With changes ranging from the rise of lethal autonomous weapons in warfare to the immense economic dynamism being unleashed by tools like ChatGPT, national power is being transformed in the age of artificial intelligence (AI). However formidable these changes have been so far, it is almost certainly still too early to know precisely which ones will be the most significant in relation to power, or even to correctly imagine them. History shows that it takes decades, if not generations, for truly transformative technologies to reshape the contours of power and for a new balance to be consolidated in its dynamics.
For example, the transformative battlefield potential of the invention of the stirrup, which allowed combatants to harness the power of horses in ways previously unknown, could have been recognized early, but it would be many years before that the feudal order and the new stratagems of conflict would allow that technology to reach its full potential. Assessing a contestant's power in terms of number of tempers might have been tempting as a measure of new war power; but of course that ultimately proved less important than the overall strength of the cavalry in the armies that would then appear and the strength of the political authorities in managing the new class of heavily armed knights.
Similarly, Europe's gunpowder boom had obvious potential, but it was not until states and armies co-evolved with that new technology that the most distinctive elements of gunpowder age power truly emerged. The same could be said of the eras of steam, electricity and nuclear energy. And today too: in the age of AI, raw computing power may be an important part of the power equation for states moving forward, but the more consequential vectors of power may still need to take shape as states move forward. States, societies and economies co-evolve alongside advances in AI capabilities.
Despite uncertainty about the ultimate form that power will take in the age of AI, there are strong indicators of what will determine success; especially considering the two AI superpowers (China and the US) vying for the upper hand in that revolution. Similarly, while the ultimate effects of such competition remain largely unknown, it is increasingly clear that the power unleashed by AI will dramatically alter fundamental aspects of the human condition, from war to politics to life itself. human biology. Now, to begin to evaluate how national power will change in the age of AI, we must first understand what actually makes AI powerful, what the determining factors are for its strength, and what the ultimate challenges will be. of their mature abilities.
According to the US Department of Defense's 2018 Artificial Intelligence Strategy, AI is “the ability of machines to perform tasks that typically require human intelligence (for example, recognizing patterns, learning from experience, drawing conclusions, making predictions, or undertaking actions) either digitally or as intelligent software used in autonomous physical systems.” By this broad definition, AI is not a new technology: it was developed in fits and starts between the 1970s and 2012 by engineers who struggled to find ways to synthesize intelligence into computer code.
However, the power of AI has expanded rapidly since 2012 thanks to what is sometimes called the deep learning revolution, a radical shift in AI development made possible by three converging advances in the field. .
First, the researchers came up with a new way to operate an algorithmic architecture that surpassed all other paradigms related to how AI systems are structured. Loosely inspired by the human brain, neural networks refract data through layers of interconnected nodes capable of establishing sophisticated and dynamic ways of interpreting and using that same data. Now, as machine learning engineers discovered, neural networks are capable of working on many tasks only if the constructed network is very large: with numerous layers, most of which include a multitude of nodes, also called parameters, and which are now used to calibrate the size of an AI model.
Secondly, the Internet began to produce sufficient volumes of data to feed neural networks and for them to take flight. Unlike human intelligence, it turns out that neural networks can establish very robust representations of data if – and only if – they have enormous amounts of examples in the data on which to train.
Finally, the processing power of computers, or computing, began to reach levels that facilitated the emergence of robust AI. Unsurprisingly, running the huge amounts of data required through the large number of parameters of neural networks generated gigantic computing needs, hence the requirement for mature supercomputing power. The computing needs of today's cutting-edge AI systems are so great that they require thousands of the world's most advanced microprocessors to operate non-stop for months at a cost exceeding $100 million. These three factors (algorithms, data and computing) form an AI triad that continues to drive the current revolution.
An opaque artificial brain
Although the results of the deep learning revolution are unquestionable, the quality of the advance has been unusual, since the researchers who have made the advance possible cannot fully explain how neural networks work internally. It is as if researchers have found a way not so much to build, but to grow increasingly larger and more powerful digital brains, but without fully understanding how those brains work inside.
Combining the right ingredients results in a larger, more capable brain (often more capable than inventors' brains in certain areas); but, although the inventor can conduct experiments with the resulting brain, the exact mechanisms of the brain that produce intelligence remain opaque. Despite this opacity, the engineer knows how to make larger brains grow, which sometimes surprise their creators with unexpected capabilities. And while more fundamental advances may be on the way, for now there appears to be a long road ahead to make enormous progress just by increasing the amounts of data, computation, and parameters used to create AI systems. In other words, the more data, more computation, and more neural network parameters engineers can gather, the more robust AI tools will be for a wider variety of more sophisticated applications, with no clear end in sight to the potential for improvement. .
