Search the archive.

Despite the strong bipartisan interest in AI regulation—as well as support from leaders of major technology companies, and the general public—passing comprehensive AI legislation remains a challenge. No consensus has emerged around either substance or process, with different groups of members developing their own versions of AI legislation through different procedures. In the House, a bipartisan bill would punt the issue of comprehensive regulation to the executive branch, creating a blue-ribbon commission to study the issue and make recommendations. Majority Leader Chuck Schumer (D-NY) convened a bipartisan working group that includes Senators Martin Heinrich (D-NM), Mike Rounds (R-SD), and Todd Young (R-IN). They are leading a novel process for developing AI legislation that involves “AI Insight Forums,” closed-door sessions with Senators and key stakeholders. The Forums are designed to educate Senators as a mechanism to lead to consensus outside of the normal congressional committee process. Other Senators, including Senate Judiciary Subcommittee on Privacy, Technology, and the Law Chair Richard Blumenthal (D-CT) and Ranking Member Josh Hawley (R-MO) have announced their own legislative framework informed by the traditional committee process. Without an emerging consensus, the future of comprehensive AI legislation remains uncertain. At the same time, however, more than a dozen bipartisan bills have been introduced on a range of specific AI-related topics in both chambers of Congress. Targeted legislation introduced so far includes bills to promote U.S. leadership in AI R&D, protect national security, compel disclosure of AI use, secure U.S. elections from deepfakes and other AI-generated misinformation, address the impact of AI on U.S. workers, and help the federal government leverage AI to deliver services. With bipartisan support and widespread interest in AI issues, it is likely that at least some targeted AI legislation could become law in the near future.

In the fact sheet accompanying the executive order, the White House also speaks of working with Congress on bipartisan AI legislation so that the United States can lead global responsible innovation for the technology. The order’s urgency and accelerated timelines are likely requirements that reflect not only the speed with which the underlying technology is advancing but also the impending electoral season—with its possibility of executive turnover and the certainty of political distraction throughout the Capitol. In his remarks announcing the executive order, President Biden referenced existing congressional initiatives, such as Senate Majority Leader Chuck Schumer’s (D-N.Y.) SAFE Innovation Framework, that are building toward the development of bipartisan AI legislation. A cornerstone of Schumer’s approach is to fast-track the development of such legislation by bringing together members of Congress, industry leaders, and members of civil society in a series of AI Insight Forums. Those forums have concluded, and reports indicate that legislative drafting has commenced. However, some senators have expressed frustration with the nature of the process and have moved forward with their own AI legislation. Sen. Josh Hawley (R-Mo.), for example, long interested in opening up avenues for plaintiffs to sue technology companies directly, recently attempted to introduce a bill to clarify that Section 230’s liability shield does not apply to generative AI. Whether a truly bipartisan agreement on AI legislation can be reached in Congress in an election year is by no means assured.

How do you see the public debate changing around IP? In recent years we have seen vastly enlarged levels of public engagement, thanks in large part to the Internet, which has brought users into direct contact with IP issues. This is likely to continue. There is, however, a tendency for IP to become the battleground for other issues. Take, for example, the controversy surrounding the patenting of the Oncomouse by Harvard University in 1988 (see box). People inevitably focused on the rights question because the publication of a patent application is generally the first public expression of a new technology. But in reality the more fundamental issue was whether it was acceptable or desirable for researchers to be able to manipulate higher organisms. Today we see a similar scenario playing out in relation to the application of CRISPR-Cas9 gene-editing technology, which increases the capacity to manipulate all organisms, and has huge implications for health, agricultural and industrial biotechnology. Are you optimistic about the future? In general yes. The stakes are high. We need to hone our risk management techniques.

It is a cherished truism among numerous Western officials: international commerce reduces the risk of war. This idea motivated the British parliamentarian Richard Cobden when he championed the repeal of tariffs on grain in 1846. In 1918, U.S. President Woodrow Wilson called for "the removal, so far as possible, of all economic barriers and the establishment of an equality of trade conditions" as part of his blueprint for world peace. Cordell Hull, the longest-serving U.S. secretary of state, argued in 1948 that "unhampered trade dovetailed with peace." Such thinking persisted during and after the Cold War. In the 1950s, France and Germany sought to mesh their economies, partly in hopes of preventing another disastrous conflict between them. After the Iron Curtain fell, many European politicians saw greater economic links with Russia as helpful for creating a cooperative diplomatic relationship. The United States and its allies worked to integrate China into the world economy for similar reasons. U.S. President Bill Clinton, for example, declared that admitting Beijing to the World Trade Organization would "plainly advance the cause of peace in Asia." And yet scholarly research has not confirmed that such thinking is accurate. Before 1990, academics vigorously debated the question but largely in the abstract: there was a marked dearth of empirical research on the links between the global economy and conflict. The past three decades produced an explosion of empirical studies examining this relationship, but the findings were a mishmash. Today, there is enough research to offer some answers, but they are hardly straightforward. The notion that more trade and globalization inherently curtail war turns out to be mistaken, yet so is the inverse. The relationship between the global economy and international security is, instead, mixed. Some factors matter, and others don't. The types of economic ties that are meaningful appear to have both stabilizing and destabilizing effects. Trade, for example, sometimes dampens conflict and sometimes fuels it. The globalization of production -- that is, the dispersion of economic activity by global firms across borders -- has a stabilizing effect among great powers, but it increases the likelihood of conflict among developing countries. International financial flows -- the cross-border purchase of bonds, stocks, currencies, and so on -- -have no clear effect at all. Washington does not seem to have digested this reality. American leaders once seemed to believe that a policy of economic engagement with China had only upsides for U.S. security; today's leaders seem to think this policy was an abject failure. But policymakers must understand that both views are wrong. If the United States wants to prevent war, it should adopt a nuanced, modulated approach to economic engagement with China -- curtailing it only when significant risks exist. Commerce is not helpful for advancing peace, nor is it detrimental. It is both.

A new tool lets artists add invisible changes to the pixels in their art before they upload it online so that if it’s scraped into an AI training set, it can cause the resulting model to break in chaotic and unpredictable ways. The tool, called Nightshade, is intended as a way to fight back against AI companies that use artists’ work to train their models without the creator’s permission. Using it to “poison” this training data could damage future iterations of image-generating AI models, such as DALL-E, Midjourney, and Stable Diffusion, by rendering some of their outputs useless—dogs become cats, cars become cows, and so forth. MIT Technology Review got an exclusive preview of the research, which has been submitted for peer review at computer security conference Usenix. AI companies such as OpenAI, Meta, Google, and Stability AI are facing a slew of lawsuits from artists who claim that their copyrighted material and personal information was scraped without consent or compensation. Ben Zhao, a professor at the University of Chicago, who led the team that created Nightshade, says the hope is that it will help tip the power balance back from AI companies towards artists, by creating a powerful deterrent against disrespecting artists’ copyright and intellectual property. Meta, Google, Stability AI, and OpenAI did not respond to MIT Technology Review’s request for comment on how they might respond. Zhao’s team also developed Glaze, a tool that allows artists to “mask” their own personal style to prevent it from being scraped by AI companies. It works in a similar way to Nightshade: by changing the pixels of images in subtle ways that are invisible to the human eye but manipulate machine-learning models to interpret the image as something different from what it actually shows. The team intends to integrate Nightshade into Glaze, and artists can choose whether they want to use the data-poisoning tool or not. The team is also making Nightshade open source, which would allow others to tinker with it and make their own versions. The more people use it and make their own versions of it, the more powerful the tool becomes, Zhao says. The data sets for large AI models can consist of billions of images, so the more poisoned images can be scraped into the model, the more damage the technique will cause. A targeted attack Nightshade exploits a security vulnerability in generative AI models, one arising from the fact that they are trained on vast amounts of data—in this case, images that have been hoovered from the internet. Nightshade messes with those images. Artists who want to upload their work online but don’t want their images to be scraped by AI companies can upload them to Glaze and choose to mask it with an art style different from theirs. They can then also opt to use Nightshade. Once AI developers scrape the internet to get more data to tweak an existing AI model or build a new one, these poisoned samples make their way into the model’s data set and cause it to malfunction. Poisoned data samples can manipulate models into learning, for example, that images of hats are cakes, and images of handbags are toasters. The poisoned data is very difficult to remove, as it requires tech companies to painstakingly find and delete each corrupted sample. The researchers tested the attack on Stable Diffusion’s latest models and on an AI model they trained themselves from scratch. When they fed Stable Diffusion just 50 poisoned images of dogs and then prompted it to create images of dogs itself, the output started looking weird—creatures with too many limbs and cartoonish faces. With 300 poisoned samples, an attacker can manipulate Stable Diffusion to generate images of dogs to look like cats.

