Introduction
‘AI is the new electricity’, as the common refrain goes. It is now standard for social scientists and policymakers to compare artificial intelligence (AI) with electricity, the quintessential general-purpose technology (GPT). Military innovation scholars acknowledge this comparison, yet they have done little systematic research into the implications of GPTs. Much work treats AI as a relatively narrow technological advance, in the mould of nuclear weapons or aircraft carriers.Footnote 1 It does not reckon with AI as a GPT, differentiated by its pervasiveness, scope for continual improvement, and strong synergies with other technologies. The comparison is gestured at but not seriously examined.
How do GPTs, like electricity and AI, influence the military balance of power? Taking economic productivity as an analogue for military effectiveness,Footnote 2 we extend insights on GPTs and economic growth to the implications of GPTs for military transformations. Differing from narrower technologies, GPTs influence military effectiveness through a pathway characterised by three features: breadth of impact spread across many military innovations, delayed timeline of widespread adoption, and indirect productivity spillovers. We call this process a general-purpose military transformation (GMT).
Equipped with a better understanding of how GPTs shape military effectiveness, we posit that GMTs differentially advantage militaries connected to a robust industrial base in the associated GPT. Regarding narrower technologies, differentials in their military adoption can be explained by the fit between a single military innovation and a military's culture, financial resources, organisational capital, tactical incentives, etc.Footnote 3 Whiles these factors affect the adoption of specific military innovations linked to a GMT, a military's ability to draw on a robust industrial base in the GPT affects all military innovations linked to a GMT. To effectively exploit a GMT, militaries must draw on talent, industry, and infrastructure in the civilian realm, where the momentum for a GPT's development lies.
To empirically support our reasoning about GMTs, we examine the evolution of electricity in military affairs. Surprisingly, very little scholarship directly examines the military consequences of electricity – widely recognised as one of the most significant technological innovations in history.Footnote 4 In the latter half of the nineteenth century, a cluster of electrical innovations, including the electric dynamo (1866) and the transformer (1886), helped create a versatile energy system with many industrial applications in lighting, communications, transportation, and machinery. A GPT was born. Looking back after the Second World War, informed observers ranked electricity among the three great influences on naval warfare in the twentieth century.Footnote 5 Eliminating the ‘naval’ qualifier would not be a stretch.
Military electrification exhibited the three theorised features of a GMT. First, the impact of electricity on military power materialised through a broad range of military applications, including communications, fortifications, transportation, and weapon control systems. Second, electricity significantly upgraded industrial productivity, which increased military production potential. Third, like the slow progression of electrification across economic sectors, the spread of electrical innovations across military branches and divisions took many decades.
In line with our hypothesis, the extent to which militaries took full advantage of electrification depended on their connection to a robust base of electrical talent, industry, and infrastructure in the civilian economy. Taking advantage of a GMT goes beyond acquiring a single electric dynamo or adopting one electricity-related military innovation. The Russian military, for instance, pioneered the use of electronic countermeasures in combat in 1904. However, unlike Britain, a leader in military electrification, Russia's weak industrial base of electrical technology prevented it from fully exploiting the electrification GMT.Footnote 6 Evidence of Russia's ability to keep pace with Britain in adopting submarines, a non-GPT, help separate the effects of the GPT dimension from other factors that could explain this military electrification gap.
This article directly engages with key academic and policy debates. First, we develop a novel explanation for how GPTs alter the military balance of power. The limited literature on GPTs and military power posits that because GPTs are characterised by private sector dominance and relatively low fixed costs, they rapidly diffuse from technological leaders to laggards, thereby narrowing gaps in military capabilities.Footnote 7 We argue, instead, that laggards do not inevitably catch up in GMTs. Rather, the long, slow process of a GMT differentially advantages militaries that can tap into a robust industrial base in the GPT. By highlighting the unique effects of GPTs, we contribute to a growing approach to studying the impacts of emerging technologies on international security, which focuses on specific technological dimensions such as complexity,Footnote 8 disruptiveness,Footnote 9 and dual-use.Footnote 10 Moreover, departing from the tendency to focus on relatively narrow technological developments, such as new weapons systems,Footnote 11 our study of electricity broadens the universe of cases for investigating the military implications of technological change.Footnote 12
Second, our research directly bears on current debates over how AI could shift the balance of military power.Footnote 13 To date, much of the discussion emphasises the narrow effects of specific AI applications, such as autonomous weapons.Footnote 14 AI's potential influence on military power through its effects on industrial productivity is rarely considered. Some scholars also suggest that AI will significantly transform military effectiveness in a relatively short timeframe, and that AI will enable rising powers to leapfrog the US in military strength. As the conclusion will show, a GMT-based approach points towards different conclusions on all these fronts.