Unprecedented capabilities
The improvements are unlocking a series of unprecedented capabilities that are about to revolutionize virtually all areas of human activity in the sciences, economies and militaries. This revolution is based on three fundamental capabilities.
The first is superhuman pattern analysis: AI systems have demonstrated remarkable abilities to accurately identify phenomena in images, text, and other types of data, a characteristic evident in the ability of AI systems to detect ailments from medical images more accurately than human specialists.
A second area is that of amazing strategic skills. AI systems have learned to far outperform humans in more and more strategy games, from chess to poker, Diplomacy to StarCraft, and now outperform humans in real-life strategic areas, such as air combat in the army. Another area is the resolution of very complex multifactorial problems. In the same way that the invention of differential and integral calculus made much of modern physics possible by opening new ways to measure, calculate and express the dynamics of movement, AI is also making it possible to solve for the first time very complex problems that require attention to more variables or dimensions that conventional methodologies have been able to manage. The quintessential example is protein folding, where AI has reduced the time invested by researchers to understand the form and function of a protein from weeks or months to seconds and has revolutionized a problem that has plagued biology for many years. .
Presented abstractly, with only a handful of examples, it may be difficult to fully appreciate how transformative these AI capabilities can be in pattern recognition, strategic analysis, and complex multifactor calculations. But these novel skills are applicable to almost any field, they have suddenly boosted humanity's capabilities in a series of areas and already facilitate advances in biology, medicine, linguistics, agriculture, robotics, logistics, weapons and more areas.
Generative AI
Such capabilities alone would be enough to predict profound changes in economies, armies and societies. Now, as anyone who has experienced ChatGPT knows, there is another element of AI's emerging capabilities that deserves special consideration: its ability to generate original content. Companies like OpenAI, Anthropic, and Google DeepMind have figured out how to build generative AI systems that not only recognize patterns in language, images, and other data, but also have such mastery over those patterns that they can produce text, art, code computer technology and convincing and original speeches at high speed and scale.
It remains to be seen where advances will ultimately go, but many managers of the most advanced laboratories in the world are driven by the almost religious vision of building from generative AI a form of artificial intelligence that surpasses humanity in all domains. of knowledge. According to that paradigm, generative AI systems like ChatGPT are already developing powers of reasoning, creativity and problem-solving simply by digesting all the vast information available on the internet and elsewhere, and will soon absorb the expertise of generative AI systems as well. More focused AI mentioned before. Proponents of this goal envision systems that, having internalized the entirety of human knowledge, will be able to take charge of large swathes of economies by autonomously planning and executing large-scale economic and scientific projects.
Regardless of how close AI pioneers get to that grand vision of artificial superintelligence, the tens of billions of dollars devoted to the effort each year have already produced powerful tools that have immense potential to change the contours of the world. national power and that they will continue to change it. In particular, the wide range of tools and capabilities that have already emerged from the AI revolution are sufficient to catalyze transformative shifts in national power as they spread across militaries and economies, whether or not visions are yet achieved. more glorious. In that sense, no political leader has summed up this evidence better than Vladimir Putin in a televised speech to schoolchildren: “Whoever becomes a leader in this sphere will become ruler of the world.”
The fight for AI supremacy
In light of the enormous potential of AI technologies in the coming years, the US and China are busily engaged in a battle for supremacy. Business and government leaders in both countries have placed big bets on AI as a major source of their future power and are far ahead of all other countries in the strength of their AI sector. Now, although the reality of a Sino-American struggle over AI is clear (and although the dimensions of that struggle are evident), what remains an unknown in the coming years will be the likely way in which this struggle will develop.
Shortly after taking office, China's top leader Xi Jinping lamented that Western countries have managed to dominate the world thanks to their use of the most advanced technologies: “the sharp weapon of the modern state.” Consequently, China set in an official statement in 2017 the goal of overtaking the US as the world leader in AI by 2030. Similarly, the US Government's National Security Commission on AI noted in 2021 that AI is “the most powerful tool in generations” and advocated a robust strategy to thwart Chinese efforts to overtake the US. As geopolitical competition between the two countries intensifies (with fears of a war over Taiwan in the horizon), this dynamic only accelerates.