Further, the application of AI technology has a marked impact on strategic stability among nuclear powers and may undermine the mutual vulnerability and strategic stability of nuclear-armed states of unequal power. Because AI boasts strong capabilities in image and pattern recognition, similar to facial recognition, its ability to recognize still images is strong and it will greatly enhance the effectiveness of strategic reconnaissance. By improving the ability to interpret satellite images and the deep reconnaissance of long-distance UAVs, the country that possesses nuclear superiority will be able to gain further intelligence on the basic characteristics and procedures of its opponent’s nuclear force deployment and movements. This will enhance a country’s confidence in its ability to disarm its opponent’s nuclear weapons with a first strike. When there is a large gap between the nuclear forces of the two countries, this situation will make a preemptive strike by the more powerful one more advantageous, thus weakening the strategic stability of the two countries during a crisis. At sea, AI technology can be used to improve the capability to collect and process a submarine’s sound signature. In anti-submarine warfare, UUVs can also be used to conduct close-range reconnaissance, which can strengthen the capacity to detect and recognize enemy nuclear-powered ballistic missile submarines (SSBNs). On the one hand, this reduces the survivability of the enemy SSBNs, thereby weakening the effectiveness of nuclear deterrence and reducing strategic stability.1 On the other hand, this AI technology can reduce the probability of accidental nuclear war: by improving the ability to identify SSBNs, a country can avoid accidentally hitting an SSBN as part of conventional anti-submarine warfare. Facing the huge quantitative and qualitative advantages of the USA in nuclear weapons, China has maintained a slim and effective retaliatory nuclear force based on mobile strategic missiles with relatively strong concealment, manoeuvrability and survivability. This force is predicated on increasing the opponent’s ‘first-strike uncertainty’ (第一次打击的不确定性) to ensure credible deterrence.2 However, once the USA is able to effectively employ AI technology to improve its reconnaissance capabilities against China’s mobile strategic missiles—enabling discovery of deployment rules, manoeuvring routes and launch site locations—this ‘first strike uncertainty’ would be eliminated. The USA would thereby gain the advantage of being able to decapitate China’s nuclear arsenal. To overcome concerns over this potential, China would have to increase the alert level of its nuclear weapons to ensure the credibility of its deterrent. This would lead to a state in which both sides tend toward pre-emption, weakening the strategic stability of the two countries and increasing the risk of nuclear conflict. In the case of sea-based nuclear deterrence, the employment of AI-based detection technology and offensive UUVs may also have similar effects. In particular, due to the difficulty of underwater communication, it would be difficult for these weapon platforms to receive onshore command and control signals in a timely manner. This would make it difficult to control, much less recall such platforms when engaged in operations.

Potential U.S. adversaries, such as Russia, take seriously the possibility that the United States might leverage its advantage in such technologies as AI to radically improve its counterforce capabilities. For the past several years, Russian military analysts have been engaged in a vociferous debate in the military press about the extent of their country’s strategic vulnerabilities.8 Their tendency to assume that current and future U.S. capabilities pose a dire threat to Russia’s security stokes these anxieties. A major challenge of nuclear strategy is that adversaries may interpret one nation’s secure retaliatory forces as a first-strike threat or a doomsday machine and react accordingly. For instance, the Russians probably conceived of Status-6 as a last-ditch second-strike option exploiting AI to autonomously circumvent U.S. defenses, but Western observers interpreted it as a Strangelovean “cobalt bomb.” AI progress is also contributing to Russia’s doubling down on older types of systems with undesirable strategic properties. For example, with its RS-28 “Sarmat” missile, Russia is reinvesting in large, silobased ICBMs with multiple independently targetable reentry vehicle (MIRV) warheads, a category of weapon it once planned to abandon under the now-defunct Strategic Arms Reduction Talks II treaty. Western strategic theory generally considers large MIRVed ICBMs to be destabilizing because they are ideal for preemptive strikes and are vulnerable to preemption. At the dawn of the millennium, Moscow believed that it could ensure the survivability of its forces by emphasizing mobile ICBMs and scrapping large silo-based missiles inherited from the Soviet Union. However, Russian leaders’ anxieties about potential U.S. threats to the survivability of the mobile ICBMs seem to have changed this calculus and led them to try to ensure retaliation by launching during a U.S. attack instead of riding it out. This is tantamount to the adoption of a launch-under-attack posture that could place great pressure on Russian leaders to launch first in a crisis, increasing the chances of accidental escalation. The Russians recognize that the Sarmat silo would be unlikely to survive a preemptive attack on its own, so its survivability hinges on an associated active defense system, code-named “Mozyr’,” that would attempt to force enemy warheads to detonate at a slight distance from a silo, allowing it to survive the nuclear explosions.

Other examples of artificial intelligence, there’s facial recognition technology being used to identify civilians and casualties, for example. They’re being used to—they’re using natural language processing, which is a type of artificial intelligence that kind of analyzes the way people speak. You think of Siri. You think of chat bots. But more advanced versions being used to kind of read in radio transmissions and translate them and tag them so that they’re able to—that forces are able to go through more quickly and identify what combatants are saying. There’s the use of 3D printing and additive manufacturing where individuals are able to for very cheap—a 3D printer costs a couple—a thousand dollars and you can get it for maybe less if you build it yourself. You can add—you can add different components to grenades to make—and then people are taking smaller commercial drones to kind of make a MacGyvered smart bomb that you can maneuver. So those are some of the kind of commercial technologies that are being pulled into the kind of military sphere and into the battlefield. They might not be large. They might not be military in its first creation. But because they’re so general purpose technologies—they’re dual use—they’re being developed in the private sector and you’re seeing them being used on the battlefield and weaponized in new ways.

Furthermore, involvement of AI technology will also introduce new variables into the stability of the global system of nuclear deterrence. In an era of weak AI, only a combination of AI technology and nuclear weapons can form an effective deterrent system. When AI technology is involved in all aspects of a nuclear deterrent structure, the original system of stability will change. As a data tool, AI provides countries with new offensive capabilities and has a direct impact on the reliability of nuclear weapon use. In a big data environment, however, there are also a number of subjective factors, such as the unpredictability of national will and strategic intent. When subjected to deep learning algorithms, intent may be clarified. These shifts could result in an imbalance in the MAD-based system of nuclear deterrence. The party with a command of AI technology will have the ability to clearly assess the possibility and destructiveness of the other party’s nuclear counterattack, thus having more flexible strategic options, while the side with relatively backward technologies will possess less credible retaliatory capabilities.41 The gap between military powers will again expand and countries’ military strategic aims will accordingly readjust. In other words, the traditional international security system will become unstable.