In what follows, we first deduce key propositions about the impact of GPTs on military affairs by adapting insights from economic and historical studies of GPTs. Our theory of GMTs is composed of three features influencing how and when GPTs affect military affairs, along with an explanation for why GMTs differentially advantage certain militaries. Leveraging primary and secondary accounts, we then illustrate the explanatory value of our theory with a historical case study: the evolution of electricity in military affairs. Finally, we conclude by discussing the limitations of our analysis and reflect on the military implications of artificial intelligence, which is plausibly a GPT as impactful as electricity.
1. Theory: GPTs and military power
Economists and economic historians largely agree that GPTs are defined by three characteristics. First, GPTs offer great potential for continual improvement. While all technologies offer some scope for improvement, a GPT supports an entire research paradigm that drives forward adaptations and modifications. Second, GPTs are characterised by their pervasiveness. A wide range of sectors apply GPTs in a wide variety of uses. Third, GPTs have strong technological complementarities, which means their full benefits come from adjustments in related technologies.Footnote 15 For instance, complementary advances in machine tools were critical to factory electrification, enabling a power distribution system in which electric motors drove individual machines.Footnote 16 Tying all three characteristics together, David describes GPT adoption as an ‘extended trajectory of incremental technical improvements, the gradual and protracted process of diffusion into widespread use, and the confluence with other streams of technological innovation’.Footnote 17
Capitalising on this extended GPT trajectory is neither easy nor automatic. Variation across domestic institutions, international linkages, and other factors can generate significant cross-national differences in GPT adoption, even between advanced economies. Take, for example, the gap in computer usage between the US and Japan from 2000–05, decades after initial innovations in personal computing. During this period, according to our calculations based on the Cross-country Historical Adoption of Technology (CHAT) dataset, the US's computerisation rate nearly doubled Japan's figure.Footnote 18
Translating GPT theory to military transformation
By adapting insights from studies of GPTs and economic productivity, we theorise about how GPTs transform military effectiveness. Taking military divisions and branches as application sectors for a GPT, we translate the pattern of how a GPT spreads across a national economy to a military context. We first deductively articulate three key features of GMTs. Equipped with a more complete view of GMTs, we then pinpoint why different militaries are better able to exploit GMTs.
For our translation to work, we must establish that the characteristics of GPTs also apply to military transformation. GPTs possess great potential for continual improvement, become pervasive in their wide variety and range of military applications, and have strong technological complementarities with existing military technology systems. The computer is one such example. Upon entering military systems, it continually improved along many technical dimensions, provoked significant structural changes, and eventually found a wide variety of uses across many branches and units.Footnote 19
We can differentiate GPTs from militarily significant innovations that meet some but not all three of the GPT criteria. Consider dual-use technologies, which have both commercial and military applications. Some dual-use technologies, such as aircraft, exhibit continual improvement and offer strong technological complementarities. Aircraft propulsion systems, however, have a limited variety and range of applications, which means we should not expect them to give rise to GMTs. The breadth of the transformations produced by GPTs often warrant the ‘-ization’ suffix. The computerisation, or digitalisation, of the military describes how computer advances prompted a wide array of innovations across the entire military. There is no equivalent for aircraft engines.Footnote 20
Of course, GMTs differ from GPT-driven economic transformations. Unlike the market dynamics of a nation's economy, intra-military competition involves different units pursuing a nominal shared mission and budget flows from one primary source. Military effectiveness and economic productivity are analogous but not the same; assessing the impact of technological innovations in the military realm is much more difficult than in the civilian economy.Footnote 21 Despite these differences, some of the ways in which GPTs interact with military organisations follow a similar trajectory as GPTs in economic systems. In particular, the first two features of GMTs, regarding their broad impact pathway and prolonged timeline of diffusion, draw directly from stylised facts in the existing GPT literature.
Like translations of all kinds, adapting the foundations of GPT theory to military affairs requires some modifications. The full effects of GMTs must include GPT-induced boosts to military production capabilities. Additionally, militaries are typically reliant on the civilian economy to advance the GPT's development, particularly after its initial incubation, which means the trajectory of a GMT will depend on a military's connection to the evolution of a GPT in the civilian economy.Footnote 22 The rest of this section describes the three key characteristics of GMTs in more detail.
GMTS: Three features of GPT trajectories in military affairs
First, GPTs directly enhance military effectiveness by spurring a wide range of military innovations. Studies of military innovation often gravitate to the most visible and graphic part of warfare: the projectile or other mechanism of force.Footnote 23 In contrast to the impacts of weapons technology on military effectiveness, which materialise through a relatively narrow pathway, GPTs produce many downstream applications with various uses across the entire military. For instance, one function of computers, the capability to process large amounts of data, bears on military targeting, logistics management, decryption, and many other domains. The impact of a GPT on military effectiveness depends on the distribution over all such functions.
