The case for global benefit from AI

In the early 1600s, the Chamberlens, who were a family of surgeons living in London, invented a device that dramatically improved survival rates for mothers and children during difficult births: the obstetric forceps. Previously, the best known ways to salvage obstructed births were extraordinarily dangerous, involving crude caesarean sections or attempts to break the mother’s pubic bone. However, the family also made a fateful choice. Knowing their design had commercial value, they went to extraordinary lengths to keep it a secret: hiding the forceps in a gilded case, banishing witnesses, and even blindfolding mothers during childbirth.

In 1723 the same device was independently invented by Jan Palfijn, a Flemish doctor, who gave public demonstrations of the technology and published details of its design and use. Forceps soon became a part of standard medical practice, first in Europe and then more widely. Finally, in 1813, a visitor to the old Chamberlen home in Essex noticed a concealed trapdoor, and beyond it, three pairs of the original forceps. At this point, the moral cost of the original decision by the Chamberlens became clear: humanity could have benefitted from this life-saving technology a full century earlier.

In 1959, the distribution of life-saving technology took a different path. A Swedish engineer named Nils Bohlin, who worked at Volvo, invented the modern three-point seat belt. Bohlin had observed that earlier, simpler versions had led to a pattern of serious injury with people falling forward and hitting their heads upon impact. After a short period of discussion, the company decided to release the patent openly so other car manufacturers were free to use it. Current estimates suggest that the modern seat belt design saved more than 300,000 lives in the United States alone.

As these examples make clear, the choices made by those who develop and distribute new technologies can have a decisive impact upon the lives of millions of people. Effects of a similar, if not greater, scale are likely to hold for potentially transformative technologies such as AGI. Like the steam engine, electricity, or the microchip before it, AI is fast becoming a general-purpose technology. As a result, its economic impact is likely to be profound, and it can be expected to spur complementary waves of scientific innovation, shaping opportunities on a global scale.

The global view

This leads to an important question: who should benefit from major technological advances and in what way?

In this essay, we’ll explore a set of arguments in support of what we term the “global view”:

Of course, advanced AI also poses significant risks and potential for harm. We’re not arguing that the benefits of diffusing advanced AI will always outweigh those harms or dangers. Indeed we think it’s critically important to mitigate the risks and downsides. Here, though, we focus on the other half of the equation: whether and how to ensure the benefits of advanced AI are shared broadly.

Two features of the global view deserve emphasis. First, it is fully universal: it holds that everyone has a moral claim to benefit from advanced AI, even if they didn’t directly contribute to developing it. Second, the view is substantial: it holds that people should actually be able to benefit from the technology – if this is what they want. It therefore reaches beyond the notion that people only need to have the opportunity to benefit from AI. In this sense, its central concern is with real benefit for all.

To motivate support for this view we explore four supporting arguments, each grounded in a different value: rights, reciprocity, fairness, and beneficence. With this groundwork in place, we then ask what it would take to make globally beneficial AI a reality.

Rights to technology

The modern human rights framework, formalized in the 1948 Universal Declaration of Human Rights, holds that all individuals are morally entitled to basic standards of life, health, and physical security. Article 27 also explicitly states that everyone has the right “to share in scientific advancement and its benefits.”

What does this right to science and technology encompass? Suppose an entrepreneur invents a new kind of 3D printer. Surely they don’t then become responsible for making sure everyone gets to access the device, or the objects it creates? If there were rights to any given technology, they would be impractical and ultimately unhelpful.

Instead, we need to ask which technologies are essential for unlocking core human capabilities. For example, as long as there is a universal right to a basic education, it follows that there is also a right to the tools and materials that are required for it to be fulfilled. Otherwise, the right itself would lack substance.

Moreover, technologies that were once held to be cutting-edge can quickly become important for ensuring such needs are met. In disaster response or epidemiological management, it’s now reasonable to expect access to GPS and real-time communication. Similarly, mobile phones in low-income countries frequently function as foundational infrastructure, being the gateway to banking, medical information, and emergency services, especially where traditional physical infrastructure is absent.

In this sense, advanced AI may well follow the trajectory of cell phones, GPS, and the internet. AI applications are currently being deployed to anticipate extreme weather events, accelerate drug discovery, and coordinate humanitarian logistics, creating widespread benefits. When a technology becomes necessary to secure key goods and services, it becomes more reasonable to treat access to the technology, directly or indirectly, as a right.