Now, the competition over AI differs radically from previous technological rivalries between superpowers. The space and nuclear races of the Cold War were races directed by states in pursuit of very defined objectives that required very specific technologies. Instead, AI development is led by the private sector and covers a very wide range of applications. Therefore, the challenge facing the US and China is how to cultivate an AI ecosystem that is more competitive than that of the rival and that can ultimately facilitate advantages in terms of national strength.
Data, computing and talent
National security expert Paul Scharre identifies in Four battlegrounds: Power in the age of artificial intelligence four areas that will define the strength of the US and Chinese AI ecosystems. Two of them have a direct relationship with the elements of the triad that has facilitated the current revolution: data and computing.
For years, many observers believed that China had a natural data advantage in that its authoritarian, hierarchical state was able to consolidate seemingly endless amounts of data on its own population, helped in part by its enormous surveillance apparatus. However, recent analyzes have complicated that assessment. On the one hand, the quality of Chinese data may not be useful in many cases for AI training. On the other hand, as generative AI becomes an increasingly important element of new AI capabilities, the fact that less than 2% of the internet is written in Chinese may also represent a significant disadvantage. Still, there remain some data areas where China has an unquestionable advantage; For example, in the area of biodata, where the Government has adopted an industrial approach focused on curating huge data sets. The consequences of the advantages and disadvantages of each country will be seen in the coming years.
In terms of computing, the US appears to have, at least for the moment, a clear advantage over China. The ultra-advanced microprocessors that are the backbone of cutting-edge AI development rely heavily on designs, software and hardware produced by the US and its allies (especially Taiwan, Japan and the Netherlands). In fall 2022, the Biden administration banned the sale of critical advanced chips and chip manufacturing equipment to China, which has thwarted China's long-term plans for AI development. Two months earlier, the Science and Chips Act had been enacted, aimed at strengthening the US semiconductor sector. Still, China could find ways around those obstacles by speeding up its own chip industry or finding inexpensive ways to achieve similar computing power with less advanced chips. Chinese companies are investing huge resources in both lines of effort and are showing some progress.
The third area of competence is talent, which is similar to the third element of the AI triad, algorithms. There is a shortage of advanced engineers in the world capable of making adjustments, improvements and innovations to AI algorithms. There, too, the picture is mixed: China has surpassed the US in the number of most cited AI publications by multiple metrics, and produces more AI engineers than any other country. However, the most decisive advances in machine learning continue to come from the US and its allied countries (especially the United Kingdom and Canada). Furthermore, much of Chinese AI talent is trained in the US and would prefer to stay in that country after studying. If the US manages to direct its immigration towards AI, it would substantially increase its workforce in that sector while decreasing China's.
The institutions
Institutions represent the fourth and final area of competition in AI, as government departments, militaries, businesses, and societies work to find and effectively integrate AI solutions to recognized problems. This area differs in some ways from the other three: while data, computing and talent are mainly related to the expansion of AI capabilities, institutions are the key to disseminating this technology to specific applications. According to some estimates, this dimension of competition in AI is the most important: the country that leads technological innovation will not reap the fruits of said innovation if it fails to effectively disseminate the power of innovations.
Historically speaking, militaries that have enjoyed relative supremacy have often had difficulty overcoming institutional and psychological inertia and meaningfully assimilating new technologies; And this points to a possible advantage for the Chinese People's Liberation Army, which assumes that it must harness technology to catch up with US military capabilities. On the other hand, early indications also indicate that the dynamism of American private companies developing AI capabilities is much more effective than that of their Chinese counterparts. Again, only time will tell.
Political and industry leaders agree that the coming years are likely to prove very formative in the fight for the edge in AI due to a relatively clear path for dramatic advances. However, the extent to which the next few years will determine the long-term future is still subject to controversy. At one extreme, optimists who believe that frontier AI will lead to the development of capabilities that exceed the collective intelligence of humanity assume that whoever achieves that level of superintelligent AI will achieve a potentially insurmountable advantage. From such a point of view, achieving superintelligence or something similar in the next 5-10 years could trigger a virtuous circle of self-improvement and would relatively quickly leave all other competitors behind. Others suspect that the trajectory of AI development is unlikely to be so focused on general-purpose hyperintelligent systems, and instead envision a world of diverse expertise in different applications, all evolving organically. In such a world, some countries are likely to specialize in specific areas of AI development or use, with strategic implications that would evolve over time. Of course, the reality could end up being something in between or completely different.
