蘭德：人工智能時代的戰略競爭人工智能軍事應用的新興風險和機遇（英文版）（144頁）.pdf

《蘭德：人工智能時代的戰略競爭人工智能軍事應用的新興風險和機遇（英文版）（144頁）.pdf》由會員分享，可在線閱讀，更多相關《蘭德：人工智能時代的戰略競爭人工智能軍事應用的新興風險和機遇（英文版）（144頁）.pdf（144頁珍藏版）》請在三個皮匠報告上搜索。

1、Strategic competition in the age of AIEmerging risks and opportunities from military use of artificial intelligence James Black,Mattias Eken,Jacob Parakilas,Stuart Dee,Conlan Ellis,Kiran Suman-Chauhan,Ryan Bain,Harper Fine,Maria Chiara Aquilino,Mlusine Lebret,Ondrej PalickaFor more information on th

2、is publication,visit www.rand.org/t/RRA3295-1About RAND Europe RAND Europe is a not-for-profit research organisation that helps improve policy and decision making through research and analysis.To learn more about RAND Europe,visit www.randeurope.org.Research Integrity Our mission to help improve pol

3、icy and decision making through research and analysis is enabled through our core values of quality and objectivity and our unwavering commitment to the highest level of integrity and ethical behaviour.To help ensure our research and analysis are rigorous,objective,and nonpartisan,we subject our res

4、earch publications to a robust and exacting quality-assurance process;avoid both the appearance and reality of financial and other conflicts of interest through staff training,project screening,and a policy of mandatory disclosure;and pursue transparency in our research engagements through our commi

5、tment to the open publication of our research findings and recommendations,disclosure of the source of funding of published research,and policies to ensure intellectual independence.For more information,visit www.rand.org/about/principles.RANDs publications do not necessarily reflect the opinions of

6、 its research clients and sponsors.Published by the RAND Corporation,Santa Monica,Calif.,and Cambridge,UK 2024 RAND CorporationR is a registered trademark.Cover:Adobe StockLimited Print and Electronic Distribution Rights This publication and trademark(s)contained herein are protected by law.This rep

7、resentation of RAND intellectual property is provided for noncommercial use only.Unauthorised posting of this publication online is prohibited;linking directly to its webpage on rand.org is encouraged.Permission is required from RAND to reproduce,or reuse in another form,any of its research products

8、 for commercial purposes.For information on reprint and reuse permissions,please visit www.rand.org/pubs/permissions.iBackground to this studyArtificial intelligence(AI)holds the potential to usher in transformative changes across all aspects of society,economy,and policy,including defence and secur

9、ity.The UK aspires to be a leading player in the rollout of AI for civil and commercial applications,and in the responsible development of defence AI.This necessitates a clear and nuanced understanding of the emerging risks and opportunities associated with the military use of AI,as well as how the

10、UK can best work with others to mitigate or exploit these.In March 2024,the Defence AI&Autonomy Unit(DAU)of the UK Ministry of Defence(MOD),and the Foreign,Commonwealth and Development Office(FCDO)jointly commissioned a short scoping study from RAND Europe.The goal was to provide an initial explorat

11、ion of ways in which military use of AI might generate risks and opportunities at the strategic level conscious that much of the research to date has focused on the tactical level or on non-military topics(e.g.AI safety).Follow-on work will then explore these issues in more detail to inform the UK s

12、trategy for international engagement on these issues.This technical report aims to set a baseline of understanding of strategic risks and opportunities emerging from military use of AI.A standalone summary report focuses on high-level findings for decision makers.About RANDThis study was conducted b

13、y a mix of RAND staff on both sides of the Atlantic.With offices in the UK,Belgium,and the Netherlands,RAND Europe is the European arm of RAND,a non-profit research institute and the largest policy research organisation in the world.RANDs mission is to help improve public policy and decision making

14、through objective research and analysis,having delivered over 75 years of classified and unclassified studies for UK,US and other allied governments.RAND has been involved in research into the military and strategic implications of AI since the 1950s,having also played a vital role in developing gam

15、e theory,deterrence theory,and nuclear strategy.For more information on the study,this report or RAND,please contact:James Black Assistant Director Defence and Security Research Group RAND Europe e.jblackrandeurope.orgPrefaceiiStrategic competition in the age of AI:Emerging risks and opportunities f

16、rom military use of artificial intelligenceThe advent of AI is ushering in profound changes to competition and conflictAI is best understood as a dual-use set of general-purpose technologies,hardware-enabled but software-based.Unlike traditional military technologies,they are highly democratised and

17、 proliferating fast.Innovation is driven by the private sector for commercial uses,not by government or defence.Collective understanding of military applications and implications is improving,but from a low base.Too often,debate prioritises certain high-profile issues e.g.lethal autonomous weapon sy

18、stems(LAWS)or artificial general intelligence(AGI)at the expense of other topics.It focuses on the tactical at the expense of the strategic;risks at the expense of opportunities;or the immediate consequences of military AI at the expense of the second-and third-order effects that might be most impac

19、tful in the long run.To address this,the MOD and FCDO commissioned this study to develop a conceptual framework mapping the strategic risks and opportunities arising from military AI.Figure 0.1 Framework:strategic risks and opportunities of military use of AISource:RAND Europe analysis(2024).Summary

20、Net results of AI-related risks and opportunities at the global levelRisks and opportunities for each actor(e.g.UK):Risks and opportunities of AI across the full continuum of cooperation,competition and conflictE.g.alliances,resilience,sub-threshold,deterrence,crisis,conventional warfighting,nuclear

21、,de-escalation,peace.International systemIntensity and dynamics of strategic competitionStrategic stability and escalation ladder.SocietyEconomyDefence enterpriseMilitary capability(e.g.C4ISTAR,fires,mass,logistics,training and exercises,MDI/MDO etc.).AI affects POTENTIAL for advantage(i.e.raw poten

22、tial in terms of net accounting of strengths/weaknesses)Impact on:Impact on:AI affects PROPENSITY for advantage(i.e.ability to translate potential into performance by marshalling resources in pursuit of strategic objectives)Superpowers verses othersDemocracy verses authoritarian Non-state(incl.AI fi

23、rms)(e.g.How can a medium power such as UK navigate intensified superpower competition in an age of AI?).Choices differ by actor type:Strategy makingOrchestrationImplementation(across all instruments of power e.g.DIME levers)Signalling/perceptions.Impact on:iiiAI poses complex,significant and undera

24、ppreciated risks to defence and securityOf the many risks and opportunities explored in detail in this report,the most pressing include:Information manipulation,such as AI deepfakes,which could not only drive political,economic and social problems but also skew military decision making in times of c

25、risis.Empowerment of non-state actors with asymmetric capabilities that challenge the dominance of state militaries or,in the worst-case scenario,new tools of mass destruction(e.g.bioweapons).The interlinked impacts of AI on the offencedefence balance between adversaries,on escalation dynamics towar

26、ds warfighting,and on the stability of nuclear deterrence.These issues are especially concerning amidst intensifying superpower rivalries and in a world already grappling with other drivers of insecurity(e.g.Ukraine,IsraelIran,Taiwan,migration,climate change).The potential catastrophic safety and se

27、curity risks associated with any future advent of AGI.Table 0.1 Priority risks and opportunities for actionSELECTED PRIORITY RISKS AND OPPORTUNITIESINITIAL ASSESSMENT OF IMPACTFramework categoryIssueSignificanti.e.potential to disadvantage in sub-thresholdSeverei.e.potential to disadvantage in conve

28、ntional warCatastrophici.e.potential for catastrophe or existential threatNationalEconomic disruption and warfareInformation-manipulation(e.g.deepfakes)Changes to defence productivity,mass and lethalityInternationalBy actor typeErosion of RBIO and governance institutionsAI-enabled repression(and exp

29、ort thereof)Empowerment of non-state actors(e.g.bioweapons)By conflict typeChanges to military offence-defence balanceImpact on escalation dynamicsImpact on nuclearMacro-trendsProspects for AGI and non-alignmentSource:RAND Europe analysis(2024).ivStrategic competition in the age of AI:Emerging risks

30、 and opportunities from military use of artificial intelligenceAt home,there are also major issues to contend with in terms of disruptive impacts on domestic politics and the economy.These shape the ends and means available to Defence.Abroad,AI could similarly have profound implications for the heal

31、th of the rules-based international order,depending on whether and how effectively nations,industry and civil society work together to manage its effects.There is significant concern among AI experts about the extent to which AI could tip the balance in favour of repressive and authoritarian modes o

32、f governance in many parts of the world,while simultaneously threatening to subvert democratic politics,pollute the information environment and undermine societies will-to-fight.Equally,leadership on military AI could deliver outsized benefitsMany of these potential risks could also manifest as oppo

33、rtunities for strategic advantage.The balance of pros and cons from the rollout of AI hinges on how quickly and effectively nations are able to adapt institutions such as their Armed Forces to exploit AIs benefits.Similarly,it depends on how well governments can exert influence internationally to sh

34、ape global behaviours on military AI in a direction that suits their interests and values.This means being willing to deliver the significant investments,organisational reforms and cultural changes needed to transform Defences approach to new technologies.Urgent action is needed to mitigate emerging

35、 risks and exploit opportunitiesTo address these challenges,nations must urgently develop a comprehensive action plan that considers the complex interplay of technical advances in AI,geopolitical competition over and through AI,and evolving norms around AI in the international system.This should dra

36、w upon a toolkit of mechanisms to influence different audiences,employing all diplomatic,information,military and economic(DIME)levers to bring together a proactive set of:Efforts to boost the responsible uptake of AI and maximise its benefits to Defence.Efforts to limit the adoption of military AI

37、by non-state and terrorist actors,or hostile/rogue states,while also imposing costs on them to influence their actions.Efforts to shape global,minilateral and bilateral governance arrangements for military AI.vTable 0.2 Toolkit of mechanisms for shaping global defence AI developmentsCATEGORY OF TOOL

