It is increasingly probable that the next U.S. presidential term could see the development of Artificial General Intelligence (AGI). If that happens, everything will change, and generative AI—the artificial intelligence which can produce images and text through ChatGPT and other applications—will seem like the Kitty Hawk Flyer compared with the B-21. AGI, possessing cognitive abilities that equal if not exceed a human’s, will be capable of performing virtually all tasks. It will revolutionize the economy, turbocharge scientific discovery, propel the quality of life to unimagined heights, and grant near invulnerability to national security. Humanity stands on the verge of a new, potentially golden era. But AGI could also create a much darker world.
The path that AGI takes will depend in large measure on who develops it, and how. If it is done in the United States, and done responsibly, then its benefits could be immense. This new, novel, and immense intelligence can be brought to bear on a swathe of problems and tasks, leading to the situation where we could eventually see AI models with the capabilities of a Nobel Prize winner assisting sectors ranging from manufacturing to national security. If, however, AGI is inaugurated in Beijing then the situation could be very different, and its effects likely malign. Chinese possession of AGI would give its troops the edge on every battlefield, its businesses the advantage in every market, and its security services the capacity to enforce a level of surveillance and repression that exceeds anything yet attempted by an authoritarian state.
Amidst the colossal domestic and international challenges that will confront the next U.S. president—Ukraine, Taiwan, the Middle East, the economy, and a politically charged environment—ensuring that the United States wins the AGI race will be seen, retrospectively, as the most important. The U.S. government’s full resources must be harnessed and then channeled into this effort, in support of those private companies working to achieve it. If Washington does what it needs to do, and it succeeds, then when the history of the 21st century is written, 2025 to 2029 will be seen as one of the most consequential periods in the country’s history.
Developing AGI will necessarily depend upon the private sector, but Washington can assist it through a combination of funding and technical support on a scale similar to that provided during the Cold War. Although ChatGPT and other landmark AI achievements were developed without government assistance, AGI likely cannot. Its requirements may be similar to AI’s, but their scale is far greater. Beijing will be doing everything it can to assist its scientists and companies, and Washington must do the same. The U.S. government must make itself a key partner in the effort to create AGI, for only Washington can meet the new technology’s data, infrastructure, workforce, and energy demands.
The first step should be the creation of an AGI-focused commission, similar to the National Security Commission on Artificial Intelligence that Congress created in 2018 at the beginning of the AI race, to make recommendations to the White House and Congress on how to boost the development of artificial intelligence and other key technologies. This new commission, with a remit spanning academia, government, and business, should be tasked with ensuring that AGI models are created in the United States by 2029. The commission should be created by Congress before the president is inaugurated to establish bipartisan support, and to ensure that legislators understand the importance of its success. The commission should channel funding and expertise to those technologies, where the United States already has a competitive advantage, such as natural language processing and reinforcement learning. Basic research, as well as development of new AGI architectures and training methods, should also receive funding and support.
The initiative should lead Washington’s drive to clear all obstacles from the path of businesses. The largest of these relate to data. Developments in AGI will require ever larger quantities of information, and current models will soon exhaust what is available on the internet. The problem is not a lack of data. Rather, there has never been more of it. In the past three years alone, more data has been created than in all of human history. The challenge is helping companies access it. That is a role for the government, and Washington should begin by creating a national register of companies and organizations holding large data sets. It should then develop economic incentives for sharing this data, in a manner consistent with privacy rights and intellectual property protection. That is not all. To process these colossal quantities of data, companies will need efficient and scalable storage facilities, which the United States lacks. Washington must ensure that this storage is built. At the same time, the administration must fund research into new technologies, including DNA storage and quantum storage, which will be critical to AGI’s development.
AI is already an energy-intensive technology, with a single ChatGPT query, for example, requiring 10 times the energy that goes into generating a Google search. The computational power necessary to sustain AI’s growth doubles every three months or so, meaning that the pressures on U.S. energy infrastructure will exponentially grow as AGI gets closer. Washington has time to prepare for these increases, and should get ahead of them, to develop the infrastructure and technology necessary to cope with surging demand. Investments must be made, particularly in renewable energy technologies, and this can be done through a combination of tax incentives, research grants, and public-private partnerships. Efforts to increase energy supply should be accompanied by research into energy-efficient computing technologies that can reduce the environmental impact of the drive to AGI.
Finally, Washington must ensure continued access to the high-performance computing resources necessary for AGI development, including CPUs, GPUs, and specialized AI accelerators. This will involve investment in research and the development of next-generation computing technologies, as well as ensuring that these resources are accessible to a wide range of researchers, developers, and academics, rather than just large corporations. Additionally, economic incentives should be introduced to increase domestic production of these critical components, reduce reliance on foreign suppliers, and mitigate supply chain risks.
AGI will also require a workforce skilled in AI, computer science, and engineering. To develop domestic expertise, Washington should expand and improve training programs from K-12 through to professional development, crafting new curricula, establishing scholarships, and proactively supporting research initiatives in education. Apprenticeship programs must be created, research partnerships established, and funding provided for joint initiatives that address the workforce needs of the AI industry. At the same time, efforts to attract top global talent must be redoubled. Crucial to this will be reforming immigration policies, to make it easier for foreign expertise to come to the United States, and those scientists from foreign countries being trained in U.S. universities must be given the incentives and means to remain in the country after graduation.