Now, what is unlikely is that there will be a slowdown or a significant pause in the development of AI. Even if there are unexpected blockages in fundamental research, there are enough proven innovations in the private sector that are likely to remake national power in the years to come, so the stakes for humanity are high.
What is at stake in AI
Nothing so reflected what was at stake during the Cold War as the fact that the struggle itself transformed fundamental aspects of the human condition. It was more than evident in the nuclear field, an area in which the weapons advances achieved during World War II accelerated in the Cold War arms race and ended up giving rise to humanity's ability to (for the first time) annihilate itself. to herself. As much as political and military leaders have been able to sow destruction in the past, never before has power been so concentrated, and in the hands of so few, that weapons could intentionally or accidentally wipe out the entire nation. humanity and the environment in which we live.
With similar importance, human beings ceased to be terrestrial creatures, a feat that may seem less world-altering now than it did at the time, when not even a lifetime had elapsed after the Wright brothers' first flight. Now, with the planetary colonization visions of Elon Musk and others, the importance of the “great leap for humanity” may increase.
However, what is surprising about humanity's two new capabilities to destroy the world and to live outside of it is that, despite reshaping the collective horizons of what is possible for humanity, these advances depended largely on the technological relationship between two countries. The combination of animosity and cooperation between the Soviet Union and the US ultimately determined the timing, conditions, and management of civilization-altering technological capabilities for the rest of the world.
The same thing, although in different areas, will happen in the age of AI. Some of these changes are speculative: there is no doubt that the dominance of AI will have enormous economic implications, but the extent to which it will displace or redefine, as some suspect, human work through automation remains to be seen. Still, in four specific areas, the fight between the US and China for the edge in AI means that major historical changes in the human condition are at stake:
1. The war
The field of war is the most recognized. Although technology has tended to move combatants further away from direct combat, lethal autonomous weapons represent a radical change in the dynamics of large-scale violence. Some rudimentary forms of lethal autonomous weapons have been around for a few years, but advances in AI and robotics mean that the slow evolution from warriors to soldiers to operators will soon reach its endpoint: war engineers, thereby that the role of human beings will be completely reduced.
Of course, giving machines the autonomous ability to distinguish, select, and kill human beings raises immensely complex moral questions. But even if ongoing efforts to ban or very strictly limit killer robots succeed, the progress of AI will mean that war – one of the most enduring and pernicious features of human existence – will no longer require, at least in theory, to be fought by human beings, at least not by those from technologically advanced countries.
2. Governance
Governance represents a second category in which the ground is fertile for fundamental transformations. Much has already been written about the growing competition between the Chinese model of AI-based techno-authoritarianism and an American democracy struggling to confront technology-induced polarization and misinformation. However, beyond that competition, it is perhaps easy to overlook a fundamental transformation in the history of the political order. As Jamie Susskind has convincingly pointed out in Future politics, AI technologies provide states with the opportunity to project a new form of power distinct from the police, military and social structures that have conventionally been their main levers for imposing order. Incorporating AI technologies into smart cities, public services, surveillance devices and social media platforms can provide states with a means to impose their will with minimal human intervention. The first glimpses of such an evolution can already be seen in China's far western Xinjiang province, where totalitarian AI-based smart cities are already playing an important role in imposing Beijing's wishes through highly advanced techniques. surveillance and population management.
But there may also be promising developments in the opposite direction, though the US is now struggling to cope with AI's formidable capabilities in misinformation, disinformation and surveillance in its democracy. Nothing prevents AI from also being able to enhance more democratic forms of government. Given the growing capabilities of AI in natural language, there may still be ways to more dynamically and effectively engage public opinion on a scale that empowers citizen voices and better builds national consensus. In any case, depending on how the US-China rivalry plays out, it is quite possible that some populations will be functionally governed both by the force of sovereign AI systems and by bureaucracies, police and military forces.