Historically, states have been largely responsible for upholding human rights among their populations. However, some governments aren’t well-equipped to provide access to AI-based tools and resources, while technologists and technology companies may be better-placed to support those efforts. In cases of this kind, if there is a right to access AI then the responsibilities to help ensure access may “fan outwards” encompassing a wider set of actors.

Reciprocity and the tree of knowledge

The story of technological progress is sometimes retold as a series of discrete breakthroughs by lone geniuses. But it’s more useful to think of it in terms of a global tree of knowledge – as a cumulative and intergenerational store of ideas produced through common effort.

For instance, the insights that unlocked modern AI methods owe their inheritance to mathematical traditions tracing back centuries – developed by thousands of researchers and scientists (and those who supported them in turn). The modern AI revolution has turned binary logic and algorithms, proposed by mathematicians like Leibniz and Boole, into infrastructure which helps run the modern world.

What kind of responsibility, if any, do technologists incur by drawing upon this knowledge base? Somewhat controversially, the philosopher Robert Nozick argued that those who passively benefit from a public good do not have a responsibility to give back to those who produced it. For example, it may be kind to pay a street busker, but if you can’t help hearing their music while you walk down the street, it’s fine not to contribute (even if you enjoyed the performance).

Some might think technologists who draw upon preexisting knowledge as an ingredient for invention are in the same situation. We think this is mistaken. Crucially, when AI researchers draw upon this resource, they actively benefit from, and create value by participating in, a practice that took many centuries to develop. Under these conditions, reciprocity suggests they should support the practice of knowledge creation, including the norms that enable it to function.

At a minimum this includes reinvesting in science. Yet, when the breakthrough is of great value it may involve something more: supporting the ecosystem that makes science possible, including the education of wider populations and institutions of learning.

Beyond this historical debt to science, modern AI systems also depend on a vast system of coordinated action. Their training requires planetary-scale infrastructure, an educated workforce, and an expansive global supply chain. Yet these networks do not necessarily result in a return to those who sustain them. Consider the school teachers and education systems around the world that were needed to train leading AI researchers. If these contributions to building AI are global and diffuse, then reciprocity also holds that the benefits of the technology should likewise be global and diffuse.

Fairness and the lottery of birth

What factors determine how well a person’s life goes overall? Of course, talent and effort make a difference. It also matters that they’re well-fed and free from disease, receive a decent education, and have opportunities for well-paid work. Taken together, one simple fact influences a person’s life more than any other: where in the world they are born. But nobody chooses this: our lives begin with the spin of a roulette wheel.

There’s a school of thought in political philosophy that grapples with this unfairness, called “luck egalitarianism”. On this view, disadvantages that come about through dumb luck, like the circumstances of our birth, are unfair. As such, they’re something we should aim to address – especially when the stakes in the lottery are high.

Modern history is a story of great technological advances, unevenly distributed. Before the Industrial Revolution, almost everyone on Earth lived in a condition of absolute poverty. In the following two centuries, sustained economic growth lifted billions of people into conditions of relative abundance. But for many people today, living standards are no higher than they were for the rest of humanity in 1800: the so-called “Great Divergence” in national fortunes.

Some commentators argue that advanced AI might drive global and domestic inequality to even higher levels (at least without a policy response to address it). Others argue that high AI adoption rates in countries such as India mean that AI might contribute to a rising tide that lifts all boats. The rising cost of training and running frontier AI systems could bring to a close the era in which they are reasonably affordable — or the rising prominence of lower-cost open models could expand access to these tools more widely.

Given such uncertainty, how can we address potential divergence?

While international aid and cash transfers can partially level the playing field, money alone cannot resolve the structural challenges posed by rapid technological change. This is because access to technology is itself an important driver of growth, bestowing capacities upon those who use it – in the realm of science, technology, and invention – that may otherwise remain out of bounds. Against this backdrop, initiatives designed to distribute AI and its complementary infrastructure widely, and to use it to address social challenges, can help ensure the benefits of technology are broadly shared.

The challenge of squandered human potential

The final ethical perspective we consider focuses directly on the enormous difference that technology can make to human life and flourishing overall. Advanced AI has the potential to make the world radically better: for example, ending diseases by accelerating medical advances, and elevating global living standards. Nonetheless, those benefits could also end up not being widely distributed.

Putting aside fairness, justice, or political instability, this would simply amount to a huge waste of potential human flourishing. Indeed, it would be a loss even if everybody’s basic rights (to food, shelter, primary education, and so on) were met. The world would still be so much worse than it otherwise might have been.