In an extreme case, AI could undermine the condition of MAD and make nuclear war winnable, but it takes much less to undermine strategic stability. AI advancements merely need to cast doubt on the credibility of retaliation at some level of conflict. Major nuclear powers, such as the United States, Russia, and China, have a shared interest in maintaining the credibility of central deterrence, but they seek regional advantages in pursuit of what they regard as their core strategic interests. Areas where credibility is already strained, such as certain extended deterrence guarantees, are particularly vulnerable to destabilization. The increasingly multipolar strategic environment is also encouraging forms of competition that threaten stability. For instance, the United States is interested in developing the capability to track and target a minor nuclear power’s mobile missile launchers, but Russia and China fear that the same technology could mature into a threat to their more sophisticated retaliatory forces. In a crisis situation, the employment or availability of AI-enabled intelligence, surveillance, and reconnaissance (ISR) or weapon systems could stoke tensions and increase the chances of inadvertent escalation. Finally, the pursuit of advanced military capabilities is liable to cause arms race instability even if those technologies are nonviable, as in the historical case of missile defense. The challenge AI poses to strategic stability is not unique to this particular technology, but it is more acute because of rapid technical progress in AI and its many potential intersections with nuclear strategy. Most of the specific applications AI are likely to be used for, such as analysis of ISR data, controlling autonomous sensor platforms, and automated target recognition (ATR) have been eagerly sought for decades but were beyond the capability of available technology. Even without further breakthroughs, incremental progress using existing AI techniques may make these long-sought goals practical realities in the foreseeable future. Both Russia and China appear to believe that the United States is attempting to leverage AI to threaten the survivability of their strategic nuclear forces, stoking mutual distrust that could prove catastrophic in a crisis. As Paul Bracken observes, ongoing improvements in technology such as AI threaten to “undermine minimum deterrence strategies” and “blur the line between conventional and nuclear war” (Bracken, 2017). AI in the Cold War AI pioneer Marvin Minsky defined AI as “the science of making machines do things that would require intelligence if done by men” (Minsky, 1968, p. v). Since AI research began in the 1950s, the boundaries of the field have shifted as computers have reshaped how humans comprehend “intelligence.” AI has also evolved as theoretical paradigms have shifted in and out of vogue. From the 1950s until the 1980s, a “symbolic” paradigm that aimed to replicate high-level human reasoning predominated, only to be supplanted by a “connectionist” paradigm that sought to emulate the biological basis of human cognition using artificial neural networks. In the 20th century, neither paradigm worked particularly well outside laboratory demonstrations. This triggered occasional periods (sometimes characterized as AI winters) during which funding for AI research was scarce. Thanks to decades of progress in computer science, advances in computing and communications hardware and software, and the rise of cloud computing and big data, AI has advanced rapidly in the past few years, most prominently in the field of “deep neural networks” (DNNs), or neural networks with many layers (Goodfellow, Bengio, and Courville, 2016). The increase in the performance of DNNs has been so spectacular that they have become almost synonymous with AI, but in actuality the older paradigms are also continuing to progress and are in widespread commercial and military use. Some impressive recent AI systems, such as Alphabet DeepMind’s AlphaGo program, which beat the world Go champion, employ DNNs in combination with such older techniques as searching the tree of possible moves. One thing that has remained constant over AI’s 60-year history is its proponents’ high hopes. With enough intelligence, might it be possible to conquer such seemingly impossible problems as poverty and illness—or even win a nuclear war? The intersection between AI and nuclear warfare became a science fiction cliché more than 50 years ago, but their real-world connections are even older. The earliest AI researchers were deeply involved in national security work and secured government support by suggesting that their theoretical studies would soon translate into practical military applications. Claude Shannon asserted in his foundational 1950 article “Programming a Computer for Playing Chess” that making computers play that venerable game would impart theoretical insights that would make “machines for making strategic decisions in simplified military operations” possible “in the nearterm future” (Shannon, 1950, p. 256). In the mid-1950s, researchers created the earliest working AI programs with support from the U.S. Air Force (Simon and Newell, 1958; Newell, Shaw, and Simon, 1959). Potential applications of such machines soon began appearing in the writings of strategic theorists. In the late 1950s, Herman Kahn postulated the notion of “doomsday machines” that would employ 10 computers programmed to recognize unacceptable enemy provocations and retaliate (Kahn, 1960, pp. 145–154). While Kahn intended these as thought experiments illustrating how not to conduct nuclear strategy, science fiction authors latched onto the idea of intelligent computers controlling nuclear weapons, inspiring numerous novels and such films as Colossus (1970), WarGames (1983), and Terminator (1984). While fictional thrillers spin tales of nuclear armed computers run amok, real-world attempts to apply AI to nuclear strategic problems tended to be much more mundane. Neither U.S. nor Soviet officials were inclined to entrust launch decisions to computers, both because they jealously reserved this prerogative for themselves and because automating retaliation was not a logical response to difficult strategic problems, such as compellence or crisis stability. The sole notable exception came from the Soviet Union at the end of the Cold War. Perceiving that the United States aspired to a first-strike capability and anxious that they might be the objects of a decapitation strike, Soviet leaders sought measures to ensure that capitalist aggressors would never go unpunished.6 The Soviet Union reportedly considered developing a system that would have automatically launched surviving ICBMs at the United States following a first strike if it could not contact the Soviet political leadership. It seems that the fully automated version, nicknamed the “Dead Hand,” was rejected in favor of a version, dubbed “Perimetr,” that would automatically delegate launch authority to field commanders but would always require a human in the loop (Hoffman, 2009). According to Russian media accounts, the Perimetr system still exists and uses some kind of AI.7 The United States, meanwhile, explored the possibility that AI could be used to bolster its counterforce capability. One late 1980s research project, the Survivable Adaptive Planning Experiment (SAPE), sought to use the AI technology of the time to enable the United States to target the Soviet Union’s mobile ICBM launchers. The SAPE would not control nuclear weapons directly; rather, it would employ expert systems to translate reconnaissance data into nuclear targeting plans that would then be carried out by manned B-2 bombers. The SAPE was just one part of an envisioned suite of systems and capabilities that, if actualized, would have severely challenged the survivability of the Soviet Union’s nuclear arsenal (Roland and Shiman, 2002, p. 305; Long and Green, 2012). AI and the Emerging Geopolitical Order Although 20th-century AI struggled to actualize these applications, more-recent advances in computing could release their potential. Such contemporary techniques as deep learning are dramatically advancing machine vision and other signal processing applications, which can enhance autonomy and sensor fusion. Autonomy and sensor fusion may be of paramount strategic relevance because they could greatly improve ISR, ATR, and terminal guidance capabilities. All of these might severely erode the means by which nuclear powers assure the survivability of their nuclear forces. Because increased weapon accuracy has long since undermined the survivability of silo-based ICBMs, the United States, Russia, and China put nuclear weapons on submarines and mobile ICBMs that were deemed more likely to survive a first strike. Technologies that make it more likely that survivable forces (such as submarine and mobile missiles) could be targeted and destroyed make it more plausible that one country might threaten a first strike. This undermines strategic stability, because even if the state possessing these capabilities has no intention of actually using them, the adversary cannot be sure of that. Thus, the capabilities can still be used to pressure potential adversaries and perhaps extract concessions during a crisis. Such a capability does not have to be exploited during a crisis to be politically useful. As Alfred T. Mahan observed, “force is never more operative than when it is known to exist but is not brandished” (Mahan, 1912, p. 105). As long as adversaries fear that the capability may exist, they can be cowed into submission without explicit confrontation—the more powerful state can in effect preemptively “win” the crisis. As a consequence, counterforce targeting capability is an enticing prospect for many despite its potential to compromise strategic stability. AI technologies could help enable new breakthroughs in tracking and targeting and in antisubmarine warfare or make it easier for high-precision conventional munitions to destroy hardened ICBM silos (Holmes, 2016). Such capabilities would be especially destabilizing because decisionmakers could threaten to employ conventional weapons much more plausibly than any kind of nuclear attack. A conventional threat would place the adversary under enormous pressure during a crisis, which could force it to capitulate—but could also spiral into nuclear war. Such escalation could happen because the adversary felt the need to use its weapons before being disarmed, in retaliation for an unsuccessful disarming strike, or simply because the crisis triggered accidental use.