38、KITPRIORITY ACTIONSMechanisms to boost AI adoption and benefits for UK DefenceAccelerate investment in and adoption of AI across Defence,while increasing resilience against hostile or accidental misuse of AIMechanisms to restrict AI adoption and benefits for adversariesAdopt a campaigning approach t

39、o restrict,slow,or increase the costs to adversaries(state or non-state)of deploying military AIMechanisms to shape emerging governance arrangements for military AIPlay a leading role in awareness raising,problem finding,and sharing learning about military AI risksDevelop transparency and confidence

40、 building measures with key allies(e.g.US)and competitors(e.g.China)to reduce escalation risksPromote an inclusive,participatory approach to build an emerging global consensus on norms of responsible behaviour around military AI,as a prelude to more robust binding agreements in futurePromote paralle

41、l development of minilateral mechanisms for reducing urgent nuclear-and bio-related AI risksInvestigate ways to incorporate AI into verification and compliance mechanisms,and vice versaOver time,consolidate the current fragmented landscape of AI governance initiatives into a more concrete architectu

42、reSource:RAND Europe analysis.This should also build on lessons from other domains as examined in this report and the momentum of recent high-level initiatives on AI.Prominent examples include the Bletchley Summit,the Responsible AI in the Military Domain(REAIM)summit,and the Political Declaration o

43、n Military AI.viStrategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligencePreface iSummary iiTables viiiFigures ixBoxes xAbbreviations xiAcknowledgements xiiiChapter 1.Introduction 11.1.Research scope,objectives and methodology 1Chapter 2.Tow

44、ards a framework for strategic impacts from military AI 32.1.Problems with current understanding of the impacts of military AI 42.2.Contours of a possible conceptual framework 72.3.Summary 11Chapter 3.Impact at the national level 123.1.Understanding strategic advantage 133.2.AI impact:potential for

45、advantage 143.3.AI impact:propensity for advantage 213.4.Summary 30Chapter 4.Impact at the international level 334.1.Understanding the international system 344.2.AI impact:actors,goals,and power 344.3.AI impact:global governance 364.4.AI impact:strategic competition dynamics 374.5.Summary 42Chapter

46、5.Implications by competition type 435.1.Implications for alliances and partnerships 445.2.Implications for Defence engagement and capacity building 455.3.Implications for resilience and emergency preparedness 465.4.Implications for sub-threshold operations 475.5.Implications for deterrence 485.6.Im

47、plications for crisis management 525.7.Implications for conventional warfighting 525.8.Implications for nuclear warfighting 54Table of contentsvii5.9.Implications for de-escalation,peacebuilding and reconstruction 555.10 Summary 56Chapter 6.Implications by actor type 576.1.Implications for different

48、 types of state 586.2.Implications for different systems of government 666.3.Implications for non-state actors 686.4.Summary 71Chapter 7.Priority issues 737.1.Towards a prioritisation of strategic risks and opportunities 73Chapter 8.Lessons from other domains 778.1.Existing models of risk management

49、 788.2.Transferrable learning 82Chapter 9.Toolkit of measures to exert influence 899.1.Mapping risks and opportunities against the toolkit 909.2.Mechanisms to boost AI adoption and benefits for Defence 909.3.Mechanisms to restrict AI adoption and benefits for non-state and terrorist actors,and hosti

50、le and rogue states 959.4.Mechanisms to shape and influence governance arrangements 989.5.Summary 104Chapter 10.Conclusion and next steps 105References 107Annex A.Methodology 123A.1.Research approach 123A.2.Data collection methods 123Annex B.List of interviews 126viiiStrategic competition in the age

51、 of AI:Emerging risks and opportunities from military use of artificial intelligenceTable 0.1 Priority risks and opportunities for action by the UK iiiTable 0.2 Toolkit of mechanisms for shaping global defence AI developments vTable 3.1 National level:AI impacts on society 15Table 3.2 National level

52、:AI impacts on the economy 17Table 3.3 National level:AI impacts on the Defence enterprise 18Table 3.4 National level:AI impacts on military capability 19Table 3.5 National level:AI impacts on strategy implementation using DIME levers 26Table 4.1 System change:Impacts,risks and opportunities from AI

53、 35Table 4.2 Systemic change:Impacts,risks and opportunities from AI 36Table 4.3 Impact of AI on intensity of strategic competition 39Table 4.4 Interactions change:Impacts,risks and opportunities from AI 41Table 6.1 Superpowers:Impacts from military AI 63Table 6.2 Middle powers:Impacts from military

54、 AI 64Table 6.3 Small states:Impacts from military AI 66Table 7.1 Prioritising risks and opportunities for action 75Table 8.1 Potential models from other domains and sectors 78Table 8.2 Potential transferrable lessons from other domains and sectors 83Table 9.1 Mapping of priority issues for governme

55、nts against the toolkit 91Table A1.1 Workshops or webinars incorporated into RAND study 125Table A2.1 List of interviews 126TablesixFigure 0.1 Framework:strategic risks and opportunities of military use of AI iiFigure 2.1 Depiction of overlaps between AI,ML and data science 4Figure 2.2 Conceptualisi

56、ng the AI lifecycle from development to adoption to impact 9Figure 2.3 Framework:strategic risks and opportunities of military use of AI 10Figure 3.1 Relationship between an actors potential and propensity for strategic advantage 14Figure 3.2 Strategic signalling and(mis)perception with AI through a

57、 game-theoretic lens 28Figure 3.3 Example of virtuous or vicious cycles emerging from strategic impacts of military AI 31Figure 4.1 Feedback loops across the continuum of cooperation,competition and conflict 38Figure 5.1 Sub-threshold operations in the grey zone of competition and conflict 47Figure

58、5.2 Examples of impact from AI on classical deterrence theory 51Figure 6.1 Global AI Index 2024 59Figure 6.2 Factors that contribute to the stability or instability of a superpower rivalry 62Figure 8.1 Wicked problem of global governance of military AI 85Figure 8.2 Toolkit:mechanisms to shape risks

59、and opportunities 88FiguresxStrategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligenceBox 2.1 Summary of findings:Chapter 2 3Box 2.2 Definition of AI in the UK Defence AI Strategy 4Box 2.3 Design considerations for a possible conceptual frame

60、work 8Box 3.1 Summary of findings:Chapter 3 12Box 3.2 Definition of strategic advantage 13Box 4.1 Summary of findings:Chapter 4 33Box 5.1 Summary of findings:Chapter 5 43Box 6.1 Summary of findings:Chapter 6 57Box 8.1 Summary of findings:Chapter 8 77Box 9.1 Summary of findings:Chapter 9 89BoxesxiAbb

61、reviationsAIartificial intelligenceAIPfDAI Partnership for DefenceAGIartificial general intelligenceASIartificial super intelligenceAUKUSAustraliaUnited KingdomUnited StatesC2Command and ControlC4ISTARCommand,Control,Communications,Computers and Intelligence,Surveillance,Target Acquisition and Recon

62、naissance DAUDefence AI&Autonomy UnitDIMEDiplomatic,Information,Military,EconomicDoDUS Department of DefenseEUEuropean UnionFCDOForeign,Commonwealth and Development OfficeFVEYFive EyesGC-REAIMGlobal Commission on Responsible AI in the Military DomainGGEGroup of Government ExpertsGPTgeneral-purpose t

63、echnologyIHLinternational humanitarian lawLAWSlethal autonomous weapons systemsLLMlarge language modelLOACLaw of Armed ConflictMCFmilitary-civil fusionMLmachine learningMODMinistry of DefenceNATONorth Atlantic Treaty OrganizationxiiStrategic competition in the age of AI:Emerging risks and opportunit

64、ies from military use of artificial intelligenceNGOnon-governmental organisationPAGPartners Across GovernmentREAIMResponsible AI in the Military DomainTCBMtransparency-and confidence-building measuresTTCEUUS Trade and Technology CouncilUKUnited KingdomUNUnited NationsUSUnited StatesxiiiAcknowledgeme

65、nts This short study would not have been possible without contributions from several institutions and individuals.The authors are especially grateful to Mike Gibson,Sam Phillips and Oliver Mahoney of the DAU in the UK MOD,and to Anthony Stanley and Sian Evans of the FCDO,for their sponsorship of the

66、 research and their feedback throughout the study.Within RAND,thanks are owed to Kiera Addair from Knowledge Services for her support with the extensive literature review,to Jess Plumridge and Hannah Beelam for their graphic design on the final report;to Ben Plumridge and Abi Saffrey for their copy

67、editing;and to our quality assurance reviewers,Erik Silfversten and Luke Huxtable,for providing comments and feedback on drafts.Crucially,too,the RAND team are grateful to the insightful contributions made by the 54 government stakeholders and academic,think tank or industry experts who took part in

68、 research interviews within the tight timeframes for this study.Notwithstanding all these valued inputs,all errors or omissions in this draft report remain the sole responsibility of the authors.xivStrategic competition in the age of AI:Emerging risks and opportunities from military use of artificia

69、l intelligence1Chapter 1.Introduction 1.1.Research scope,objectives and methodology1.1.1.Adoption of AI by state militaries and armed non-state groups is ushering in significant changes to the character of competition and conflictDevelopment,integration and use of AI for military purposes could have

70、 profound implications for the future of warfare and for international peace and security more generally.This presents both opportunities and risks,whether at the tactical,operational or strategic level.Already,uncrewed robotic systems are being deployed in unprecedented numbers and with increasing

71、levels of autonomy across the battlespaces of Ukraine,IsraelGaza and the Red Sea.AI is similarly being integrated into intelligence analysis,command and control(C2),targeting,fires,training,simulation,equipment monitoring,and logistics.Such trends have driven debates over the impact and possible tra