As a result, the foreseeability of a GPT's effect on the conduct of warfare is very limited. GPTs will influence military effectiveness in unanticipated ways, often through lengthy causal chains that involve complementary innovations in adopting military sectors. Although a certain degree of unpredictability applies to the effects of all technologies on military affairs, the breadth of military applications affected by technological changes can vary. Some innovations present a very limited set of applications, though these applications can interact with the strategic landscape in many ways (Figure 1). Nuclear fission technology, for instance, had a relatively bounded set of military applications – namely, nuclear weapons – but the interaction between nuclear weapons and the strategic landscape evolved in multifaceted, unpredictable ways. However, the set of possible military applications for GPTs is much larger than the corresponding set for other technologies.
The military technology stack – two impact pathways.
Second, GPTs indirectly affect military effectiveness by significantly upgrading industrial production potential. This constitutes a key element of military power.Footnote 24 Analysing hundreds of battles and wars from 1898 to 1987, Beckley finds a positive relationship between economic productivity and military effectiveness. Separate from its total economic output, a more economically efficient nation will translate its economic resources into superior weapons and military organisation.Footnote 25
Assessed on their own merits alone, some of the most transformative technological changes do not tip the scale far enough to significantly affect overall industrial productivity.Footnote 26 As ‘engines of growth’, GPTs are different because their impact on productivity comes from accumulated improvements across a wide range of complementary sectors. Put simply, they cannot be judged on their own merits alone.Footnote 27 According to empirical studies of the steam engine, electricity, and information and communications technologies, the arrival of GPTs precedes a wave of economy-wide productivity growth.Footnote 28
Third, the most consequential impacts of GPTs on military effectiveness occur only after a long period of gestation. In the 1980s, many observers bemoaned the computer's failure to induce a surge of productivity growth, leading to Robert Solow's famous quip: ‘We see the computers everywhere but in the productivity statistics.’ Eventually, we did see the computers in the productivity statistics. Compared to other technologies, GPTs exhibit a more pronounced diffusion lag due to their substantial demands for complementary innovations, organisational changes, and skills adjustments.Footnote 29 We expect similar delayed timelines in the military domain.
This extended trajectory raises difficult questions regarding when a GMT ends. One could trace recent advances military electronics and precision warfare back to nineteenth-century electrical innovations. For our purposes, we date this to when a GPT spreads across a wide range of military applications.Footnote 30 Oftentimes, other fundamental advances, such as transistor innovations in the 1940s, initiate a new GPT trajectory. Thus, advances in precision warfare, though built on a base of electrical advances, are more connected with the transistor GMT than the electrical one.
GMTs and differential advantages
We hypothesise that militaries more connected to a strong industrial base in the GPT are better positioned to exploit GMTs. The broad applicability of GPTs across many sectors, combined with the fact that the civilian economy presents many more application scenarios than the military realm, means that the momentum for a GPT's evolution lies in the civilian realm. Successful adaptation to a GMT requires militaries be able and willing to accommodate civilian-guided GPT development, including by adopting commercial technical standards and managing imbalanced distribution of talent. This distinguishes GPTs from dual-use technologies like aircraft parts and nuclear power, where military and civilian development trajectories significantly diverge, and military capabilities are less dependent on civilian ones.Footnote 31
This proposition engages with claims that GPTs level the military balance of power. Arguing that GPTs rapidly diffuse from technological leaders to laggards, Daniel W. Drezner posits that ‘general purpose tech has a greater leveling effect than prestige tech’.Footnote 32 Michael Horowitz also links GPTs to a levelling effect. ‘If commercially-driven AI continues to fuel innovation, and the types of algorithms militaries might one day use are closely related to civilian applications, advances in AI are likely to diffuse more rapidly to militaries around the world’, he writes. ‘This could change the balance of power, narrowing the gap in military capabilities not only between the United States and China but between others as well.’Footnote 33
A clearer understanding of GPT adoption suggests modifications to arguments about levelling effects. In our view, a military's acquisition of a single electric dynamo should not count as successful adoption, as this does not capture whether the military has meaningfully incorporated the GMT associated with electricity. That GPTs are driven forward by civilian applications does not necessitate that they will diffuse quickly to militaries that are technological laggards. Instead, adapting to GMTs is a protracted, challenging process that differentially advantages militaries able to tap into a robust industrial base in the associated GPT. Therefore, a GPT will retrench existing military balances if the countries with strong militaries are also those that can tap into robust civilian sectors linked to the GPT.Footnote 34
2. Case study: The electrification of warfare (late nineteenth and early twentieth centuries)
Research design and case selection strategy
To evaluate our theory, we investigate the impact of electricity on military affairs. Electricity is considered the prototypical GPT, making it a representative case for studying GMTs. The frequent comparisons made between AI and electricity serve as an additional advantage of our case selection. The electrification of warfare is also substantively important. Despite being widely recognised as one of the most significant technological changes in history, electricity's military impact has not received much scholarly attention.Footnote 35
If the evidence from the electricity case supports our theory, we should observe two main sets of implications. First, the impact of electricity on military effectiveness should exhibit the three features of a GMT: broad impact pathway, indirect productivity benefits, and prolonged gestation period. Second, we expect differentials in military electrification based on militaries’ connections to a robust industrial base in the GPT.