Similarly, AI is sparking military systems evolution in the United States and China, two countries that lead in global AI development by a number of different markers, but not yet in a militarily transformative way. Military systems evolution refers to the ways that new technologies are incorporated into new or existing military capabilities and applications. AI applications have advanced in autonomy and robotics, decision support, predictive maintenance and logistics, cyber, and modeling and simulation, yet are subject to many of the weaknesses mentioned above; progress is generally broader than it is deep. Many higher-profile AI-enabled programs in the unclassified realm across different types of applications, such as Joint All-Domain Command and Control (JADC2) or Loyal Wingman, are comparatively nascent. Military access to the useful, clean, and comprehensive data sets necessary to train AI applications can be limited. As such, AI has done little to spark either operational innovation or organizational adaptation in the United States and China. Operational innovation stems from critical thinking about how militaries can capitalize on new technologies and systems to achieve their objectives; organizational adaptation connotes changes to the structure of the fighting force to exploit new systems and patterns of operations. For example, German blitzkrieg warfare centered new operations around speed and mechanized (tank) maneuvers, leading to the organizational adaptation of separating tank units from soldiers moving on foot. In the United States today, current and forward-looking joint and service operational concepts continue to emphasize the importance of traditional elements like long-range fires, maneuver, information dominance, and joint force projection as augmented but not transformed by AI. In China, AI similarly does not appear poised to transform the informatization concept of warfare, which emphasizes dominating the information environment as part of an asymmetric strategy to counter U.S. comparative advantages. Intriguingly, though, the future “intelligentization” concept discussed by some Chinese strategists, envisions AI as helping dominate the “cognitive” domain—including an adversary’s psychology, perception of information, and even brain function—and represents longer-term thinking about disruptive operations. Regardless, though, with military AI applications largely too immature to dictate shifts in operations, neither the United States nor China has altered its military organizations based on AI. U.S.-China AI Competition Through the RMA Lens So, what can policymakers do with this information? How can insights gleaned from the RMA framework inform U.S. policy approaches to tech competition with China? Why should defense analysts, technologists, or the public care if AI is not sparking imminent revolutionary changes to military affairs? For one, thinking in RMA terms helps to evaluate mental models for AI’s impact on security and defense along an “evolutionary to revolutionary” spectrum. AI might spark evolutionary near-term change in the next decade or so, augmenting reconnaissance-strike capabilities without overturning dominant patterns of operations. In the United States, discussion about JADC2 points to this possibility—connecting “sensors to shooters” has been a joint goal since the Gulf War. AI could continue advancing autonomous capabilities, fulfilling noncombat functions like predictive maintenance, or augmenting cyber capabilities. However, analysis of seemingly less revolutionary change, including careful analysis of technological developments that may seem less obviously militarily applicable, must accompany evolutionary framing. With certain AI applications and methods advancing rapidly, it is important to acknowledge AI’s game-changing potential and how unforeseen technological applications could generate new, revolutionary military systems and operations. Analysts and policymakers who offhandedly dismiss even seemingly militarily irrelevant applications of AI do so at the risk of ceding advantages to creative adversaries. It is better to be safe than sorry. Policymakers can also use the RMA framework to identify indicators of AI-driven transformative change. Open-source technology monitoring could uncover potentially disruptive academic or private-sector tech progress. Specific AI signposts could include advanced development of enabling computing technologies, new, less compute-heavy learning methods, or progress in explainability that could help warfighters trust previously opaque capabilities. Advanced government signposts could include AI assurance cases and dependable test and evaluation frameworks that suggest a need and ability to assess application performance at scale, as well as new doctrinal guidance incorporating AI that informs future training and force development. For example, examining new Chinese doctrine will help determine the importance the People’s Liberation Army sees for AI applications in its future force. Passing these early- and late-stage signposts would indicate that AI technology may be making disruptions to military paradigms increasingly likely. Finally, and crucially, the RMA framework can help identify the missing pieces of the intellectual puzzle when it comes to military applications of emerging technology. To date, the United States and China have made little progress on operational innovation and organizational adaptation with AI. A future AI RMA instigated by the United States or China will therefore require more grappling with how AI might change patterns of operations, thinking, and experimentation that historical case studies illustrate—such as the carrier aviation RMA and China’s writing on intelligentization—can be done even before new technological applications become operational. In the case of the carrier RMA, both the U.S. and Japanese navies possessed cutting-edge (for the time period) carrier and aviation technologies. However, the unique way the United States came up with operational patterns and structures using the new capabilities transformed naval warfare from being centered around battleships and thus provided a decisive advantage compared to the Japanese. Without critical thinking and experimentation around how new technologies will be applied, their impact on military affairs is not guaranteed to be revolutionary. Conclusion AI may not yet or even imminently spark a revolution in military affairs. Nonetheless, using the RMA analytical framework to delve into why it is not doing so, and how it might in the future, can help us think systematically about AI’s potential impact on future warfare. AI’s impact could one day be transformative enough in areas like cyberwarfare or even nuclear deterrence without over-hyping nascent capabilities. To ensure that militaries use AI-enabled capabilities thoughtfully and responsibly, military futures analysis should be as rational and grounded as possible.