72、jectories of algorithmic warfare,known as intelligentised warfare by Chinas Peoples Liberation Army(PLA).These discussions do not only concern the immediate military,ethical and legal impacts.There is also an increasing recognition of broader strategic implications.Examples include concerns about th

73、e effects of AI on state competition or conflict escalation,extending to the risk of nuclear warfare.1 Faced with a nascent understanding of these cascading implications,governments must 1 Kania(2019);Liu&Maas(2021).work with industry,academia and civil society to improve understanding of how to acc

74、entuate the benefits of military AI while reducing potential risks and hazards at the strategic level.For its part,the UK Government published a Defence AI Strategy in June 2022,with a focus on responsible development and deployment of AI in a defence context.This entails not only accelerating adopt

75、ion of AI by the MOD and Armed Forces,but also proactively shaping global AI developments to manage both benefits and risks.The UK thereby aims to play a leading role in shaping governance of this fast-moving technology by collaborating with allies and partners,engaging neutral countries and,where i

76、nterests align,working with adversaries to build a shared understanding of the possible strategic implications and how to manage AIs global impact.1.1.2.The MOD and FCDO commissioned RAND to build a framework for thinking about these emerging strategic risks and opportunities from military use of AI

77、Against this backdrop,RAND was asked to deliver an exploratory study into the potential impacts,both beneficial and detrimental,of military AI on the strategic level.Specifically,this one-month initial study aimed to identify components of a potential conceptual framework that could aid in comprehen

78、ding the strategic effects of AI to help facilitate an informed response from UK Government,including Defence.As discussed in Chapter 2,the UKs Defence AI Strategy defines AI as a collection of 2Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial i

79、ntelligencegeneral-purpose technologies,each of which has the potential to empower machines to execute tasks that would typically necessitate biological intelligence.Machines can also learn from large datasets how to perform these tasks,such as identifying patterns,acquiring knowledge from experienc

80、es or making predictions.Currently,the state-of-the-art in AI is advancing rapidly,with much uncertainty over its future directions.As such,any framework for conceptualising the risks and benefits of military AI needs to consider the wide range of current applications and possible future development

81、s.This does not mean making predictions about the most likely trajectories for technological progress,as these are likely to be wrong.Rather,it means mapping out different potential strategic risks and opportunities that could come to pass depending on how both technology and governance mechanisms e

82、volve.These should be grouped based on common features to enable a systematic approach to thinking about issues such as deterrence,strategic stability or proliferation.Armed with this mapping of possible impacts of military AI at the strategic level,this exploratory study was then tasked with high-l

83、evel identification of possible ways to influence these risks or opportunities.This included incorporating any insights emerging from other sectors(e.g.nuclear arms control,space,biotech)or the approaches and thinking of other countries around the world.1.1.3.The research team employed a multi-metho

84、d approach,combining literature review and interviews with government,industry,think tanks and academiaTo inform this initial exploratory study,the research team drew on:A narrative literature review of 200 academic or grey sources,derived from a long-list of 1,500.Semi-structured interviews with ov

85、er 50 stakeholders and experts from across government,UN,NATO,defence industry,AI firms,academia,think tanks and non-governmental organisations.Seven external workshops or webinars and two parliamentary inquiries held alongside the study.Iterative development of a conceptual framework in consultatio

86、n with the MOD and FCDO.More information on the methodology and interviewees can be found in Annexes A and B respectively.The findings presented below are not intended as definitive,but rather as the basis for further research and discussion conscious especially that RANDs study was undertaken in on

87、ly four weeks,placing tight constraints on the time available both for data gathering and for framework development or testing.This technical report details the studys findings,beginning with the logic behind the proposed conceptual framework(Chapter 2),before moving to different categories of impac

88、t from military AI(Chapters 36),priority issues for action(Chapter 7),lessons from other domains(Chapter 8),and a toolkit of measures to address them(Chapter 9),before concluding with next steps(Chapter 10).3Chapter 2.Towards a framework for strategic impacts from military AI This chapter outlines t

89、he case for a more structured and multidisciplinary approach to mapping potential strategic risks and opportunities arising from military use of AI.In doing so,it delves into fundamental aspects of this technology,including its dual-use nature encompassing both civil and defence applications.Box 2.1

90、 Summary of findings:Chapter 2AI should be understood as a set of general-purpose technologies(GPTs),hardware-enabled but software-based.Unlike traditional military technologies,they are highly democratised and proliferating fast;innovation is being driven by the private sector for civil and commerc

91、ial uses,rather than by governments or defence establishments.Our understanding of the applications and implications of these technologies is improving,but from a low base.Despite a lot of hype around AI,there are significant gaps in both our theoretical understanding and our empirical data on the p

92、otential benefits,drawbacks and risks of different use cases for AI,including in a military setting.This has prompted intense and at times highly ideological debates among global AI experts,and left policy makers grappling with high levels of uncertainty around the likely pace and direction of futur

93、e advances.Crucially,this uncertainty not only exists in relation to the technical dimension of AI,but also its human element.As well as a set of GPTs,AI needs to be understood as a complex socio-technical system.Military applications will be shaped as much by the operational,organisational and cult

94、ural context in which AI technologies are developed and deployed as by the underlying characteristics of the technology itself.Despite the deficiencies in current evidence and understanding,policy makers cannot afford to wait for perfect clarity before acting on AI rollout and governance.Given this

95、urgent need to bring structure to thinking about military AI,this chapter proposes an initial and high-level categorisation of different types of strategic risk and opportunity,with different sub-categories then elaborated upon in subsequent sections of this report.Source:RAND Europe analysis.4Strat

96、egic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligence2.1.Problems with current understanding of the impacts of military AI2.1.1.AI is a family of GPTs,a broad scope that complicates discussions of impact given the diversity of AI techniques a

97、nd applicationsAI is increasingly recognised as a set of GPTs,in some ways akin to the combustion engine,electricity or the Internet.GPTs are technologies that have the potential to drastically impact productivity across many sectors,including defence,and significantly transform societal structures

98、and individual lifestyles.AI is therefore much more than any single given technology(for example,large language models LLMs such as ChatGPT).Rather,AI is a heterogenous group of different systems,methods and applications,each with their own developmental trajectories and implications.There are also

99、strong overlaps with fields such as data science and links to hardware technologies such as computing or robotics.Box 2.2 Definition of AI in the UK Defence AI Strategy a family of general-purpose technologies,any of which may enable machines to perform tasks normally requiring human or biological i

100、ntelligence,especially when the machines learn from data how to do those tasks.Source:UK MOD(2022).Figure 2.1 Depiction of overlaps between AI,ML and data scienceSource:UK MOD(2022).Machine Learning(ML)Natual Language Processing(NLP)Data based classification modelsPredictive analysisRule enginesData

101、managementDescriptive analyticsVisualisationNon-ML based NLPSymbolic logicSpeechVisionDataScienceArtificialIntelligenceOverlapping technologiesAI:Machines that perform tasks normally requiring human intelligence,specially when the machines learn from data how to do those tasks.UK National AI Strateg

102、yML:Computer algorithms that can learn by finding patterns in sample data and then apply this to new data to produce useful outputs,often using neural networks.Alan Turing InstituteData Science:Research that involves the processing of large amounts of data in order to provide insights into real-worl

103、d problems.Alan Turing Institute 5At a high level,AI can be further differentiated into Narrow,Broad and Strong AI:Narrow AI,also sometimes known as Weak AI,refers to AI systems that are designed to perform a narrow task(e.g.facial recognition or Internet searches)and can only operate under a limite

104、d predefined range.2 They are specialised systems that excel in their specific tasks but lack the ability to understand or apply knowledge beyond their programming.Broad AI refers to an approach to AI that focuses on creating systems capable of generalising knowledge and skills across multiple tasks

105、 and domains.These systems would be able to adapt to tasks,but not at the level of sentience or comparable to human performance.Strong AI,such as artificial general intelligence(AGI),refers to those systems able to understand,learn,adapt,and implement knowledge across a broad range of tasks at a lev

106、el equal to or beyond human capabilities.AGI,or the related concept of artificial superintelligence(ASI),is a long-term goal of many research programmes,but largely theoretical at this point.Problematically,debates over the impact of AI suffer all too often from imprecision and conflation between th

107、ese different types of system,or misunderstandings between technical and policy communities who come at the issue with different perspectives,assumptions and skillsets.32 Sheikh et al.(2023).3 David Galbreath,interview by the authors,19 March 2024;Giacomo Persi Paoli,interview by the authors,2 April

108、 2024.4 Schaefer et al.(2021).5 Anonymous,interview by the authors,5 April 2024;Chris Spedding,interview by the authors,15 March 2024.6 Mouton et al.(2023).2.1.2.AI forms part of a complex socio-technical system,with deep uncertainty about both the trajectory and pace of development and real-world a

109、doptionAI should not be seen in isolation but as a complex socio-technical system with a significant human element across the lifecycle,from initial research and development(R&D)through to commercialisation and real-world deployment.4 Humans directly and indirectly affect the strengths,weaknesses an

110、d biases of AI systems,for example through initial engineering,training data for algorithms or the quality of prompts used when tasking LLMs.In turn,the impact of AI is not solely determined by the technology itself,but also by how it is integrated with other systems,how it is used by humans,and how

111、 it is perceived and governed by institutions,regulators and societies.5 This combination of AI with other technologies whether legacy systems or other novel technologies such as biotech,quantum or robotics may result in some of the most significant impacts.6 Yet understanding the nuances of technol

112、ogy convergence requires systems thinking and a diverse mix of interdisciplinary expertise that can be hard to achieve.Uncertainty extends not only to technological progress but also to the absorptive capacity of organisations,such as Defence,to translate new technologies into applications and innov

113、ation.The pace of change will be influenced by a wide range of factors.These include advances in the technology itself,but also policy decisions,societal acceptance and economic conditions.Many national or defence AI strategies focus 6Strategic competition in the age of AI:Emerging risks and opportu