It will, however, be impossible to draw all the best scientists, technologists and developers to the continental United States. Nor will it be possible to develop all the necessary infrastructure in the United States. Collaboration will be necessary to fill the gaps in workforce, capacity, and expertise. Considering the stakes, this must be done carefully, and partners for close collaboration must be chosen deliberately. Those countries that are close U.S. allies with security standards matching Washington’s—Canada, Japan, South Korea, and the United Kingdom, in particular—should be designated Tier A countries, and their capacities and infrastructure capabilities should be mapped against those in the United States, to identify opportunities for collaboration. Canada, for example, has been a close partner on energy, and cooperation should be intensified.
A second group of Tier B partners, whose security protocols are not yet up to par, should be empowered to develop local solutions based on Tier A technology. Smaller or non-aligned countries—Tier C—should be encouraged to adopt Tier B technologies or develop their own. Lastly, Tier D countries—those whose values do not align with democracies’—should be subject to strict export controls and sanctions on key technologies, ranging from semiconductors to advanced manufacturing products. The objective should be to empower the United States and its allies and partners to collaborate to develop AGI by 2028.
This approach to Tier D countries—China in particular—will be critical. If the competition between Beijing and Washington is conceived of as a track race, the U.S. runner should do everything possible to outrun the Chinese, including—if possible—by tripping them up. Export controls, sanctions, and all the economic means necessary to restrict access to sensitive AGI technologies and expertise must be introduced and driven hard. The intelligence services must be a part of this effort, and work closely with allied countries to curtail illicit supplies of essential technologies to Beijing. Meanwhile, funding must be allocated to institutions working to mitigate AGI-powered threats from adversaries’ efforts. Finally, the United States must take powerful, proactive measures to counteract persistent Chinese misinformation campaigns in unaligned countries. To combat anti-American messages being pushed across Africa and Asia, Washington must launch its own information campaigns, to communicate its desire to share AGI’s benefits and see a world at peace.
Uncontrolled AGI carries its own risks, and its design, deployment, and development must be regulated in a manner that does not restrict innovation. The key step will be the creation of a regulatory framework that is sufficiently flexible to adapt to AGI development’s rapid pace, without allowing it to stray into those areas of life and technology that must be protected. The United States should not imitate the European Union’s regulatory approach, which imposes limits on the scale and power of models and which will as a consequence stunt innovation. Instead, Washington should try something different, and introduce a tiered regulatory system, which applies varied oversight depending on a model’s capabilities and risks. Some aspects of AGI may be relatively benign and require limited oversight. Others, including those with military applications, carry far greater risks and will require intensive oversight. Transparency will be critical, requiring companies to disclose information regarding their systems, including their algorithms and training data, and this whole regulatory edifice must be equipped with the necessary tools of enforcement, including oversight and heavy fines for non-compliance.
Regulation will not be enough, though. A major risk is that AGI, even if correctly programmed, begins to exhibit unexpected behaviors that threaten humans or society. This will be a technical challenge for Washington, requiring the development of adversarial testing methods to identify vulnerabilities and biases in AGI systems. All technologies touching areas that, if they were to go wrong, would have catastrophic effects must be subject to regular and persistent testing, spanning a wide range of scenarios. Red-teaming will be essential, and leaders will be tasked with developing contingency strategies if bugs begin to show. Elements of this approach have already been applied to AI models but, due to the radically more powerful nature of AGI, these methods must be adopted widely and intensified accordingly.
Companies must also be encouraged to develop AGI in an ethical manner. Policing by consent is always the most effective way to operate, and Washington should pursue deep collaboration with business in this effort. The president should regularly gather the leaders of the key companies, to stress Washington’s support, and to establish collaborative methods to ensure ethical behavior. That way, the relationship between business and government will not become bogged down in unending discussions over what constitutes “safe” or “robust” practices, while innovation proceeds unhindered.
There is one further consideration. A U.S. victory in the AGI race will mean nothing if China promptly purloins the technology and bends it to its own ends. To stop that from happening, Washington must take steps to ensure the security and integrity of AGI systems and data, to guard against cyberattacks and data breaches. This must be done in collaboration with businesses, to ensure that their processes are as tight and technologically secure as the intelligence services.
AGI has the potential to transform society in ways that, historically, only governments have had the power to accomplish. That is a potential problem, because the public currently have no ability to challenge it, shape it, support it, or oppose it. This is a novel challenge for democratic nations. Although technological progress has generally been driven by the private sector, and has been largely beneficial to consumers, it has until now generally only affected lives if an individual chose to adopt it.
AGI is different. Its effects will touch everybody, irrespective of a person’s desire to use the technology. That creates a novel and potentially dangerous democratic challenge, whereby every member of the public will find their lives altered, in a manner over which they may feel that they have no control, and in a direction in which they have no choosing.
Public feeling about AI is already negative. In 2023, a Gallup poll found that 21 percent of Americans trusted businesses to use AI responsibly, and only 6 percent were confident that AI would lead to higher numbers of jobs. This public doubt and concern can be neutralized through the development of novel structures that empower the public to discuss and debate AGI, as well as influence its development. The next U.S. president should, therefore, institute an advisory Citizens AI Council to provide public input on AGI policy decisions, and make recommendations on the means by which this technology can be harmoniously introduced. That effort should be matched by a drive to explain the technology and its advantages to the general public through programs like U.S. Senator Mike Rounds’ proposed AI literacy strategy.
In 1946, less than a year after atomic bombs were dropped on Japan, Winston Churchill spoke at Westminster College in Fulton, Missouri. Surveying the development of nuclear technology, he warned that “the dark ages may return, the Stone Age may return on the gleaming wings of science, and what might now shower immeasurable material blessings upon mankind, may even bring about its total destruction.” The United States faces a similar challenge today. Let us work to ensure that AGI showers immeasurable material blessings upon mankind—and that it is made in the United States of America.
The post The Artificial General Intelligence Presidency Is Coming appeared first on Foreign Policy.