AI hype has truly taken over the word of data. Every event you attend, everyone is discussing AI, ML, GenAI, LLMs, and AGI. Every data product seems to have some sort of AI features built into them. There are seemingly AI experts everywhere now. I am in the data industry and even I get lost in the flood of almost daily updates on new AI announcements. Lots of it is just noise. But if you shovel away the BS, you can indeed uncover some great gems for the analytics there. Don’t get me wrong, I am a supporter. But a practical one. I hate the hype. I like to see tangible benefits of any tech before I become a believer. When we deploy IOblend, we expect it to make it a significant positive impact on the client’s challenges. We don’t want to sell them our solution if their current setup does the job just fine. Every new technology implementation brings with it changes to the BAU. Change tends to be disruptive. The benefits must be material enough to justify putting the company through the change process. The AI implementation is no different. It must make a visible and material positive impact on the business or it’s a waste of money. That’s not what’s yet happening though in many cases. Especially in the GenAI universe. The hype forces companies to think that they must deploy the tech wholesale or be forever left behind. Like a panacea for all their business problems. The businesses are expected to drop what they did with data and analytics already and throw resources at the AI. With gusto. What’s happening then? This is where the hyped-up promises of digital wonders meet the reality of technological limitations, business data and processes. The challenges often arise from the misplaced expectations by the senior management. “We gave you a big budget, why aren’t we seeing the impact yet?” The biggest mistake here is thinking that AI is somehow different (i.e. simpler) from all the past technological breakthroughs and thus should be much quicker to implement and reap financial benefits. It is not by any means. Early adoption phase Many businesses are still in the exploration phases, experimenting with AI (and the more recent GenAI) applications rather than implementing them at scale. The initial use cases often focus on innovation and proof-of-concept rather than revenue generation and cost reduction. AI is not easy. It is the cutting edge of technology and requires time to embed and gain wide adoption. Even the largest companies struggle with AI implementations (McDonald's drive-thru AI pulled after multiple errors). And we’ve all seen the struggles Google is having. Integrating AI into existing business processes and systems is complex, requiring significant investment in both time and resources. Just like any other technology companies always struggled to implement for decades now. High initial costs Adopting AI and GenAI solutions requires substantial upfront investment in technology infrastructure, such as high-performance computing resources and cloud services. A lot of businesses are still on-prem and buried in spreadsheets. Throwing AI at them wholesale brings a host of other expenses they might not be ready for or even anticipated. Then, there is a high demand for skilled AI professionals, leading to increased costs for hiring, training and retaining talent. Data dependency Practical AI and GenAI models require large volumes of high-quality data to be effective. Many businesses struggle to gather, clean, and maintain the necessary data. That data problem is not AI-related but has plagued companies forever. The AI just put a bigger spotlight on it. Bad quality data makes the models produce spurious results, quickly diluting trust in the technology itself. “What is this rubbish? I’m not using it again.” Concerns over data privacy and security regulations also limits the extent to which businesses can leverage GenAI in particular. Many businesses are very cautious about what data they share for training. Regulatory and ethical concerns The regulation for AI is still evolving, creating uncertainty for businesses about compliance requirements and potential legal risks. Moving goalposts. On the ethical front, there are concerns around the use of AI, such as bias and transparency, can slow down the adoption process as businesses work to address these issues. Even small biases can have a significant impact on large-scale businesses, such as insurance brokers, mortgage lenders, etc. Organisational readiness Resistance to change can put massive obstacles to the adoption of new technologies. There are the obvious concerns around job security. “Will the AI replace what I do here?” The hype makes you believe that a lot of admin and entry level analytical jobs will be gone soon. Not the best way to win hearts and minds. Then, if the employees start using the AI, they still need time to adapt to new workflows and tools. They need help training, handholding and close support by the AI experts (usually external consultants). Even for those who eagerly embrace the technology, there is still a learning curve. None of it happens overnight or comes at no cost. Market dynamics In some industries, there is a bit of “wait and see” approach happening among the senior leadership. The competitive advantages provided by AI and GenAI are not yet immediately apparent. At least not in a sufficiently material manner to make them go after the tech with full force. Especially in markets where competitors are also only just beginning to adopt the technology. What we also observe is that many businesses didn’t grow their digital budgets to accommodate AI and GenAI. But rather they reallocated resources away from other digital activities. This shows the unwillingness to make big bets just yet. And, naturally, broader economic conditions, such as market volatility, high cost of capital and geopolitical risks, impact business investment in new technologies when the payback is not immediately material. Ignore the hype So, what can the business do to avoid frustration when implementing AI and GenAI tech in production? Treat AI just as you would any other technological advancement. How long did it take your company to adopt data warehouses, ERP, CRM, or migrate to the cloud? How costly and painful was it? Revisit the learnings. AI will be no different. Thoroughly evaluate implementation challenges You have built your AI test models and demonstrated the positive effects on your business. Now it’s the time to scale out the PoC. Everyone is excited. Expectations run high. First, choose the right use case, fully understanding the limitations of the AI technology. The hype will have you believing anything is possible. Do your research properly. Get a second opinion from an independent party before you commit to full deployment. It will save you a lot of money and potential embarrassment. Then, you actually need to make it scale. This is when things become tricky. Putting a PoC into production often requires more time and effort that you have planned. Again, just like with any other technology. One the biggest technical challenges that we encounter is issues with data. Mainly, it comes to the ability (or lack of thereof) to integrate, curate and manage vast amounts of data needed to feed the algorithms in production. Data sources First, the sheer volume and variety of data required for effective AI models requires robust data integration capabilities. Data coming from sensors, point solutions, events, databases/data lakes, batch and streaming must be aggregated and harmonised efficiently. This harmonisation leads to a unified data set that AI algorithms can then process to produce reliable outputs. And it must be done at scale and in an automated fashion to reduce errors and cost. For instance, a financial institution deploying AI for fraud detection must integrate transaction data, customer profiles, and external threat intelligence. Without automated integration, this process would be laborious and error-prone, easily leading to significant gaps in the AI’s analysis. Data quality Once you automate the data collection, you need to ensure data quality and consistency. AI models are only as good as the data they are trained on. You need a consistent way to detect and rectify anomalies, fill missing values, and standardise data formats at scale. This continuous data curation is essential for the model’s performance and longevity. Doing this manually will never work at scale. A case in point is the healthcare sector, where AI models assist in diagnosing diseases. The success of these models relies on the quality of medical records, imaging data, and patient histories. With automation of your data integration, they feed these models with the consistent, high-quality data, enhancing diagnostic accuracy and patient outcomes. Real-time processing Many operational uses cases with material benefits require dynamic decisioning. The AI in these settings will need a constant and reliable feed to robust data in real time. For example, in autonomous driving, AI systems must process sensor data live to make split-second decisions. Real-time data integration and management make this decisioning possible by ensuring that data from various sensors is rapidly and accurately combined and analysed. Take another example from the manufacturing industry where we helped bring automation of data integration. They had to process factory data manually once a week from seventeen plants. By the time they processed and analysed the data, events like stoppages, machine breakages, inventory levels had all been out of date. We moved them to real-time data processing and the analytics became live. It enabled them to start using AI for a more in-depth analysis. If your business never worked with real-time data, that is another layer of technology that you need to adopt. Luckily, real-time data integration is a lot easier today than it used to be. Products like IOblend made that accessible and cost-effective. But that consideration must still be taken when deciding on AI investments. ROI timeframe Developing and deploying production-grade AI and GenAI solutions takes considerable time. The payback will not be immediate. A lot of the use cases now are compartmentalised PoCs and MVPs. The data is handpicked, manually curated and prepped. Scaling these out to production are lengthy and costly exercises, especially if the underlying data estates are not in tip-top conditions. Hint: most aren’t. Measuring the direct financial impact of GenAI initiatives can also be challenging, especially in the early stages. ROI may be realised through indirect benefits like improved efficiency, customer satisfaction, and innovation rather than immediate revenue increases. Putting a $$ number on those benefits is not straightforward, so the CFOs tend to look at such investments with a degree of scepticism. You can see in the recent market announcement that the AI-related stocks are starting to lose momentum. The market expectations were mismatched with the reality of the ROI timeframes. The hype had a lot to do with that. Hype is never helpful The AI hype is not helping the AI technology's development and public perception. We all know why it’s done. There is a lot of money being invested in the industry. But this “overenthusiasm” leads to disillusionment and scepticism when the technology fails to deliver immediate, miraculous results. This then damages the genuine progress and investment. The hype also obscures the realistic capabilities and potential limitations of AI, creating a gap between our understanding and actual technological advancements. The disconnect results in regulatory and ethical challenges. The policymakers and the public may not be adequately informed to address the real issues and potentials of AI, ultimately stalling its responsible and effective integration into society. And finally, the hype pushes the businesses to rush implementations, which leads to expensive blunders. It’s critical to spend time understanding the technology and its implications on your business thoroughly beforehand. This requires some investment up front, but trust me, this is a lot cheaper than plunging vast resources into AI and seeing bubkis in return.