114、nities from military use of artificial intelligenceas much,if not more,on overcoming barriers to AI adoption as on progressing the technical state-of-the-art.7 These include a lack of workforce skills,fragmented approaches to data management or sharing,and antiquated procurement systems.The latter s

115、truggle to deal with the software-driven nature,iterative development cycles or sheer pace of change in AI technologies.8Cultural barriers compound these uncertainties.9 Only a minority of AI technical specialists understand the policy process,or the myriad ways in which AI may affect different orga

116、nisational contexts,including the military.10 Other issues include a lack of security clearances for many AI experts and,as demonstrated by the high-profile backlash within Google against its work for the Pentagon on Project Maven,an ethical aversion of some tech firms and employees to working with

117、Defence.In turn,few political leaders,civil servants or military personnel have deep technical knowledge in AI or related fields such as data science.This means it can be hard to bridge these interdisciplinary,organisational and cultural divides.112.1.3.Discussions of AI in a military context often

118、focus narrowly on the ethics and tactical impact of lethal autonomous weapons systems,or existential threatsFor the reasons outlined above,and more,the future direction and pace of change in AI technologies is uncertain.Similarly unclear 7 UK MOD(2022).8 Andrew van der Lem,interview by the authors,2

119、2 March;Heather Roff,interview by the authors,27 March 2024.9 FCDO official,interview by the authors,19 March 2024.10 Horowitz&Kahn(2023).11 Scharre(2023).12 Anonymous,interview by the authors,8 April 2024;Anonymous,interview by the authors,22 March 2024.13 Meerveld et al.(2023).14 Scharre&Lamberth(

120、2022);Mouton et al.(2023).are what the cascading second-or third-order effects of AI may be,beyond the more obvious direct impacts.12 Much of the defence debate focuses on two extremes:On the one hand,the tactical military impacts and associated ethical,legal and policy dilemmas associated with the

121、combination of AI and autonomy with advances in robotics most notably in relation to lethal autonomous weapons systems(LAWS)or so-called killer robots.13 On the other hand,the potential existential or global catastrophic risks(GCRs)arising from AI,either in terms of AGI(which may not be directly mil

122、itary in nature,but could nonetheless pose a threat to the human species if its goals and values did not align with our own survival)or in terms of AIs interplay with weapons of mass destruction(with the bulk of literature in this area focusing on nuclear escalation risks and on the convergence of A

123、I with bioweapons).14Both issues are important,but the focus on these topics to the exclusion of other risks and opportunities may serve as a barrier to,and distraction from,development of a more holistic and nuanced understanding of the strategic impacts of military AI.And while many of the contrib

124、utors to debates on both LAWS and GCRs bring deep technical knowledge on the realities of AI,popular views are often shaped 7by reductive preconceptions emerging from science fiction(e.g.Terminators,Skynet,etc.).The focus of many policy makers,analysts and civil society organisations on the ethical

125、implications of LAWS has usefully served to underscore the need for governments to develop,integrate and roll out military AI in a responsible matter,and to build social acceptability and political legitimacy for increasingly autonomous systems.15 Equally,being responsible also means adopting milita

126、ry AI at a sufficient pace to deter and defeat aggression from adversaries,state or non-state,who might threaten international peace and security if they themselves exploit AI to gain a decisive battlefield advantage.Such debates can overlook the more prosaic but no less significant impacts that AI

127、can have away from any fighting,for example through integration into procurement,logistics or personnel management systems across Defence.16 Discussions about whether the human should be in,on or out of the loop for decision making are important,especially when considering lethal force.17 So too,tho

128、ugh,are deeper conversations about how to build the best humanmachine teams given the strengths,weaknesses and cognitive biases of both human and machine agents,as well as the demands of different specific tasks or situations(e.g.the speed and nature of the decisions to be made).18 This goes far bey

129、ond LAWS or AIs use in targeting and affects the technologys adoption across all aspects of the wider defence enterprise.15 Hoadley&Lucas(2018).16 Joe Wang,interview by the authors,21 March 2024.17 Wong et al.(2020).18 Schaefer et al.(2021).19 Scharre&Lamberth(2022).Similarly,research and action on

130、GCRs is essential.Even if low-probability,any AI-related developments that could pose an existential risk merit proper analysis,modelling and proactive risk mitigation measures,given the global scale and dire consequences of their potential impacts.19 Equally,though,the fierce debate that has emerge

131、d within the AI community between those focused on existential risk and those focused on nearer-term risks(e.g.concerns around bias,privacy,inequality,etc.)poses a false dichotomy to policy makers.It is imperative to address both types of risk,and this should be feasible with the collective resource

132、s and political bandwidth of major governments and tech firms.This means iteratively developing solutions to the immediate practical challenges posed by AI adoption(e.g.developing governance arrangements to mitigate concerns around safety and bias and accentuate the technologys benefits)while also b

133、eing mindful about any longer-term trends and path dependencies that could lead to GCRs(see Chapter 7).2.2.Contours of a possible conceptual framework2.2.1.The initial framework presented in this report is informed by a set of design criteria,assumptions and caveatsGiven the shortcomings in understa

134、nding outlined in Section 2.1,the MOD and FCDO asked RAND to develop a structured way of thinking about and categorising the strategic risks and opportunities(collectively,8Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligenceimpacts)ari

135、sing from military use of AI.To this end,the research team undertook a review of existing conceptual frameworks and typologies within academic literature,a search which emphasised the lack of any such comprehensive or universally agreed tools.To help develop a new framework,the literature review and

136、 interviews with officials and experts emphasised several design considerations.The remainder of this chapter provides a brief description and visual depiction of a possible framework that aims to meet these design considerations.Chapters 3 to 5 then elaborate on each of the main levels of the frame

137、work,and the categories and sub-categories therein.This initial framework is intended as a guide to further research,analysis and policy discussions.It is certainly not intended to be definitive,given both substantive issues(i.e.,the complexity,uncertainty and rapid pace of change in AI as already m

138、entioned)and practical considerations(i.e.,the fact that this research was undertaken in a very short period).Instead,the framework is designed as the basis for further debate,iteration and refinement to incorporate further insights and learning over time especially in terms of prioritising one stra

139、tegic risk or opportunity over another or improving understanding of timelines to certain breakthroughs in AI which is beyond the scope of this short exploratory study.Box 2.3 Design considerations for a possible conceptual framework Need to recognise AI as a set of GPTs and a complex socio-technica

140、l system with a crucial human dimension.Need to reflect the deep uncertainty that exists around the future trajectories and pace of progress in AI technologies,as well as around organisational,cultural,financial and other barriers to real-world adoption.Need to focus on the under-scrutinised strateg

141、ic level of defence,while acknowledging that tactical and operational-level impacts from military AI may have aggregate effects on the strategic level.Need to move beyond the important but potentially distracting high-profile debates on LAWS or GCRs to consider a wider range of possible impacts aris

142、ing from military use of AI.Need to be flexible and future-proof,with the framework able to accommodate rapid changes in AI technologies rather than being tied to near-term priorities(e.g.LLMs such as ChatGPT)and thus soon rendered obsolete.Need to be coherent with theory(e.g.around strategy,deterre

143、nce or warfare),accessible to a non-technical audience,precise with language and,where possible,orthogonal in categorisation of types of impact.Need to accommodate not only direct impacts from military AI but also potential second-and third-order effects,and the feedback loops that may occur across

144、levels or areas(e.g.military and non-military levers).Source:RAND Europe analysis(2024).9Figure 2.2 Conceptualising the AI lifecycle from development to adoption to impactSource:RAND Europe analysis(2024).2.2.2.The highest level of the framework addresses different types of AI-related risk and oppor

145、tunity,of which the impacts of military AI are a subsetFigure 2.2 depicts different categories of strategic impact,conscious of the dual-use nature and characteristics of AI(e.g.as a set of GPTs)and uncertainty about the pace and direction of future change in the technological state-of-the-art.At th

146、e highest level of abstraction,risks/opportunities can be:General(e.g.AI safety);Military-specific;or Domain-or issue-specific(e.g.nuclear deterrence).Crucially,as noted above,AI does not exist in a vacuum.Rather,it is best understood as a complex adaptive socio-technical system,with an important hu

147、man component.As such,the framework also emphasises the need to think about how various barriers/enablers could shape how quickly and in what ways new AI technologies are absorbed into ministries of defence and military organisations.2.2.3.Subsequent levels of the framework address the interplay of

148、impacts at both global and actor level,across the full continuum of competition and conflictFigure 2.3 moves into more detail on strategic risks and opportunities specifically relating to military AI and breaks this down further across several dimensions.It builds on prior conceptual work for both t

149、he US Government(the Department of Defense,and Office for Net Assessment)and the UK MOD(the Secretary of States Office for Net Assessment and Challenge,and Defence Science and Technology)around the dynamics of strategic competition and how new technologies can affect a given actors strategic advanta

150、ge or Nature and characteristics of AI:General purpose Focus on cognition Rapid page of change etc.Changing state of AI:AI now AI next AI in the future Narrow verses general etc.Adoption of AI:Types of strategic AI risk or opportunityGeneralMilitaryDomainor Issue-specific Pace of change Level of amb

151、ition Understanding AI as a socio-technical system.(e.g.Land,Maritime,Air,Cyber/EM,Space)(e.g.Nuclear)Barriers or enablers of adoption:Pan-DLOD Levels of autonomy Policy,legal and ethical etc.Applications of AI:Descision support Process automation Robotic systems etc.10Strategic competition in the a

152、ge of AI:Emerging risks and opportunities from military use of artificial intelligencelack thereof.20 The framework thereby aims to capture how military use of AI could impact:The international system;The intensity and dynamics of strategic competition or collaboration within that system;and The pot

153、ential and propensity of individual actors to achieve strategic advantage within that persistent global competition.The framework further differentiates between how AI-related impacts manifest differently:Between actor type:differing between superpowers(e.g.US and China),medium 20 Mazarr,Rhoades et