To control for competing explanations of how GPTs affect the military balance of power, we compare military electrification to the impact of the submarine on military affairs. These two technologies differ with respect to whether they qualify as a GPT but are similar across most other relevant features.Footnote 36 The submarine was not a GPT. While it generated some technological complementarities with advances in weapon systems and underwater propulsion, submarine technology did not have a wide variety and range of uses and had limited potential for continual innovation. Still, like electricity, the submarine was a disruptive innovation that changed the conduct of warfare.Footnote 37 Moreover, the introduction of submarines into military affairs occurred in the same period as the introduction of electricity into military affairs.Footnote 38 If we find that the two technologies affected the military balance of power in different ways, then these differences cannot be accounted for by technology-agnostic factors, such as the organisational competencies and cultures of the militaries in question, or time-dependent factors, such as the distribution of military power and the nature of military competition at the time. These overlapping trajectories help pinpoint the technological dimension of GPTs.
The electrification of warfare: background
Much of our empirical analysis relies on primary sources, supplemented by historical work on electricity and military modernisation. These include trade journals like The Electrician and The Electrical Review, which discuss military electrification from the perspective of engineers. Archival evidence, such as military service appeal tribunal records, help evaluate the extent of military dependence on civilian electric power systems. Lastly, we benefit from writings and speeches by military innovators involved with incorporating electricity into military operations.
Our analysis mainly covers the period from the mid-1800s to the end of the First World War. In the late nineteenth century, a cluster of electrical inventions – including the first practicable dynamos, the transformer, and the steam turbine – enabled the widespread application of electric power. This versatile energy system transformed systems of lighting, manufacturing, transportation, and electronic devices. By tracing the evolution of military electrification, we show that the versatility of electrical applications in the economic realm extended to the military domain (Figure 2).Footnote 39 Concretely, we evaluate whether military electrification was consistent with the three theorised features of GMTs.
The military electrification GPT tree (early applications).
GMT feature 1: Broad impact pathway
The versatility of military electrification eclipsed the constricted predictions of contemporary observers. As advances in electricity were emerging, experts and popular commentators alike envisioned that their main impact on military effectiveness would manifest through a narrow pathway: war-winning weapons.Footnote 40 Engines would deliver electric shocks ‘of infinite variety’ on the battlefield, one publication hypothesised in 1889;Footnote 41 electric rays of destruction would work ‘revolutionary effects on the art of modern warfare’, predicted another in 1896;Footnote 42 the Gatling gun's inventor claimed that a powerful electrical weapon would bring peace to the world.Footnote 43 A 1911 edition of Technical World magazine painted a particularly vivid picture of what electric-powered warfare would look like in 1950:
The old War God hurling his thunderbolts will seem impotent beside man wielding the forces of nature for weapons. Magazines exploded without warning by darting, invisible, all-penetrating currents of electricity … Guns, explosives, and projectiles will sink into the past, even as have the bow and arrow, giving place to howling elements clashing under man's direction.Footnote 44
The electric rays of destruction never materialised. Instead, electrification transformed militaries through many subtle, diffuse channels. The main applications spanned lighting (e.g., searchlights that helped to guard harbors and forts against surprise night attacks),Footnote 45 power (e.g., electric firing of guns),Footnote 46 and communications (e.g., telephones and wireless telegraphy).Footnote 47 As Carolyn Marvin argues, ‘Actual as opposed to fantasy developments in electrical warfare were mostly in the realm of communications rather than destructive weaponry.’Footnote 48
Indeed, electrical communications became a new arena for conflict. Described by Stephen Peter Rosen as ‘one of the most significant military technological innovations of the modern era’, electronic warfare was first demonstrated in the Russo-Japanese conflict of 1904.Footnote 49 During the First World War, British radio intelligence was effective in tracking German communications at sea. Many assessed this capability as ‘the most important single factor in the defeat of the U-boats in 1914–1918.’Footnote 50 Arguably, the most powerful demonstration of electricity's mark on the First World War was the British penetration of German diplomatic codes, which facilitated the leakage of the Zimmerman telegram.Footnote 51
Eventually, electrical applications in communications, lighting, transportation, and weapons control converged in complete electrical systems. On naval ships, electrical systems incorporated dynamos that supplied energy, searchlights that guarded the ship from surprise attacks, electric-powered velocimeters that calculated the positions of enemy ships, and electric technology that transmitted the commander's communications.Footnote 52 To underscore this point, electricity was central to the dreadnought, a new type of battleship often deemed the key military innovation of this time period.Footnote 53 Electrical communications and range finders supported centralised fire control, which facilitated long-range shooting with heavy gunnery – the crucial advance from pre-dreadnought ships to dreadnoughts.Footnote 54 Wireless telegraphy also supported coordination between dreadnoughts and the overall battlefleet.Footnote 55