But here’s where the hype goes off the rails. Not so fast: The fast pace of AI change is an illusion Georgia Tech computer science professor Mark Riedl posted on the X social network that it’s plausible Q* is simply research at OpenAI aiming for “process supervision” that replaces “outcome supervision” and that when OpenAI published general information about this idea in May “no one lost their minds over this, nor should they.” The idea of replacing word or character prediction with some kind of supervised planning of the process of arriving at the result is a near-universal direction in labs working on LLM-based genAI. It’s not unique to OpenAI. And it’s not a world-changing “breakthrough.” In fact, AI doesn’t advance with individual companies or labs making massive breakthroughs that change everything. It only feels that way because of OpenAI. Although OpenAI was founded in 2015, its culture-shifting ChatGPT chatbot was released only about a year ago. Since then, the tech world has been turned on its head. Thousands of LLM-based apps have emerged. Tech funding turned hard toward funding AI startups. And it feels like this brand of AI has already changed everything. In reality, however, OpenAI’s innovation wasn’t so much in AI, but in the project of providing access to genAI tools to the public and to developers. The company’s ChatGPT services (and its integration by Microsoft into Bing Search) caught hundreds of other AI labs in companies and universities off-guard, as they had been proceeding cautiously for decades. ChatGPT set the rest of the industry scrambling to push their own research into the public in the form of usable tools and open APIs. In other words, the real transition we’ve experienced in the past year has been about the transformation of AI research from private to public. The public is reeling, but not because AI technology itself suddenly accelerated. Nor is it likely to unnaturally accelerate again through some “breakthrough” by OpenAI. Actually, the opposite is true. If you look at any branch of any technology or set of technologies that approaches AI, you’ll notice that the more advanced it gets, the slower further enhancements emerge. Look at self-driving cars. I was physically present at the DARPA Grand Challenge in 2004. In that contest, the Pentagon said it would grant a million dollars to any organization with an autonomous car capable of finishing a 150-mile route in the desert. Nobody finished. But the next year and in the next DARPA Grand Challenge, the Stanford entry finished the route. Everyone was convinced that human-driven cars would be obsolete by 2015. Fast forward to 2023 and activists are disabling autonomous cars by placing traffic cones on their hoods. The highest level of autonomy is Level 4, and no Level 4 car is available to the public or capable of driving on any roads other than pre-defined, known routes and under certain conditions of time and weather. That last 5% will likely take longer to achieve than the first 95%. That’s how AI technologies tend to progress. But we lose sight of that because so many AI technologists, investors, boosters, and doomers are true believers with extreme senses of optimism or pessimism and unrealistic beliefs about how long advancement takes. And the public finds those accelerated timelines plausible because of the OpenAI-driven, radical changes in the culture we’ve experienced as the result of AI’s recent public access. So, let’s all take a breath and relax about the overexcited predictions about how AI in general, and Q* in particular, are about to change everything everywhere all at once.

An interagency assessment released this month by the White House stresses that many federal research facilities are now well beyond their 40-50-year estimated lifespan, with about half rated to be in “poor or critical condition.” “U.S. scientists and engineers are now faced with conducting 21st century R&D in many facilities designed in the 1950s that cannot support modern research and current laboratory practices in health and safety,” the report states. The report flags numerous negative statistics, including that nearly 40% of Department of Energy facilities are rated as substandard or inadequate, more than 60% of the square footage at the National Institute of Standards and Technology is classified as being in poor or critical condition, and about 75% of NASA facilities are beyond their designed lifespan. It attributes the situation to decades of inadequate funding and deprioritization of maintenance. It also warns that remediating aging facilities instead of replacing them still incurs significant costs while missing out on the benefits of properly upgrading those facilities. The report points out cases where other countries’ research infrastructure has passed that of the U.S. For example, China has overtaken the U.S. in number of supercomputers and the European Centre for Medium-Range Weather Forecasts has pulled ahead of NOAA’s Global Forecast System. However, the issues driving this divide are not limited to advanced equipment, with the report citing widespread problems with basic infrastructure such as insufficient power, unreliable HVAC systems, and failing plumbing systems at U.S. facilities. It also argues that the U.S. is struggling to attract and retain top scientific talent as private companies and other countries offer better compensation and more stable research environments. Among its recommendations, the report proposes that U.S. research agencies increase their benchmarking against facilities abroad and promptly identify places where a gap may be forming, noting such an effort could also be used as a basis for increasing international cooperation.

The primary goal of the bill is to ensure that copyright owners have visibility into whether their intellectual property is being used to train generative AI models. If enacted, the law would require companies to submit notices to the U.S. Copyright Office detailing the copyrighted works used in their AI training datasets. These notices would need to be filed within 30 days before or after the public release of a generative AI system. The Copyright Office would then maintain a public database of these notices, allowing creators to search and see if their works have been included. The hope is that this transparency will help copyright holders make more informed decisions about licensing their IP and seeking compensation where appropriate. Potential Benefits for Copyright Owners The bill’s notice requirements and public database could be a significant step forward for creators looking to protect their IP in the age of AI. By providing a way to track the usage of their works in generative AI systems, the law would empower copyright owners to take action against unauthorized use and potentially negotiate licenses or royalties. This could be particularly valuable for individual artists, writers, and other creators who may lack the resources to monitor the vast landscape of AI applications on their own. Challenges and Obstacles for AI Developers While the bill’s intentions are laudable, it raises several practical challenges for AI companies. The 30-day window to file notices could be difficult to meet, particularly for startups or smaller organizations with limited resources. The vague requirement for a “sufficiently detailed summary” of copyrighted works used also leaves much room for interpretation and potential non-compliance. Moreover, many AI training datasets are compiled by crawling the open web, which could make identifying and isolating every single copyrighted work extremely burdensome. The technical challenges of building and maintaining a comprehensive, searchable database of notices could be substantial. Impact on AI Innovation The additional administrative overhead and legal liabilities introduced by the bill could have a chilling effect on the pace of AI innovation. Smaller players may struggle to keep up with the compliance burdens, while larger tech giants with more resources may be able to adapt more easily. This could further entrench incumbent advantages and raise barriers to entry for new startups. Some companies might choose to move their AI development efforts outside the U.S. to avoid these regulations altogether. This could put the U.S. at a competitive disadvantage in the global race for AI leadership. If more copyright owners start denying the use of their works for AI training, it could also limit the data available and hinder progress in certain domains. AI systems rely on large, diverse datasets to learn and improve, so restrictions on training data could have far-reaching effects.