154、al.(2022);Black et al.(2023);Heath et al.(2024).powers(e.g.UK)and small states;democracies and authoritarian regimes;or state and non-state actors.Across the full continuum of cooperation,competition and conflict:from alliance-building through to deterrence,crisis management,conventional warfighting

155、 or even nuclear exchanges.Crucially,the non-linear and relational nature of strategic competition means that the above continuously interact with and impact each other,with a series of feedback loops between impacts at different levels.Figure 2.3 Framework:strategic risks and opportunities of milit

156、ary use of AISource:RAND Europe analysis(2024).Net results of AI-related risks and opportunities at the global levelRisks and opportunities for each actor(e.g.UK):Risks and opportunities of AI across the full continuum of cooperation,competition and conflictE.g.alliances,resilience,sub-threshold,det

157、errence,crisis,conventional warfighting,nuclear,de-escalation,peace.International systemIntensity and dynamics of strategic competitionStrategic stability and escalation ladder.SocietyEconomyDefence enterpriseMilitary capability(e.g.C4ISTAR,fires,mass,logistics,training and exercises,MDI/MDO etc.).A

158、I affects POTENTIAL for advantage(i.e.raw potential in terms of net accounting of strengths/weaknesses)Impact on:Impact on:AI affects PROPENSITY for advantage(i.e.ability to translate potential into performance by marshalling resources in pursuit of strategic objectives)Superpowers verses othersDemo

159、cracy verses authoritarian Non-state(incl.AI firms)(e.g.How can a medium power such as UK navigate intensified superpower competition in an age of AI?).Choices differ by actor type:Strategy makingOrchestrationImplementation(across all instruments of power e.g.DIME levers)Signalling/perceptions.Impac

160、t on:11For example,AI-related impacts on the stability,polarity,institutions,norms and dynamics of the international system have cascading effects on what nation states compete or collaborate over,how,where,why and with whom.In turn,the impacts of military AI on the asymmetric strengths or weaknesse

161、s of a given actor affect their ability to influence strategic outcomes and thus to reshape the international system in their favour,including through war.21 This reflects the dialectical nature of strategy(a contest between opposing wills and intelligences)and the nature of geopolitics as a complex

162、 adaptive system with emergent properties that arise from the interplay of different competing actors and other factors(e.g.environmental or technological change),as well as fog,friction and an element of chance.222.3.SummaryThere is a pressing need for a more structured,nuanced,empirically based an

163、d interdisciplinary debate over the potential strategic risks and opportunities arising from the growing military use of AI.Currently,there are significant pockets of research,analysis and discussion.But,all too often,different scholarly communities are siloed off from one 21 Anonymous stakeholder,i

164、nterview by the authors,25 March 2024.22 Black et al.(2023).another.There are consequently substantial knowledge gaps,as well as misunderstandings between technical and policy specialists.Furthermore,much of the debate focuses on certain high-profile issues(e.g.LAWS or GCRs)at the expense of other t

165、opics;on the tactical at the expense of the strategic;on the risks at the expense of the opportunities;or on the immediate consequences at the expense of the second-and third-order effects that might be most impactful.This report proposes a conceptual framework as the basis for mapping the full brea

166、dth of potential strategic risks and opportunities emerging from current or future use of AI in a military context,based on a set of design criteria derived from expert interviews and an extensive literature review.It is intended as the basis for iteration,not least as the technical feasibility and

167、real-world impacts of different applications become better known through further theoretical work,empirical research or lived experience.The following chapters delve into different levels of the framework,providing more detailed discussion of sub-categories of impact within each,and illustrating pos

168、sible strategic risks and opportunities of note.12Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligenceChapter 3.Impact at the national level This chapter considers the first category of the framework:the risks and opportunities that AI

169、poses at the level of individual strategic actors(typically,but not exclusively,nation states).Given the focus of this study on military rather than civil applications of AI,the discussion is framed through the lens of the constant competition among actors for strategic advantage.This in turn influe

170、nces the balance of power and the degree of peace,prosperity,and stability at the international level topics covered in Chapter 4.Box 3.1 Summary of findings:Chapter 3First,this chapter explores how AI could affect the potential of actors for advantage(i.e.their raw potential in terms of a net accou

171、nting of their strengths and weaknesses).This discussion moves from the higher-level impacts on society and the economy through to those on the defence enterprise.Key issues that emerge include:The potential for AI to drive sweeping economic disruption,or be weaponised to wage economic warfare,with

172、knock-on effects for prosperity,stability and security,as well as the financial settlement for Defence.The potential for AI to be used for information manipulation(e.g.highly sophisticated deepfakes),with consequences for everything from political warfare,subversion,electoral interference,crime and

173、public trust.The potential for AI to transform the productivity of the Defence enterprise and support development and fielding of military capabilities with increased mass,survivability and lethality.A recurring theme,across all these sorts of impacts,is the potential for AI to energise those govern

174、ance systems(i.e.societies,governments,militaries)that manage to adapt and integrate AI into their day-to-day functions,while exacerbating the already acute pressures on those nations or organisations that are left behind.13Second,the chapter then considers how AI might impact the propensity of indi

175、vidual actors for advantage(i.e.their ability to translate that raw potential into actual beneficial outcomes depending on how efficiently they marshal their available resources in pursuit of strategic objectives).Here,the evidence suggests:AI could bring substantial benefits across the strategy cyc

176、le,from intelligence gathering and analysis,through to decision support and consideration of alternative courses of action,as well as enabling more effective collaboration across government and with allies and partners to implement an agreed strategy.Conversely,though,the literature and interviews e

177、mphasise concern about the potential for AI bias,brittleness and failures;a lack of proper understanding of how to get the most of both sides of humanmachine teams;and limited appreciation of the limitations,vulnerabilities,dependencies or through-life support needs of military AI systems not least

178、given acute shortages of AI expertise within government.There is similarly a growing body of work on the risks associated with military AI in terms of the potential unintended consequences for strategic signalling and perceptions(or misperceptions)among different actors.AI hype and rhetoric do not h

179、elp in this regard.If not properly addressed,such issues could drive an arms race narrative around military AI,as well as increasing the chance of unintentional escalation in a crisis.23 UK Government(2021);(2023).Source:RAND Europe analysis.3.1.Understanding strategic advantageThis level of the fra

180、mework focuses on the impacts of military AI on a given actor(e.g.the UK or another country)in terms of their ability to exert an influence on the international system discussed in Chapter 4.Here,the categorisation of AI-related risks and opportunities is built around the concept of strategic advant

181、age one of the central concepts of the UKs Integrated Review in 2021,and its Refresh in 2023.23 Building on prior RAND research for the UK MOD and US DoD,strategic advantage can be understood in terms of both an actors potential and propensity for advantage,as outlined below.Box 3.2 Definition of st

182、rategic advantageA position of strategic advantage is one in which an actor is more likely than others to achieve their objectives in a given contest,crisis or conflict,having influenced the dynamics of competition in their favour and maximised the relevance of their own areas of asymmetric advantag

183、e across all levers of powers.Source:Black et al.(2023),adapted from SONAC(n.d.).14Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligenceFigure 3.1 Relationship between an actors potential and propensity for strategic advantageSource:Blac

184、k et al.(2023).24 By the UK MODs definition,the Defence Lines of Development that make up any military capability are training,equipment,personnel,infrastructure,concepts and doctrine,organisation,information,logistics and interoperability.The US DoD and NATO both have their own equivalents,if using

185、 slightly different terminology.The following sections examine AI-related impacts across each of these sub-categories,beginning with how strategic risks and opportunities arising from AI could affect an actors raw potential for advantage.3.2.AI impact:potential for advantageThis sub-category of the

186、conceptual framework address AI-related impacts in terms of:Impact on society Impact on the economy(including the technology base)Impact on the Defence enterprise(including the MOD,Armed Forces and defence industry)Impact on available military capability(in terms of Defence Lines of Development24).C

187、ollectively,these different themes aim to capture the strategic implications of AI in terms of boosting or undermining an actors(e.g.a POTENTIAL FOR ADVANTAGE:i.e.raw capacity in terms of resources,size of economy,population,geography,etc.Determines the potential levers and maximum theoretical scope

188、 for advantage.PROPENSITY FOR ADVANTAGE:i.e.ability to translate a greater or lesser%of that potential into beneficial outputs(influence,military effect,etc.),based on the effectiveness and efficiency of the state apparatus and the dynamism of national industry,academia,and wider society.Determines

189、the efficiency with which the actors potential is translated into real-world advantages in a given contest,crisis or conflict.PERFORMANCE:i.e.how the actor operates in a specific contest,crisis or conflict,and with what perceived success or failure,based on the imperfect translation of its national

190、potential into real-world effects and the influence of other variables(e.g.,other actors,luck).Determines strategic outcomes(for finite games)or trends in play(for the long game).Influence of external factors15nations)underlying capacity for both hard(i.e.military,coercive)and soft(e.g.diplomatic,pe

191、rsuasive)power,as well as how favourable or volatile their domestic position is socially,politically and economically.This forms the raw potential that can then be translated into useful outputs,and influence over the outcomes of a given competition or conflict,depending on an actors propensity for

192、advantage(i.e.the effectiveness of their governance systems and strategy making and implementation).3.2.1.AI is projected to have profound if unpredictable effects on society,shaping the infosphere,social attitudes and the effectiveness of governance systemsThe literature review and interviews condu

193、cted for this study emphasised that,given AIs 25 Johnson(2021b).26 Tate Nurkin,interview by the authors,20 March 2024;Joe Wang,interview by the authors,21 March 2024.27 Andrew van der Lem,interview by the authors,22 March 2024.role as a set of inherently dual-use GPTs,it is impossible to segregate t

194、he pure military applications of AI from the wider impacts of civilian AI systems that could have cascading effects on defence and security.25This includes the potential impacts on society itself.Prominent risks found by the research team include AI enabling an unprecedented spread of disinformation