Most moral theories agree that if an actor, or community of actors, are faced with an opportunity to bring about enormous benefits (like saving many lives) at a manageable cost to themselves, they ought to do so. AI technologists increasingly find themselves in such a position, as their decisions shape the trajectory of a technology that increasingly touches the lives of people around the world, helping to determine the terms on which it can be accessed and how it’s used.

It might be objected that the moral requirement for AI to generate global benefit demands too much of technologists and wealthy nations. After all, AI companies are already spending hundreds of billions of dollars to develop advanced AI, sometimes run on tight margins, and additional obligations could reduce incentives for further innovation. But the cost of any action must be evaluated relative to the scale of the benefits it can unlock. If AI is, as many hope, a tool that can be used to make progress on some of humanity’s most intractable challenges, can it really be right to leave so much untapped benefit on the table?

Fortunately, the choice is not always so stark. To begin with, technological invention often diffuses rapidly through national economies, with the bulk of the benefit accruing to other actors who were not themselves the original innovator – such as other firms, both small and large, which build on top of innovations or use them in other sectors. In the present case, broad-based market access to goods and services and the open interaction of ideas clearly have an important role to play. At the same time, additional measures may be needed to ensure the benefits of AI extend to low-income countries or harder-to-reach populations.

Institutions and incentives, like the global fund for AI recently proposed by the UN Secretary-General, may help technologists receive a return that encourages further investment in innovation over time while also facilitating broad benefit.

Unlocking Global Benefit

The four arguments developed here—based upon rights to technology, the tree of knowledge, the lottery of birth, and squandered potential—are grounded in distinct philosophical and political traditions. Yet they provide support for a common conclusion: there is a global claim to benefit from AI’s invention and use.

How might such broad-based benefit and access be unlocked in practice?

We don’t pretend to know the precise mix of policies, institutions, and incentives that will best achieve this outcome. However, there are several promising pathways to be explored.

First, AI companies should orient research toward the global good. Competing research pathways might be similarly promising from a commercial perspective, but some generate more public benefits. For example, AI applications for medical breakthroughs are likely to generate more material benefit than those focused on entertainment.

AI companies should also aim to ensure continued, widespread, and affordable access to their models – subject to appropriate safety and security limits and protections, which are justified when broad access increases risk of misuse. This could involve ensuring that lightweight models – which are reasonably close to the frontier – are available in as many geographies as possible. It could also mean subsidizing use-cases with special public value, such as those in schools and hospitals. Or in some cases it could mean sharing model weights. For example, Google DeepMind released AlphaFold 1, 2, and 3 for the public to use (the latter in partnership with Isomorphic Labs).

Of course, model access alone won’t always be enough to confer tangible benefits. In 2005, the MIT Media Lab launched the “One Laptop Per Child” initiative, with the goal of building and distributing affordable laptops to millions of schools in developing countries. In 2012, a rigorous empirical evaluation of the program in Peru – one of the biggest adopters – found no improvement in math or reading scores. Computers alone could accomplish little without teacher training and internet access, or when basic necessities like textbooks and running water were still missing.

To avoid repeating these mistakes, technologists could look beyond their traditional consumer bases and support developers around the world who are building applications to meet specific local needs. More substantially, governments, developers, international organizations, and nonprofits could invest in the infrastructure that will make AI genuinely useful: internet access, reliable power, technical support, and financing for larger projects.

A new age for technologists?

The power of technology to shape global outcomes is greater now than at any previous point in history. Just as earlier moral frameworks, intended to govern conduct in small communities, may no longer serve as an accurate guide to action in a globally connected world, so the power of technology may call for a new kind of ethics – one that is commensurate with its deep impact on our lives.

Technologies such as AI often bring with them unique capabilities, and hence the opportunity and responsibility to do things that could otherwise not be done at all. Even when this is not the case, technologists increasingly have an opportunity to unlock benefits in situations where others have chosen not to act or where alternative approaches appear to have stalled. The development of technologies like AI itself brings with it responsibilities for technologists, understood as co-participants in a human practice from which we all benefit and which is geared towards common ends.

If advanced AI meets its promise as a transformative general-purpose technology, then it has the potential to bring about a radically better world. But technology is not destiny. The task before all of us is to imagine, share, and iterate on the best plans for realizing the benefits for the world. The task for those who help shape AI is to choose, deliberately, to prioritise these goals, helping to bring about structural, inclusive, and enduring improvement of the human condition.