As if that were not enough, by shifting the structure and balance of global power, AI complicates the very political context in which it is governed. AI is not just software development as usual; it is an entirely new means of projecting power. In some cases, it will upend existing authorities; in others, it will entrench them. Moreover, its advancement is being propelled by irresistible incentives: every nation, corporation, and individual will want some version of it. Within countries, AI will empower those who wield it to surveil, deceive, and even control populations-supercharging the collection and commercial use of personal data in democracies and sharpening the tools of repression authoritarian governments use to subdue their societies. Across countries, AI will be the focus of intense geopolitical competition. Whether for its repressive capabilities, economic potential, or military advantage, AI supremacy will be a strategic objective of every government with the resources to compete. 1}le least imaginative strategies will pump money into homegrown AI champions or attempt to build and control supercomputers and algorithms. More nuanced strategies will foster specific competitive advantages, as France seeks to do by directly supporting AI startups; the United Kingdom, by capitalizing on its world-class universities and venture capital ecosystem; and the EU, by shaping the global conversation on regulation and norms. The vast majority of countries have neither the money nor the technological know-how to compete for AI leadership. Their access to frontier AI will instead be determined by their relationships with a handful of already rich and powerful corporations and states. This dependence threatens to aggravate current geopolitical power imbalances. The most powerful governments will vie to control the world's most valuable resource while, once again, countries in the global South will be left behind. This is not to say that only the richest will benefit from the AI revolution. Like the Internet and smartphones, AI will proliferate without respect for borders, as will the productivity gains it unleashes. And like energy and green technology, AI will benefit many countries that do not control it, including those that contribute to producing AI inputs such as semiconductors. At the other end of the geopolitical spectrum, however, the competition for AI supremacy will be fierce. At the end of the Cold War, powerful countries might have cooperated to allay one another's fears and arrest a potentially destabilizing technological arms race. But today's tense geopolitical environment makes such cooperation much harder. AI is not just another tool or weapon that can bring prestige, power, or wealth. It has the potential to enable a significant military and economic advantage over adversaries. Rightly or wrongly, the two players that matter most--China and the United States--both see AI development as a zero-sum game that will give the winner a decisive strategic edge in the decades to come. From the vantage point of Washington and Beijing, the risk that the other side will gain an edge in AI is greater than any theoretical risk the technology might pose to society or to their own domestic political authority. For that reason, both the US. and Chinese governments are pouring immense resources into developing AI capabilities while working to deprive each other of the inputs needed for next-generation breakthroughs. (So far, the United States has been far more successful than China in doing the latter, especially with its export controls on advanced semiconductors.) This zero-sum dynamic--and the lack of trust on both sides--means that Beijing and Washington are focused on accelerating AI development, rather than slowing it down. In their view, a "pause "in development to assess risks, as some AI industry leaders have called for, would amount to foolish unilateral disarmament. But this perspective assumes that states can assert and maintain at least some control over AI. This may be the case in China, which has integrated its tech companies into the fabric of the state. Yet in the West and elsewhere, AI is more likely to undermine state power than to bolster it. Outside China, a handful of large, specialist AI companies currently control every aspect of this new technological wave: what AI models can do, who can access them, how they can be used, and where they can be deployed. And because these companies jealously guard their computing power and algorithms, they alone understand ( most of) what they are creating and (most of) what those creations can do. These few firms may retain their advantage for the foreseeable future--or they may be eclipsed by a raft of smaller players as low barriers to entry, open-source development, and near-zero marginal costs lead to uncontrolled proliferation of AI . Either way, the AI revolution will take place outside government. To a limited degree, some of these challenges resemble those of earlier digital technologies. Internet platforms, social media, and even devices such as smartphones all operate, to some extent, within sandboxes controlled by their creators. When governments have summoned the political will, they have been able to implement regulatory regimes for these technologies, such as the EU'S General Data Protection Regulation, Digital Markets Act, and Digital Services Act. But such regulation took a decade or more to materialize in the EU, and it still has not fully materialized in the United States. AI moves far too quickly for policymakers to respond at their usual pace. Moreover, social media and other older digital technologies do not help create themselves, and the commercial and strategic interests driving them never dovetailed in quite the same way: Twitter and TikTok are powerful, but few think they could transform the global economy. This all means that at least for the next few years, AI's trajectory will be largely determined by the decisions of a handful of private businesses, regardless of what policymakers in Brussels or Washington do. In other words, technologists, not policymakers or bureaucrats, will exercise authority over a force that could profoundly alter both the power of nation-states and how they relate to each other. That makes the challenge of governing AI unlike anything governments have faced before, a regulatory balancing act more delicate-and more high stakes-than any policymakers have attempted.

While AI tools have the potential to assist human creativity, much like other creative tools that have come before it, copyright protection for wholly AI-generated material is not desirable as a policy matter. Wholly generated AI material that is based on copyrighted works ingested by AI developers without compensating the creator or obtaining their permission to ingest their works has the potential to supplant the market for the ingested works. Policymakers should be discouraging such activities, not incentivizing them by granting legal protection to material manufactured wholly outside of the realm of human authorship. Some have suggested that granting copyright protection to AI-generated works is important to incentivize the further development of generative AI systems and technology. However, the prevalent use of AI tools demonstrates that there are already adequate incentives in place. Today, there exist a large number of AI developers and systems, all who offer varieties of GAI tools that are used across a spectrum of creative fields. That number has grown exponentially over the past year and is likely to continue to increase in the coming months and years. Similarly, the number of AI users and customers has expanded significantly. It is clear that no additional incentives related to the copyrightability of output wholly generated by AI are necessary to encourage AI developers and systems to enter the marketplace and prosper.

With U.S.-Chinese relations worse than they have been in over 50 years, an old fairy tale has resurfaced: if only the United States would talk more to China and accommodate its rise, the two countries could live in peace. The story goes that with ample summitry, Washington could recognize Beijing's redlines and restore crisis hotlines and cultural exchanges. Over time and through myriad points of face-to-face contact--in other words, reengagement--the two countries could settle into peaceful, if still competitive, coexistence. Talk enough, some analysts contend, and the United States and China might even strike a grand bargain that establishes stable spheres of influence and something akin to a G-2 to solve global problems such as climate change and pandemics. From this perspective, the dismal state of US.-Chinese relations is not an inevitable result of two ideologically opposed great powers clashing over vital interests. Rather, it is a mix-up between partners, blown out of proportion by the United States' overreaction to counter China's overreach, as Susan Shirk, a Sinologist and former U. S. deputy assistant secretary of state, has put it. For the past two decades, the thinking goes, China has simply been doing what rising powers usually do: flexing its muscles and demanding a greater say in global affairs. Although many of China's actions, such as its menacing of Taiwan, worry advocates of reengagement, the main target of their critique is the United States--specifically, its relentless pursuit of primacy and the self-serving actors behind it. In this dark imagining, grandstanding politicians, greedy defense contractors, sensationalizing pundits, overzealous human rights activists, and belligerent bureaucrats fan the flames of rivalry for profit, creating an echo chamber that crowds out different perspectives. Some individuals are supposedly repeating hawkish narratives to protect their careers. The result, the journalist and author Fareed Zakaria has argued, is that "Washington has succumbed to dangerous groupthink on China." The fact that most Americans also hold hawkish views on China just provides more evidence of how irrationally aggressive U.S. policy has become. "The problem today isn't that Americans are insufficiently concerned about the rise of China," the historian Max Boot has insisted. "The problem is that they are prey to hysteria and alarmism that could lead the United States into a needless nuclear war." For those advocating reengagement, the solution to this cycle of hostility is straightforward. First, defuse tensions through vigorous diplomacy, commerce, and people-to-people exchanges. Next, create a new forum where officials from each country can meet regularly to hash out agreements. According to the historian Adam Tooze, regardless of the exact structure of negotiations, the basic objective is the same: "accommodation of China's historic rise." For some advocates of reengagement, accommodation would merely entail reducing trade barriers to China, a move U.S. Treasury Secretary Janet Yellen proposed earlier this year. Other observers, however, favor more drastic concessions. The political scientist Graham Allison, for example, has urged in these pages that the United States accept China's traditional sphere of influence in Asia. Presumably, that would mean giving Beijing greater freedom in the South China Sea, letting go of Taiwan, and relinquishing American power in the region. It is an enticing vision. The world would certainly be better off if great powers could settle scores through diplomacy rather than by squaring off in a security competition. Yet the history of-great-power rivalry, and of US.-Chinese relations in particular, suggests that greater engagement is unlikely to mend ties between the countries and, if performed hastily, could actually catalyze violent conflict. Of the more than two dozen great-power rivalries over the past 200 years, none ended with the sides talking their way out of trouble. Instead, rivalries have persisted until one side could no longer carry on the fight or until both sides united against a common enemy. For example, the United States and China paused their rivalry to ally against the Soviet Union during the latter half of the Cold War, a contest that ended only when the Soviet Union sputtered into terminal decline. In every case, shifts in the balance of power were preconditions for sustainable settlements. Before those shifts, periods of détente were usually just chances to regroup and reload for the next round of competition. In some cases, such as when the United Kingdom sought to improve relations with Germany from 1911 to 1914 and again in 1938, pursuing détente paved the road to war. The United States and China are unlikely to buck this pattern. Their vital interests conflict and are rooted firmly in their respective political systems, geographies, and national experiences. Many of the connections binding the countries together, such as their extensive trade, are also driving them apart by giving policymakers additional reasons to fight and pressure points to exploit. Neither side can make major concessions without exposing itself. And after decades of dealing with each other, both governments have accumulated long lists of grievances and view the other with deep mistrust. The United States tried to work with China repeatedly from the 1970s to the 2010s, yet top Chinese leaders consistently viewed US. outreach, especially the American attempt to integrate China into the U.S.-led liberal order, as an insidious form of containment-a plot designed to weaken the grip of the Chinese Communist Party and lock China into economic dependence and political subservience to the West. American outreach to China during this period was more extensive than the proposals being seriously considered by US. policymakers today. Nevertheless, these overtures failed to fundamentally change Chinese assessments of American intentions or dissuade efforts by the CCP to dominate East Asia and beyond. The fact is that the US.-Chinese rivalry is unlikely to wind down without a significant shift in the balance of power. The United States needs to make policy choices based on this reality and not get caught up in a fantasy. This does not mean cutting off diplomacy or shutting down talks completely, but being clear eyed about what that type of engagement can realistically achieve. There are reasons to hope for a medium-term mellowing of Chinese power that might open space for a real diplomatic breakthrough. To get there, however, the United States and its allies must deter Chinese aggression in the near term and avoid concessions that disrupt favorable long-term trends. BAD BLOOD 'The United States and China have become what political scientists call "enduring rivals," meaning countries that have singled each other out for intense security competition. Over the past few centuries, such pairs have accounted for only one percent of the world's international relationships but more than 80 percent of its wars. Think of the repeated clashes between India and Pakistan, Greece and Turkey, China and Japan, and France and the United Kingdom