195、,causing social upheaval and atomisation,and undermining trust in facts,institutions or democratic politics.26 Conversely,AI holds the prospect of enhancing the quality of public services and offering novel solutions to systemic challenges such as climate change,with the balance of risks and opportu

196、nities seen as dependent on how AI governance unfolds.27Table 3.1 National level:AI impacts on societyDescription Use of AI and big data helps to boost productivity and the quality of public services,leading to improvements in health,education,social care,transport and other policy goals AI provides

197、 new tools for managing the green energy transition and combating the effects of climate change and natural disasters(alleviating some of demand for military aid to civilian authorities/humanitarian assistance and disaster relief)AI supports more efficient R&D,enhancing innovation in other areas of

198、science and technology(S&T)Impacts from AI on the national economy(see Section 3.2.2)and on international security and stability influence patterns of migration and demographic change Improvement in prosperity(see Section 3.2.2)drives wider social benefits e.g.mental health Improvements in policy ou

199、tcomes due to use of AI in turn boost public trust in governance+16Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligenceDescription AI-enabled deepfakes and disinformation campaigns fuel truth decay and governance crisis Extremist organi

200、sations or hostile states exploit AI to support recruitment and propaganda Division between AI haves and have-nots leads to increasing social tension Potential for backlash against AI on ethical,privacy and other grounds Social upheaval from AI(e.g.impact of automation on jobs)undermines public orde

201、r,trust in institutions(incl.the military),and national or alliance will-to-fight Increased challenges to open,democratic societies as AI risks outstrip capacity to respond Use of AI for pervasive surveillance in more authoritarian states(see Section 3.4)Uncertain outcomes from efforts to find techn

202、ical and policy solutions to AI bias AI and its impacts on the infosphere drive unpredictable changes in cultural identities Uncertain impacts from AI on party politics and election interference Uncertain impacts from AI on the social contract Uncertain impacts from AI on public attitudes,values,and

203、 legal and ethical norms,including attitudes towards the use of human vs machine intelligence to perform different tasks Longer term,vast but uncertain implications from AGI or ASI on understandings of what it is to be human or how to reorganise society in an age of machine superintelligence Source:

204、RAND Europe analysis(2024).28 Wright(2019).3.2.2.AI is expected to affect all sectors of the economy,if at differing paces,with potential for both winners and losers from any AI-fuelled economic disruptionAI is seen as a central component in the so-called fourth industrial revolution,with machine in

205、telligence and automation affecting all aspects of value chains and touching all sectors of the economy.28 On the one hand,AI proponents argue that AI and related digital technologies such as robotics,novel compute,telecommunications or big data,could bring substantial improvements in productivity.M

206、ore efficient use of both capital and labour would then deliver better products and services and drive economic growth.This may advantage those knowledge-based economies that are best able to innovate,attract AI-related talent and develop new AI applications,as well as to extract value from data a v

207、ital commodity,with analogies often drawn to natural resources(i.e.data is the new oil).Conversely,there are risks that AI could fuel unemployment and disruption in labour markets,increased volatility in financial markets(given the rise of-?17algorithmic trading),and heightened inequality between th

208、ose countries,regions,companies or individuals able to embrace the opportunities offered by AI and those excluded from partaking 29 Sigfrids et al.(2023).30 Futter(2022).in its economic benefits.29 Equally,there are concerns that AI could be purposefully weaponised as a tool of economic warfare,caus

209、ing volatility or exerting coercive influence.30Table 3.2 National level:AI impacts on the economyDescription AI drives economic growth and regional development AI boosts productivity across various economic sectors,boosting competitiveness and exports AI boosts role of data and analysis in optimisi

210、ng supply chains,financial trades or sanctions Automation reduces need for some jobs,but also creates others,shifting focus onto those activities where humans add most value(e.g.using soft skills)AI drives better economic/fiscal/monetary policy making backed by economic modelling and foresight,helpi

211、ng to bolster economic resilience and predict,absorb and recover from shocks Globally,AI supports development of emerging economies,lifting millions out of poverty Automation drives mass unemployment in some industries(esp.white collar)AI exacerbates skills shortages(e.g.in STEM)and brain drain(e.g.

212、to Silicon Valley)Competition for advantage in AI leads to a race to the bottom on regulatory standards on issues such as data protections,algorithmic bias,harm prevention or privacy AI intensifies wealth and income inequality,concentrating the benefits of AI technologies in the hands of a few power

213、ful countries,companies or super-wealthy individuals Weaponisation of AI drives new forms of economic warfare(e.g.deepfakes or memetic engineering to disrupt financial markets,attacking models behind algorithmic trading,etc.)Abuse of AI for fraud and other economic crimes,terrorist financing or evad

214、ing sanctions Public sector and government finances are stretched by dealing with economic disruptions and the negative externalities of poorly regulated AI,even as private sector actors reap rewards Conversely,AI,coupled with automation,could potentially hinder the advancement of emerging economies

215、,as the reliance on inexpensive labour diminishes,thereby eliminating a key component of their development trajectory-+18Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligenceDescription Uncertainty over which countries best seize the eco

216、nomic benefits of AI(beyond US,China)Uncertain prospects for new forms of AI-related wealth distribution e.g.universal basic incomeSource:RAND Europe analysis(2024).31 Joe Wang,interview by the authors,21 March 2024.32 Anonymous,interview by the authors,25 March 2024.3.2.3.AI is similarly expected t

217、o have profound impacts on the Defence enterprise and on the military capability development,with strategic consequencesJust as AI is expected to reshape civil and commercial organisations,or non-defence industries,so too is it expected to transform the Defence enterprise.This sub-category of the fr

218、amework focuses on the risks and opportunities arising from AIs adoption across a ministry of defence as a military-strategic headquarters;across military commands,other top level budget holders and procurement agencies responsible for developing and acquiring new capabilities;and across the defence

219、 technological and industrial base.Here,the literature and interviews emphasised that AI could have and in some cases is already having profound implications not only on the military tasks,forces and capabilities that Defence needs to provide,but also on the oft-overlooked back office functions that

220、 support those requirements.31 Collectively,these could either transform the strategic bandwidth and competence of a ministry of defence,and the productivity,efficiency,resilience and value-for-money of defence industry,or see them left behind by more agile and innovative competitors with knock-on e

221、ffects on military and strategic competition.32Table 3.3 National level:AI impacts on the Defence enterpriseDescription Advanced AI decision support tools help to improve quality of MOD decision making as a military-strategic headquarters and management of Defences resources and portfolio Further ad

222、vances in AI enhance understanding via better intelligence analysis and prediction Process optimisation using AI tools transforms the efficiency of finance,procurement,logistics,personnel management,maintenance,infrastructure management and other IT systems etc.AI boosts productivity across the Defe

223、nce workforce,increasing strategic bandwidth Applications of AI and other Industry 4.0 technologies to defence industry boosts productivity,industrial competitiveness,exports,supply chain resilience and value for money of defence programmes Applications of AI to defence R&D help to identify and abso

224、rb new S&T at greater pace?+19Description Defence has much less significant buying power or ability to shape markets for dual-use AI technologies compared to traditional defence industry leaving it a rule-taker,not rule-maker Skills shortages restrict Defences ability to adopt and exploit AI at pace

225、 A rushed rollout of AI creates new dependencies(e.g.on foreign AI companies),vendor lock-in and other unintended consequences(e.g.impacts on staff morale or retention)Conversely,rollout of AI in Defence proves too slow,risking Defence being left further and further behind by more innovative competi

226、tors or by private sector organisations Some AI specialists in tech sector are reticent to work with Defence(e.g.as on Project Maven)Defence industry faces increasing competition for AI-related talent from other sectors Unique challenges for Defence in navigating ethical and legal sensitivities of A

227、I and autonomy Uncertain fiscal implications of economic impacts of AI for defence budgetsSource:RAND Europe analysis(2024).At the next level down,AI is expected to affect not only how military capability is delivered,but also what that capability looks like across all its constituent parts,known as

228、 the Defence Lines of Development.Table 3.4 National level:AI impacts on military capabilityDefence Lines of DevelopmentTraining:Equipment:Combination of AI and synthetic environments drives advances in wargaming,training,education,exercises and mission rehearsal Use of AI and autonomous systems red

229、uces need for live training and associated strain on platforms Use of AI drives requirement for new training and education packages,e.g.in AI bias Use of AI demands pipeline of new training data to maintain algorithms,not just training humans Shift from hardware-centric model to emphasis on software

230、 as key enabler of capability Shift from linear to spiral development models,open architectures,data,MLOps pipelines etc.AI enables increased mass through autonomy and automation across all domains AI tools(e.g.predictive analytics and equipment health monitoring)support efforts to bolster availabil

231、ity and readiness of military platforms?-20Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligenceDefence Lines of DevelopmentPersonnel:Infrastructure:Adoption of AI enables exploitation of asymmetric strengths of humanmachine teams AI sup

232、ports new people/career management tools Automation of dull,repetitive tasks boosts morale AI drives new skills requirements from workforce,as well as competition from the private sector Rollout of AI demands new focus on accessing and securing compute,as well as datasets AI tools and autonomous sys

233、tems help optimise infrastructure monitoring,prediction of faults,etc.Increased AI-enabled cyber threats to critical infrastructure,but also AI tools to aid the defenceConcepts and doctrine:Organisation:AI informs better,faster concept and force development,experimentation and testing AI supports op

234、erational analysis and lessons processes by aiding data fusion,analysis and insights into causation Continuing need for legal reviews to assess new AI and autonomous systems against IHL/LOAC Use of AI drives changes in process and accompanying organisational structures Rollout of AI requires overcom

235、ing bureaucratic and organisational cultural barriers to innovation AI places new demands on both leadership and followership at all levelsInformation:Logistics:Adoption of AI requires and reinforces the drive for transformation in Defences data strategy Hostile actors seek to poison data and algori