Ultimately, we should extend some grace to those who foretold of electrical weapons of mass destruction. After all, the broad impact pathway of electricity made it difficult to envisage how it would shape military power. An American ordnance engineer, a frontline user of electrical applications, put it best in an 1892 article, ‘Great as was the usefulness of electricity during the period of the Civil War … there was probably no one at that era whose imagination was sufficiently elastic to dream of electricity ever acquiring the compass it possesses at the present time.’Footnote 56
GMT feature 2: Industrial productivity spillovers
In addition to spurring a variety of military innovations, a GPT should also influence military effectiveness by boosting industrial productivity. It is well documented that electrification resulted in a productivity surge in the US and other advanced economies.Footnote 57 Crucially, electrification enabled mass production, as the adoption of electric unit drive in factories resulted in standardised workflows and plant capacity expansion.Footnote 58 By the First World War, the ‘capacity of civilian firms to manufacture large numbers of standardized weapons became increasingly central to the conduct of industrialized warfare.’Footnote 59 Recognising that the US needed to realise ‘the greatest possible production of needed war materials of the kind peculiarly dependent upon a cheap and dependable supply of electricity’, the War Industries Board restricted civilian uses of electricity in key industrial centres.Footnote 60
This connection between the electrification of manufacturing and military production only intensified in the Second World War. Electricity was one of the highest targeting priorities for Allied strategic bombing efforts because of its impact upon a wide range of German industrial activities.Footnote 61 Historians attribute the outcome of the war to a large extent on the Allied capabilities in mass production.Footnote 62 American industry produced more than 250,000 planes during the Second World War, which exceeded the output of Britain and Germany combined.Footnote 63 Vaclav Smil concludes, ‘the rapid mobilization of America's economic might, which was energized by a 46% increase in the total use of fuels and primary electricity between 1939 and 1944, was instrumental in winning the war against Japan and Germany.’Footnote 64
GMT feature 3: Delayed effects
In the economic realm, scholars hold up the diffusion of electricity as an example of the lag between the emergence of a GPT and its impact on national productivity. Measured by percentage of total installed horsepower in manufacturing industries, adoption of unit drive, and estimates of electricity's contributions to GDP growth, American electrification did not take off until the 1920s.Footnote 65 This was a full four decades after major advances like the dynamo and incandescent light bulb.
Do we find a similarly delayed timeline for military electrification? Though it is more difficult to track the effects of electricity on military effectiveness, we can trace how and when complementary innovations in different military branches were first introduced as military capabilities, first used in warfare, and fully adopted as a standard military capability (Table 1).Footnote 66 Combined with the previous technology tree, this mapping exercise of two electrical military innovations shows that even early movers did not achieve widespread adoption of key innovations until right before the First World War. Among later adopters, widespread adoption did not take place until the interwar period or after the Second World War. In fact, some of the most significant military applications of electricity, including radars, did not emerge until the 1940s.Footnote 67
Delayed impact of electrical military innovations.
It took time for militaries to upgrade their skill base and make organisational adjustments for electrification. Due to limitations in their ability to train and attract electrical engineers, even leading militaries relied on volunteer technical reserves to maintain and develop their electrical infrastructure through the First World War.Footnote 68 Like manufacturers reluctant to overhaul their factory layouts to optimise electricity usage, militaries were slow to adopt electrical applications that demanded structural changes. For instance, although the opportunity existed as early as 1899, the US Navy did not fully integrate radio communications until 15 years later, due to opposition by senior naval officers who saw the radio as a direct threat to their authority onboard ships.Footnote 69
Comparison to submarines as a non-GPT
Not all technologies interact with military systems in the same way. Along all three features of GMTs, the evolution of submarines in military affairs differed from military electrification.Footnote 70 First, advances in underwater submersion technology affected military effectiveness through a much narrower pathway, largely in the application of submarines as weapons platforms.Footnote 71 During the late nineteenth century and early twentieth century, underwater submersion technology had very few civilian applications, resulting in limited effects on industrial productivity. Lastly, there was a relatively short delay between the introduction of the first modern submarines, which occurred around the turn of the twentieth century, and their impact on military effectiveness, which was apparent in the years before the First World War.Footnote 72
Electrification gaps across militaries
The pervasive applicability of GPTs across many sectors, combined with the fact that the civilian realm offers a broader range of applications than the military realm, means that the military will grow more dependent on the civilian sector as GPTs develop. By comparison, military applications of some dual-use technologies also rely on the civilian sector, but the civilian and military development trajectories of many dual-use technologies, such as aircraft parts, now greatly diverge.Footnote 73 Industrial dependency, therefore, should prove especially relevant for GPTs.