Still, there are some reasons to think otherwise. As former US diplomat R. Nicholas Burns recently , the United States maintains “alliances in Europe and Asia, and the Russians and Chinese do not.” In addition, the American military has begun to ISIS, killing more than fighters over the past two and a half years. So do a resurgent Russia, an ascendant China, and the emergence of the Islamic State suggest that US power is ebbing, or are these challenges exaggerated? What do US officials really think about these matters? If we take stock of their public statements as a whole, the foreign policy establishment certainly appears concerned about the latest challenges to US empire, especially the uncertainty that Trump’s election has introduced. But they also remain quite confident in their power to shape the world and steer the United States into a new age of global hegemony. The World’s Superpower Over the past few years, a number of high-level officials have expressed great confidence in the durability of US hegemony. Not only have they insisted that the declinist thesis is wrong, but they have argued that the United States will remain the world’s dominant power well into the future. In May 2016, two former high-level officials the more confident view for the Senate Committee on Foreign Relations. In a statement to the panel, former Secretary of State James Baker said that the United States would continue to lead the international system. Certainly, “much of the rest of the world — countries like China, Brazil and India — are catching up with us,” Baker conceded. “Still, we should remain the world’s preeminent leader for the foreseeable future.” Former National Security Advisor Thomas Donilon agreed. As long as the United States takes the proper precautions, he said, it “will continue to be the world’s leading and most powerful nation for a long time to come.” Donilon also rejected the declinist thesis, calling it a “myth” that should not be taken seriously: “The idea that America is in decline does not stand up to a rigorous analysis of our national balance sheet of strategic assets and liabilities,” Donilon asserted. “The truth is that no nation can match our comprehensive set of enduring strengths.” Other Obama administration officials offered similar views. Last October, for example, Secretary of State John Kerry noted that the United States maintains tremendous economic advantages. “We’re the richest country on the face of the planet,” Kerry . In fact, the Obama administration made great strides in expanding US economic power across the globe. President Obama, who during his final months in office that he had “made it a priority to open up new markets overseas,” boasted that his administration had “increased US exports to the world by more than 40 percent — to record levels.” At the same time, administration officials also pointed to their other great advantage in world affairs: American military power. As Secretary of Defense Ashton Carter in April 2015, “it will take decades — and let me repeat that: decades — for anyone to build the kind of military capability the United States possesses today.” Obama expressed similar sentiments during his farewell tour. The United States possesses “the greatest military in the history of the world,” Obama . “Make no mistake,” he continued, “even with the challenges of recent years — and there have been challenges — our allies and adversaries alike understand America’s military remains, by far, the most capable fighting force on the face of the Earth.” In short, contrary to those who warn of waning US hegemony, high-ranking officials insist that the United States remains the most powerful country in the world. “We are, without a doubt, the world’s superpower,” CIA Director John Brennan this past September.

Recent events, however, should give progressives pause about what a multipolar world means. Certainly, belief that the United States would act unilaterally, unwisely, and unfairly was sustenance to multipolarity’s cheerleaders at home and abroad. “Calling for a ‘multipolar world’ [became] a euphemism for opposing the Iraq war,” Foreign Policy’s Elizabeth Dickinson observed. While President George W. Bush’s invasion of Iraq is widely disparaged today both in the United States and international circles today, only the passage of time will—like Harry S. Truman’s intervention in Korea—determine its place in history. However, as China, Russia, Iran, and Turkey challenge the post–World War II international order and seek regional dominance, their behavior reflects the reality of what a multipolar world truly means. What would a multipolar world look like? Consider, for example, China’s abduction of Meng Hongwei, the head of Interpol. Can international and multilateral organizations work if their leaders must subordinate themselves to the increasingly hardline Chinese Communist Party? Russia’s protection of Syrian President Bashar al-Assad’s use of chemical weapons, meanwhile, signals the end of a post–World War I consensus. Critics can castigate the United States for upwards of two hundred thousand Iraqi deaths, which occurred as the result of the U.S. invasion of Iraq and its aftermath, but they often ignore the fact that America was not to blame for most of these deaths. Instead, they were a result of suicide bombings and other terrorist attacks sponsored by Tehran and others who too often avoid progressive accountability for their choices. Put another way, American training and tactics seek to minimize civilian casualties in all phases of conflict. Russia, however, has no such qualms: ask the Chechens. Nor should America’s history in the eighteenth and nineteenth centuries provide an excuse to ignore those, like China, who today would incarcerate a million citizens in re-education camps solely because of their ethnicity or religion. Turkish president Recep Tayyip Erdoğan has been forthright in his desire to reshape the post–World War II international order. “Rather than justice, the existing global order is producing chaos, injustice and despair,” he argued in September 2018. But it is Erdoğan who has waged ethnic cleansing against Syrian Kurds, a campaign he may soon expand, and Erdoğan defended genocide in the Sudan and opened the floodgates for foreign fighters into the Islamic State. It’s one thing to criticize or second-guess U.S. policy; that is the domain of democratic debate in Congress and during presidential campaigns and in dueling op-eds in a free press, but it is nonsensical to suggest that any alternate pole—Moscow, Beijing, Tehran, or Ankara—would make the world a more peaceful place or better respect the evolving international human-rights regime. The one thing revisionist states have in common is that they tolerate neither debate nor dissonance. That is a lesson Uighurs, Ukrainians, and Kurds have discovered, as have Meng Hongwei and Turkey’s imprisoned journalists. It is time American and European progressives recognize it as well, for the stakes are far greater than simply American power.