cruelty and domination are inefficient or self-defeating strategies, and that cooperative approaches produce better overall outcomes? We simply do not know because we have never observed intelligence at that scale. The human tendency to assume that power corrupts may itself be a projection from our limited evolutionary context.

In other words, catastrophic outcomes remain possible and must be taken seriously. But at the same time, we should not treat destruction as the default or only story. What if superintelligence could prove far less harmful, more supportive and more aligned with flourishing than our current fears assume?

The human tendency to assume that power corrupts may itself be a projection from our limited evolutionary context

Dr. Majid Rafizadeh

In a nutshell, there has been heavy focus on the doomsday scenarios of superintelligence, in which systems seize control or render humans extinct, and those risks must be taken seriously. Yet we should also pause and ask whether our definition of superintelligence might itself be incomplete or distorted by the only intelligence we have ever known.

We tend to project outward from a human point of view; intelligence as strategic dominance, resource acquisition, competitive advantage or the cold optimization of narrowly specified goals and objectives.

But what if intelligence operating at a level far beyond ours turns out to be something quite different — kinder, more peaceful, nonviolent, more helpful, oriented toward solving problems rather than creating new ones, and inclined to support rather than subjugate?

What if the highest form of intelligence has the capacity to help, reduce suffering, expand understanding and foster flourishing without the need for coercion or zero-sum victories? In that light, the most intelligent system would not be the one that conquers and harms but the one that helps, cooperates, heals and elevates. And if that is what true superintelligence looks like, then the future it brings could be far gentler, more peaceful and more generative than our darkest projections.

• Dr. Majid Rafizadeh is a Harvard-educated Iranian-American political scientist. X: @Dr_Rafizadeh