236、thms Increasing use of synthetic data and secure clouds supports sharing with allies,partners,industry AI tools help predict demand patterns,optimising supply chain and stockpile management AI and robotics boost warehouse efficiency Autonomous systems for last mile logistics in contested environment

237、s cut force protection need21Defence Lines of DevelopmentInteroperability:Opportunities for countries to position themselves as an AI leader and framework nation for others Combination of AI and technologies such as cloud and edge computing,connectivity,etc.,supports rollout of digital architecture

238、for multi-domain operations and bolsters interoperability with allies and partners Conversely,divergent national approaches to AI,autonomy and cross-border data sharing(incl.standards,policy,ethical or legal differences)or varying speeds of AI adoption could undermine alliance cohesionSource:RAND Eu

239、rope analysis(2024).33 Hunter et al.(2023).The uses to which such capabilities might be put,and the implications of AI-enhanced operations across the continuum of competition and conflict,are examined in more detail in Chapter 5.3.3.AI impact:propensity for advantageThis sub-category of the conceptu

240、al framework addresses AI-related impacts on:The full strategy cycle,from strategy making to orchestration and implementation;All levers of power:diplomatic,information,military,economic(DIME);and Signalling to,and perceptions of or by,other actors.Collectively,these focus on the ways in which AI mi

241、ght help an actor(i.e.a government)to be more efficient and effective in mobilising their national resources(i.e.their potential for advantage)in pursuit of security goals in a competitive environment.This section concludes by exploring the temporal dimension and the debate over whether AI will prov

242、ide enduring first-mover advantages to those who are fastest to adopt it or least constrained ethically or whether such advantages will be fleeting as AI technologies proliferate to other actors.3.3.1.Decision support tools could first enhance,then reimagine,strategy makingGiven its focus on machine

243、 intelligence,including via gamified learning,it is unsurprising that much of the literature around the military applications of AI focuses on potential uses in decision making,including at the strategic level.AI has long since surpassed human players at certain strategic games(e.g.chess,Go or real-

244、time strategy videogames).In doing so,it has also demonstrated a capacity to develop winning strategies that no human had previously employed.33 Equally,current limitations to technical capabilities(esp.Narrow AI)means that AI systems especially those reliant on ML techniques are not yet able to rep

245、licate humans ability to generalise,adapt and make 22Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligencestrategic decisions in the face of unfamiliar,uncertain or complex circumstances.34In the near to medium term,then,the focus is on

246、use of AI alongside other technologies(e.g.data science,modelling,synthetic environments,etc.)as a decision support tool,harnessing the strengths of both human and machine intelligences to make more timely,better informed and higher quality decisions:At the beginning of the decision cycle,AI can ass

247、ist with intelligence gathering and analysis.35 AI tools can then assist with cleaning,fusing,processing and analysing vast amounts of data from diverse inputs(be they covert or open source),making sense of complexity,and then prioritising and visualising the most pertinent information to human deci

248、sion makers to help them avoid cognitive overload.36 AI tools then offer means of wargaming or Red Teaming potential strategies,helping depict how other actors may react,serving to improve the robustness of the chosen course of action(COA).37 As they become more sophisticated,AI decision support too

249、ls will also be able to suggest alternative COAs of their own.38 These can then be modelled,e.g.by running many different iterations of the same scenarios in faster-than-real-time,to build out a refined picture of causal relationships and what does and does not work stochastically,enabling selection

250、 34 Rob Solly,interview by the authors,12 April 2024.35 Futter(2022).36 Anonymous,interview with the authors,15 March 2024.37 Geist et al.(2024).38 Andrew Sharpe,interview with the authors,19 March 2024.39 Slapakova et al.(2022).of the most promising strategies.These models can then be improved over

251、 time with real-world data and ML.At the tactical level,there are many situations in which speed of decision making is the driving consideration in apportionment of roles to human vs machine intelligences(e.g.for an uncrewed combat aircraft in a dogfight,the speed of reaction and the fact that comms

252、 links back to a human overseer are likely to be jammed may suggest a need for higher levels of autonomy).39 At the strategic level,though,speed is important,but it is also about tempo,i.e.making high-quality decisions at the right time.As such,the somewhat slower pace of decision making involved in

253、 many cases(with notable exceptions,such as urgently responding to a possible nuclear attack see Chapter 5)presents greater scope for humans to be in,not merely on,the loop.Examples include strategic-level decisions about defence investment and capability development priorities,about deterrence post

254、ure or about military campaign planning.The literature and interviews reveal mixed opinions as to the pace at which Defence will likely be able to field more sophisticated AI decision support tools,as well as the desired levels of autonomy versus human control.Certainly,machine intelligence brings c

255、ertain advantages,e.g.the ability to quickly absorb vast amounts of data that would overwhelm human analysts or to make decisions without 23emotion.40 Newer language models have demonstrated theory of mind(i.e.,the capacity to derive insights about the likely perspectives of other agents)and,related

256、ly,an ability to bluff or deceive both important traits for a strategist.41 Conversely,it is important to stress that AI systems remain subject to biases and brittleness;despite recent progress,models are all too susceptible to mistakes,hallucinations or adversarial attacks.42There is currently sign

257、ificant interest in improving the flexibility of AI systems to deal with a wider range of unfamiliar contexts,tasks and decisions,as manifest in large private and governmental investments in large-scale frontier models.Given the importance of maintaining legitimacy and accountability in decision mak

258、ing,there has been considerable investment in recent years in so-called Explainable AI.These are systems that do away with old black box techniques and present users with justifications behind the machines decisions,articulated in terms that humans can understand.43 There is a significant discussion

259、 surrounding the potential of various AI methods to enhance decision making processes and foster an appropriate level of human trust.This aims to avoid scenarios where humans rely on their own biases and heuristics due to insufficient trust in AI,or conversely place excessive confidence in algorithm

260、ic outputs without a full grasp of the underlying assumptions and constraints.44 Equally,there is hope in the literature that the 40 Tate Nurkin,interview by the authors,20 March 2024.41 Payne(2024).42 Liu&Maas(2021).43 Reinhold&Reuter(2022).44 Johnson(2022);Hughes et al.(2024).45 Meerveld et al.(20

261、23).46 David Galbreath,interview by the authors,19 March 2024.47 Robles&Mallinson(2023).growing use of AI and related advances in decision science could force human strategists to also get better at articulating their own mental models and justifications for decisions.This would lead to humanmachine

262、 teams that are more cognisant of biases and more focused on iterative learning.45In the longer term,AGI could then far surpass the cognitive limitations of humans.This would open a wide range of unprecedented strategies for influencing,deterring or defeating adversaries.Equally,though,we return her

263、e to the macro risks around AI safety,alignment and meaningful human control.3.3.2.Crucially,any use of AI in strategy making will be hotly contestedIt is important to remember the dialectical nature of strategy making.It can be understood as a contest of opposing wills and intelligences in which th

264、e enemy gets a vote.For those concerned with strategic theory,there is debate within the literature and interview data as to whether AI fundamentally changes the nature,rather than merely the character,of strategy and each of the influential precepts handed down by Clausewitz,Sun Tzu and many others

265、.46 More practically,the adoption of AI into different stages of the strategy making and decision cycle also reflects an intensifying contest for decision advantage underway between the UK and its allies(esp.the US),and competitors such as Russia,China or Iran.4724Strategic competition in the age of

266、 AI:Emerging risks and opportunities from military use of artificial intelligenceConcepts of decision and information advantage have emerged as central to Western thinking about how to achieve advantage.They are prominent,for example,in the UKs Integrated Operating Concept(IOpC).48 These ideas build

267、 on the manoeuvrist approach and the legacy of mission command within NATO militaries;emphasising getting inside adversaries observe-orient-decide-act(OODA)loops to out-think them,and then presenting that adversary with multiple dilemmas(i.e.decisions where all the options are bad)so as to shape the

268、ir behaviours in directions that favour NATOs interests and preferred outcomes.49 Both Russian and Chinese doctrines emphasise similar concepts,if framed through the lens of their own cultures and historical experiences50:Russias armed forces stress the uses of AI in operationalising concepts such a

269、s reflexive control(influencing adversarys perceptions,access to information and thinking)or disorganisation(seeking to disrupt and paralyse adversarys C2 structures,especially in the initial period of war).Chinas Peoples Liberation Army(PLA)is modernising fast in preparation for systems destruction

270、 warfare(the idea that advantage comes not from destroying the enemys forces in detail,but rather from targeting key nodes and linkages in their C2 systems to confuse,paralyse and ultimately out-think them)and for informatised and intelligentised future wars using AI.What this translates into,then,i

271、s a competition for advantage between opposing AI-enabled 48 UK Ministry of Defence(2020).49 UK Ministry of Defence(2022b);NATO(2023).50 Black,Lynch et al.(2022).51 Black et al.(2024);Lucas et al.(2024).systems for what the military calls Command,Control,Communications,Computers and Intelligence,Sur

272、veillance,Target Acquisition and Reconnaissance(C4ISTAR),as well as opposing counter-C4ISTAR capabilities:Here,AI becomes a target(e.g.with each side seeking to poison training data for opponents AI algorithms,or to exploit the limitations of AI systems),a defensive aid(e.g.with AI supporting automa

273、ted cyber defence)and a tool supporting offensive action through a mix of kinetic and non-kinetic effects(e.g.via use of AI in electronic warfare,or for detection and targeting of concealed adversary C2 nodes,such as mobile headquarters,for attack with long-range fires).51 This entails a need for an

274、y integration of AI into strategic decision making processes to consider possible threat vectors through which adversaries might seek to poison,exploit or degrade that AI system.It also means having technical and procedural redundancies and reversionary modes so that humans can fall back on non-AI-d

275、ependent systems if their AI systems are disrupted.It is ultimately unclear what the net results will be of this contest between measures and countermeasures,e.g.in terms of the overall offencedefence balance,the transparency or opacity of the battlespace,or the influence of fog and friction on the