This was the case with electricity. The amount of money invested in research and development within electrical engineering departments and electric companies across the United States was ‘many times greater than that invested in all [the US's] ships put together.’Footnote 74 It was difficult for the military to retain electricians because of the breadth of opportunities in private industry. In the years before the First World War, an electrician-sergeant in the US Army earned about $40 per month; a civilian electrician with similar competencies earned about $100 per month.Footnote 75
One of the clearest demonstrations of military electrification's dependence on the civilian electric industry is the US military's switch from direct current (DC) to alternating current (AC) standards in coastal defence fortifications. Prior to 1898, all US military electrification projects preferred the use of DC, even though civilian industry was transitioning to AC systems.Footnote 76 Over time, as expenses of maintaining separate DC power systems for military posts accumulated, US policy adjusted ‘towards a goal of accommodation with civilian commercial utilities’.Footnote 77 Ultimately, in adapting AC standards, the US recognised that widespread military electrification would only be possible if the army permitted the integration of AC systems, which were dominant in commercial electrification.Footnote 78
To test our proposition about the non-levelling effects of GPTs, we evaluate whether industrial dependency accounted for differences between the British and Russian militaries in their adoption of electrical military innovations. The British demonstrated an effective system of wireless military communications and electric fire control in the First World War.Footnote 79 At the time, Russia was a first rank power and perceived to be growing rapidly in military strength. However, its military was a technological laggard with respect to electrification. According to one comparison of the signal communication systems of the British and Russian militaries at the onset of the First World War, Britain and Russia were at complete opposite ends of the spectrum.Footnote 80 If GMT theory holds, we should observe that the British military was more connected to a robust industrial electricity sector than the Russian military. To further isolate the effects of industrial dependency, we also examine developments in submarine technology, a non-GPT.
Compared to its Russian competitor, Britain's military could draw from a more robust civilian electricity sector. Measures of electricity production per capita from 1896 to 1940 depict the substantial gap in civilian electrification between these two countries (Figure 3). As earlier evidence about the US experience showed, militaries needed to tap into civilian talent pools to adopt electrical advances. Accordingly, Britain's employment of signals intelligence during the First World War, ‘used the best technology and science and electrical engineers of the day’.Footnote 81 Archival evidence from military service appeal tribunals during the First World War provide a more granular picture of British industrial dependency. Of those that appealed against conscription into the army based on the grounds that it was more conducive to national interests that they continue to engage in work, electrical occupations occupied a far larger share than other technologically progressive fields, including chemicals, steel, and submarines.Footnote 82
UK-Russia comparison of electrical industrial base.
On the other hand, Russia was unable to electrify its military. This failure came to the fore at the Battle of Tannenberg between Russian and German forces at the onset of the First World War, to dramatic effect. The Russian Army suffered a massive defeat, and 100,000 men were captured. According to George I. Back, who served as the chief signal officer of the Mediterranean theater for the US Army, and George Thompson, a military historian, Russia's setback at Tannenberg ‘was largely due to an almost total lack of signal communication’.Footnote 83 Because the Russian military lacked both the electrical equipment and the requisite technical knowledge base for using these devices and encrypting electrical communications, the Germans had access to detailed Russian communications and marching orders.Footnote 84
One key factor behind the Russian military's relative failure with adopting electricity was its weakly developed industrial base in electricity. The lack of a unified technical profession and skilled personnel functioned as bottlenecks on Russian military electrification.Footnote 85 With no leading electrical companies of their own, Russia's electrical engineering sector depended on foreign suppliers for critical components such as high voltage transformers and measuring instruments.Footnote 86 When imports were cut off amidst war, Russian industry could not independently manufacture this equipment, even with support from the Russian war industry committee.Footnote 87
If civilian dependency holds, then even a strong, technologically savvy military cannot bring about a GMT on its own. Before the First World War, the Russian military tried to promote electrification by whatever means necessary. It sponsored research and travel, trained scientists, subsidised domestic industry, gathered information about the latest developments in electric technology, and provided testbeds for materials and systems.Footnote 88 While the Russian military did achieve success in some limited, early applications of electricity, this success did not extend to overall electrification. Russian electric mine technology, for instance, was relatively advanced. Russian electric mines damaged two British ships in the Crimean War, and the Japanese sent a delegation to the Russian Mine School in 1877 to gain knowledge from Russian experts. However, this competent pool of workers was ‘one of the few such groups in Russia.’Footnote 89 Unfortunately for the Russian military, taking full advantage of a GPT necessitates much more than skill competencies in a few application sectors; it requires access to an industrial base that can facilitate widespread GPT adoption.