3. Decision making
A less valued, but no less profound, area that AI is about to disrupt is the fragility of modern societies. The ability of AI systems to analyze data and make decisions faster than humans has obvious advantages, but it has the unfortunate drawback that if AI systems go awry, they can also do so very quickly and in complex ways in the face of which human beings will have difficulty adapting. Think, for example, of the 2010 flash crash on Wall Street where unexpected, machine-speed interactions between trading algorithms caused a trillion dollars in stocks to disappear in a matter of minutes. Although the authorities have theories about the cause of what happened, in reality no one knows for sure what happened, since not even the registration system could keep up with a completely runaway algorithmic negotiation. As more advanced AI is integrated into increasingly important systems, societies are leveraging to an unprecedented degree powerful but often insecure mechanisms with failure modes for which we are ill-equipped.
According to the chairman of the Securities and Exchange Commission responsible for regulating US markets, financial markets are being driven more than ever today by AI-powered trading systems and in a way that is not sustainable. The critical electricity, water, telecommunications and gas infrastructure systems that modern societies depend on – already facing the threat of conventional cyber attacks – may become increasingly vulnerable to disastrous AI-driven cyber attacks. The US has promised that AI will never command nuclear weapons, but automated systems are increasingly finding their way into the sensors and alarms involved in nuclear decision-making. And increasingly powerful AI models used in biological and pharmaceutical research can, with a single click, be used deliberately or inadvertently to design unprecedented chemical or biological weapons. The extent to which societies are exposed to such catastrophic risks will depend on how demanding business and international competitive pressures on AI become and also on the safety standards (national and international) that become established. As in the Cold War, these issues will be determined largely by the actions of the superpowers.
4. Human biology
The biological future of humanity is the final frontier of the human condition at stake in the age of AI. Although genetic engineering technology has existed since the 1970s (and has since sparked multiple debates in bioethics), the functional scope of the possibilities for significant editing of the human genome has until now been minuscule. The enormous, ultra-complex amounts of information encoded in human DNA are too unwieldy and complicated for conventional scientific methods to accurately understand, let alone manipulate. However, they are just the kind of complex multifactorial problem that AI is uniquely suited to solve if given enough genetic data.
As AI unlocks the secrets of DNA, it will reveal which sets of genes relate to which traits, opening the doors to gene therapy to cure genetic diseases, as well as designer babies and genetic enhancement. The preemptive ban on hereditary gene editing by some seventy countries may not be able to withstand the dawn of an era in which (along with advances in gene editing techniques) AI will provide the practical means to fundamentally alter the structure of humanity in strategic ways.
Above all, because China is already ahead of itself. In 2020, the US Director of National Intelligence reported that China had already begun testing members of its military to develop biologically enhanced soldiers. Just two years earlier, a Chinese scientist went public with an experiment to produce the world's first genetically modified babies.
That means that unless Beijing sterilizes those three people, China has already introduced irreversible genetic modifications to its genetic heritage; including, unfortunately, some involuntary mutations resulting from the scientist's poor execution of the experiment. Initially, the Chinese government praised the feat, but international protests forced a reaction against the scientist, who ended up imprisoned. However, the die is cast: the Chinese scientist in question has been released and is once again defending his controversial research into genetic engineering. All while the Chinese government has taken an industrial approach to collecting the massive amounts of data needed for an AI-driven genetic engineering future; and not only from their own population, but also from the world at large through state-subsidized genetic sequencing companies with large international customer bases. If China succeeds, it may not only become a leader in gene therapies capable of treating diseases, but also in the creation of genetically modified super soldiers, in the development of IQ and in other adjustments made to the structure of humanity, which that will raise ethical and strategic questions of the first order to its rivals.
Conclusions
Humanity therefore finds itself in an awkward position at the dawn of the AI era. Since the deep learning revolution was unleashed in 2012, AI has emerged as a transformative technology with clear potential to propel states to new heights, even though the exact paths of propulsion remain unclear. While China and the United States fight for dominance of this new source of power, it is likely that the competition will be decided based on which superpower manages to best bring together the computing power, data, talent and institutions to innovate and disseminate AI technologies effectively. Now, surely the greatest manifestations of power in the age of AI (those capable of fundamentally altering the human condition) are decided by the dynamics of that competition, in a similar way to how the trajectory of nuclear technology and space exploration ended up being decided based on the dynamics of the cold war.
In an ideal world, the new powers of AI to remake humanity's relationship with war, governance, disasters, and human biology would be guided by reflective and deliberative processes between societies and countries, processes that would serve to answer some of the most important moral and political questions that humanity has faced to date. The reality is likely to be very different.
Bill Drexel is a research associate in the National Security and Technology Program at the Center for a New American Security (CNAS).