What about other factors besides industrial dependency? For instance, Russia's general economic underdevelopment vis-à-vis Britain might explain this military electrification gap. Other considerations include differences in the two militaries’ organisational and cultural attitudes towards emerging technologies. To control for alternative explanations, we examine trends in submarine technologies, which do not fulfill the characteristics of a GPT. If Russia was able to keep pace with Britain adopting submarines, then it is less likely that general economic, organisational, and cultural factors were driving differences in military electrification.
Indeed, the Russian Navy was closer to parity with the British in terms of submarine capabilities than military electrification. Russia quickly realised the strategic potential of submarines as new combat platforms, and Russian submarines played an effective role during the First World War. Before the outbreak of war, the Russian Navy was equipped with 18 diesel-electric submarines,Footnote 90 of which the Bars and Morzh class submarines were on par with the best foreign counterparts. In the Baltic Sea, for example, British submarines sank 15 merchant boats in total, while the Russian Bars class submarines sank eight in 1916–17. Footnote 91 Russian submarines also delivered results in the Black Sea, sinking and capturing 25 ships.Footnote 92
Unlike Russia's experience with military electrification, the Russian Navy could achieve isolated areas of technical competence in submarine capabilities. For instance, Russia launched the world's first submarine minelayer, the Krab, in 1912. Underwater minelaying was successful enough at restricting Turkish naval operations in the Black Sea to justify converting other submarines into minelayers.Footnote 93 Overall, the gap between Russian and British submarine capabilities was smaller than the corresponding disparity in military electrification.Footnote 94
Since submarine technology was not a GPT, a military's connection to a robust industrial base should be less significant for explaining which militaries are differentially advantaged by submarines. Like Russia's electrical base, Russia's civilian shipbuilding industry was weak, but this weakness was not as significant in determining the military effectiveness of Russian submarine capabilities.Footnote 95 Developing effective submarine capabilities required a relatively narrow talent base skilled in operating high-speed reciprocating machinery.Footnote 96 Russian submarine engineers were experienced, knowledgeable, and recognised for their skill even when the boats they were commanding lacked in quality.Footnote 97 Russia struggled more with electrification because of the demand for general upskilling drawing on a broad industrial base connected to electrical advances.Footnote 98
Conclusion and implications for AI
In sum, the evidence from the electricity case is consistent with GMT theory. Like its economic impact pathway, electricity found widespread military applications only after a protracted period of gestation. In addition to directly boosting military effectiveness through a broad array of military innovations, electricity also indirectly transformed military power through stimulating industrial productivity. Recognising the GMT associated with electricity is a prerequisite to evaluating how electricity shaped the balance of military power. Instead of levelling the military playing field, the electrification GMT advantaged militaries that were able to tap into a robust electrical industrial base.
Recognising the limitations of a single case, this article provides a promising building block in the study of GMTs. By uncovering how militaries adopted electricity, it provides the first comprehensive explanation of how GPTs affect military power. This fills a large gap in the military innovation literature, which has not adequately explored some of the most significant technological advances throughout history. To further flesh out our article's contribution, we conclude by mapping out how GMT theory challenges current discussions of how military affairs will be transformed by AI – often dubbed the ‘next GPT’.Footnote 99
To begin, it is necessary specify scope conditions when extending the implications of the electricity case to other possible GMTs. While AI and electricity are both GPTs, they differ along many other relevant characteristics. Autonomy, for instance, is a distinctive characteristic of some AI systems. Whether the lessons from history about the effects of electricity on military affairs can apply to current developments in AI depends not just on whether the technical properties of AI are comparable to those of electricity but also on the congruence between the nature of military competition and dynamics of the international system in the current period and those of the late nineteenth century. Still, our analysis can provide an initial guide for comprehending the impact of AI on military affairs, akin to how studies of electricity's effect on economic transformation framed scholarship on the impact of computers on productivity.Footnote 100
First, speculation about how AI will transform military affairs places excessive emphasis on the narrow effects of weapon systems. Possibly influenced by popular images of killer robots, both policymakers and scholars focus on autonomous weapons as the primary military application of AI. US defense intellectuals highlight how China could take advantage of AI-enabled hypersonic missile systems to leapfrog US military power.Footnote 101 In its approach to AI, the Chinese military also tends to prioritise ‘trump card’ or ‘assassin's mace’ weapons that can counter US capabilities.Footnote 102
In contrast, a GMT approach emphasises the accumulation of AI-enabled improvements across many military systems. This impact pathway will likely interact with weapons capabilities, as was the case with electricity and fire control. Overall, though, effects of AI advances will be more consequential in other military domains, including communications, decision support, intelligence, and logistics.Footnote 103 Moreover, the focus on AI weapons neglects the indirect effects of AI's potential to upgrade a nation's productive capabilities. AI applications that improve the efficiency and adaptability of manufacturing lines could have significant follow-on effects for military readiness.
Second, existing conjectures about the impact of AI on military affairs severely underestimate the timeframe for when substantial effects will occur. Recent influential articles on AI and national security converge on the next ten to twenty years as the timeframe for when AI will substantially transform military power.Footnote 104 This is reflective of a broader tendency to conflate rapid progress in a technological field, which is characteristic of GPTs, with rapid adoption across military applications, which is uncharacteristic of GPTs.
GMT theory suggests a different view. Economists have already begun to model implementation lags in the effects of AI on economic productivity.Footnote 105 A similar extended trajectory will apply in the military realm. The current wave of AI development started with breakthroughs in deep learning in the early 2010s, so if AI follows the same timeline as electricity, a prolonged period of gestation could extend until around the 2050s.Footnote 106 In addition, since the development of AI is still in its early stages, the foreseeability of its military applications is very limited. Twenty years after the introduction of the electricity dynamo, even the most astute observers of military transformation could not envision how that technology would transform military affairs. As only a decade has passed since critical breakthroughs in deep learning, any attempt to foreordain the ultimate military implications of AI should be met with deep scepticism. Our imaginations – to borrow language from the ordnance engineer quoted earlier – are not sufficiently elastic.
Lastly, GMT theory supplements existing thinking about international diffusion of military applications of AI and the effect of AI on the military balance of power. Some scholars argue that if military advances in AI continue to be closely linked to civilian applications, then military AI capabilities will rapidly diffuse to other countries.Footnote 107 Informed by a historical perspective of GMTs, we view ‘military AI technology’ as not a singular technological innovation but part of a GPT trajectory, which encompasses a broad distribution of technological applications. Just like the organisational requirements for adopting wireless telegraphy were different from those required to adopt searchlights, the adoption capacity for different military applications of AI will vary.
To more fully account for how AI advances will differentially advantage certain militaries, more attention should go to factors that apply across the broad front of a GPT trajectory. We highlight the significance of a state's industrial capacity to provide AI infrastructure and skilled labour to militaries. Specifically, militaries able to draw from a wide skill base in AI will better exploit the AI-based GMT. Crucially, the talent base required for AI differs from the talent base required for other dual-use technologies like nuclear power. GMT theory suggests that military linkages to a wide base of AI engineering talent, rather than star researchers or cutting-edge technical capabilities, are crucial to adapting generalised models to a variety of specific military applications.Footnote 108 If leading militaries have stronger connections to their civilian GPT sector than challengers, then GMTs could reinforce existing military balances.Footnote 109
The three great influences on naval warfare of the twentieth century were the aircraft, the submarine, and electricity. We have shown that electricity is not like the others. It powered a GMT. In parallel characterisation, Horowitz identifies three key technologies that could reshape the future of warfare in the twenty-first century: cyber, drones, and AI.Footnote 110 As plausibly the defining GPT of our century, AI is not like the others. After all, it's the new electricity.
Acknowledgements
For helpful comments and input, we thank: Markus Anderljung, Carolyn Ashurst, Miles Brundage Ben Buchanan, Max Daniel, Ben Garfinkel, Luke Muehlhauser, Carina Prunkl, Toby Shevlane, Duncan Snidal, Waqar Zaidi, seminar audiences at the University of Oxford and Stanford University, the anonymous reviewers, and especially Alexis Carlier and Joslyn Trager. Jeffrey is an Assistant Professor of Political Science at George Washington University. This work was supported by a postdoctoral fellowship at Stanford's Center for International Security and Cooperation, sponsored by Stanford's Institute for Human-Centered Artificial Intelligence (AI). Allan is Senior Staff Research Scientist and Head of Long-term Strategy and Governance at DeepMind, and founder and president of the Centre for the Governance of AI.
Supplementary material
To view the online supplementary material, please visit: https://doi.org/10.1017/eis.2023.1
Jeffrey Ding is an Assistant Professor of Political Science at George Washington University.
Allan Dafoe is a Senior Staff Research Scientist at DeepMind and President of the Centre for the Governance of AI.
Open access