276、decision making of different actors.Competitors will be seeking to bolster the robustness of their own AI-enabled strategy making functions while undermining 25or attacking their adversaries AI-enabled systems at the same time.523.3.3.AI also affects orchestration and communication of a strategy onc

277、e one has been formulatedLiterature and interviews also suggest that AI tools could have implications for the next stage of the strategy cycle,namely orchestration and communication of that strategy.Possible impacts include:AI,along with rollout of other digital technologies,could support realisatio

278、n of new ways of collaborating with PAGs,allies,partners,industry,academia,civil society,and NGOs.The nature of AI as a set of GPTs,and as a software-based capability that requires continuous updating,means that use of AI necessarily entails a closer partnership with commercial AI firms.They will be

279、come vital to strategy implementation and thus must be engaged in something deeper than a transactional customersupplier relationship.53 AI tools could also play a significant role in strategic communications,as well as related activities such as information and influence operations.This includes ev

280、erything from content generation through to translation or understanding and automating engagement with different target audiences,through to measures of effect to refine comms strategies.5452 Maurice Chiodo,interview by the authors,28 March 2024.53 There are parallels here to the cyber/electromagne

281、tic and space domains,where private tech firms have been integrated more directly into supporting headquarters(e.g.Commercial Integration Cell at the National Space Operations Centre).54 Joe Wang,interview by the authors,21 March 2024.55 Anonymous,interview by the authors,25 March 2024.Here,the emph

282、asis is primarily on opportunities,with AI supporting better collaboration across institutional,cultural or language boundaries,and communication of strategies to different audiences.There are nonetheless several associated risks,for example arising from a potential backlash against the legitimacy o

283、f strategies developed using AI,or from divergent approaches by different partner organisations to the rollout and regulation of AI tools,including in a military or national security context.These are examined further in Chapter 5s discussion of implications for alliances.3.3.4.AI introduces new ris

284、ks and opportunities to strategy implementation,affecting all instruments of power(DIME)AI is similarly expected to affect the last stage of the strategy cycle,namely implementation.Here,literature and interviews emphasise the potential opportunities arising from AIs role as an enabler and force mul

285、tiplier.Proponents hope AI will generate new efficiencies that increase the reach and likelihood of success when states use instruments of power(i.e.DIME levers)in pursuit of strategic goals.55Of note,research and historical case studies suggest a nations propensity for advantage i.e.their ability t

286、o realise a greater or lesser portion of their raw potential by mobilising and coordinating all parts of government,the private sector and wider society in pursuit of strategic goals hinges on a range of factors.These include the underlying characteristics of the society in question(e.g.national ide

287、ntity and will,levels of openness to free thinking,26Strategic competition in the age of AI:Emerging risks and opportunities from military use of artificial intelligencelearning and innovation,or of competitive diversity and pluralism).They also,crucially,hinge on the effectiveness(both real and per

288、ceived)of governance institutions at responding to the dominant competitive 56 Mazarr(2022);Mazarr et al.(2024).paradigm of the age,including the strategic implications of disruptive technologies such as AI.56 With this in mind,Table 3.5 outlines strategic risks and opportunities of AI across each o

289、f the DIME levers.Table 3.5 National level:AI impacts on strategy implementation using DIME leversRisksOpportunities AI impacts undermine rules-based order,state sovereignty,and institutions(e.g.UN)Shift to machine intelligence undermines value of diplomacy as a human art Backlash against stance on

290、military AI Magnification of soft power with AI Participatory approaches to tech governance that build ties with diverse new partners AI helps develop new mechanisms for building trust,ensuring compliance with treaties etc.Reduced ability to exert global influence in an AI-degraded infosphere Increa

291、sed control of algorithms or private firms over infosphere at states expense New AI-enabled means to understand and influence target audiences globally Improved measures of effect See Section 3.2 on development of the Defence enterprise and military capability See Chapter 5 on competition and confli

292、ct See Section 3.2 on development of the Defence enterprise and military capability See Chapter 5 on competition and conflict Diminishing human influence on economy AI-enabled economic warfare undermines stability of global markets AI frustrates enforcement of sanctions,or fight against terrorist fi

293、nancing and crime Improvements to government finances(and thus discretionary spending)from AI boom New AI-enabled economic statecraft tools New insights into economic data and capacity to better predict effects of policy interventions Source:RAND Europe analysis(2024).273.3.5.Experts raise concerns

294、over AIs potential unintended consequences for strategic signalling and perceptionsAs stressed in preceding sections,strategy is a cycle,with the rollout of AI bringing positive and negative effects across it.As a cycle,it is recursive,rather than linear.Strategy implementation is accompanied by mon

295、itoring of changes in the strategic environment to determine if a change in strategy is required.In part,this means adapting in response to contextual factors(e.g.changes in technology,the economy,climate,etc.).But it also means interpreting the signals that other strategic actors are sending,whethe

296、r consciously or inadvertently,and making sense of what these might mean about their perceptions,plans or decision calculus.Game theory has long provided explanations for how two or more(human)actors interact,sending and receiving signals from each other,and making inferences and judgements that the

297、n inform adjustments to their respective strategies,postures and behaviours.As depicted in Figure 3.2,this is 57 Tate Nurkin,interview by the authors,20 March 2024.based on imperfect self-knowledge about ones own objectives,information,decision making logic,assets,vulnerabilities and capabilities an

298、d even more imperfect knowledge about those of other actors.This fuels the risk of misperceptions and unintentional escalation,especially under the pressure of a crisis.AI tools bring new opportunities for addressing such issues,e.g.by improving intelligence analysis and thus understanding of other

299、actors strategic culture and decision making.But the literature and interviews also show substantial concern among experts that AIs integration into strategic decision making(above all,nuclear command and control)could have unintended consequences.57 Wargames by RAND and others have emphasised the p

300、otential for rapid escalation of crises in which actors fear,wrongly or rightly,that AI may give their adversaries a decisive advantage(e.g.the ability to launch a first strike),or where they believe hostile AI systems may target vital infrastructure(e.g.nuclear C2 systems).28Strategic competition i

301、n the age of AI:Emerging risks and opportunities from military use of artificial intelligenceFigure 3.2 Strategic signalling and(mis)perception with AI through a game-theoretic lensACTOR AACTOR BObjectivesInformationObjectivesInformationDecision-making logicDecision-making logicAssets and vulnerabil

302、itiesAssets and vulnerabilitiesCapabilitiesCapabilitiesPersonPersonMachineSignalling by Actor AKnowledge of Actor AKnowledge of Actor BSignalling by Actor BMachineImpact of win/lossImpact of win/lossBoth offensive and defensive(e.g.Five Pillars)Both offensive and defensive(e.g.Five Pillars)Source:RA

303、ND Europe analysis(2024).Note:the Five Pillars refer to defence capabilities and activities in terms of non-proliferation,deterrence,counter force,active defence and passive defence.58 Scharre&Lamberth(2022).These risks are explored in more detail in specific relation to deterrence,crisis management

304、 and nuclear in Chapter 5.But it is important to flag the broader issue of misperception at this juncture,as it is an essential consideration when looking at the impact of AI on any given actors propensity for advantage.That actor will not be able to achieve their desired strategic outcomes if they

305、send the wrong signals to others and contribute wittingly or unwittingly to escalation dynamics that work against their own national interests,let alone wider peace and security.58 Relatedly,literature and interviews raise concerns about 29the rhetoric of an arms race in military AI.It can be argued

306、 that such terminology is reductive;more polemical than an accurate description of competition over AI,not least given the deep cross-border linkages between national AI sectors(even the US and China).Equally,some experts express concern that such combative or hyped language may fuel misperceptions

307、that create a security dilemma.59For example,it is noteworthy that the open-source literature in the US,China and Russia often presents divergent narratives on military AI:expressing concern that ones own nation is falling behind others(even as other countries say the same about themselves);employin

308、g rhetoric about goals for military AI that conveys strength domestically but might raise unwanted alarm abroad;and at times taking the more bellicose language or technology hype from other countries own military AI programmes at face value.60 Intelligence analyses within government should present a

309、 more nuanced view on other actors objectives,capabilities,posture and policies(e.g.around levels of autonomy).Still,heightened geopolitical tensions and the immaturity of transparency and confidence-building mechanisms(TCBMs)around military AI increase the risk of misperceptions that undermine effo

310、rts to build a global governance architecture for AI or avoid unintended conflict more broadly.6159 Cave&higeartaigh(2019);Roff(2019);Anonymous,interview by authors,20 March 2024.60 Hunter et al.(2023);Nabibaidze(2024).61 Nadibaidze&Miotto(2023).62 Rossiter(2021).3.3.6.While actors pursue first-move

311、r advantage in AI,some advantages are likely to prove more fleeting than others,given the diffusion of AI as a set of GPTsThis chapter has focused on the theme of advantage,and how AI might impact the potential and propensity of a given actor(typically a nation)to achieve advantage,prompting both st

312、rategic risks and opportunities.There is debate within the literature and interviews,however,over how decisive and lasting those advantages may be.Here,there are four main topics of contention:First,the importance of technological breakthroughs(as in AI)to long-term strategic outcomes,either in warf

313、are or competition more broadly.Here,it can be observed that Western strategic theorists or practitioners have typically placed more stock in the decisive battlefield impact of technology and supposed revolutions in military affairs(RMAs).This brings accusations of techno-determinism,underplaying no

314、n-material factors such as will-to-fight.62 Conversely,Soviet/Russian,Maoist/Chinese and Iranian traditions have typically placed greater emphasis on factors such as political will,mobilisation of popular support and indirect approaches.Second,the extent to which RMAs at the tactical level,even if t

315、hey exist,prove decisive at the strategic level.Here,it can be observed that the US and UK have had clear technological superiority in every war they have fought since 1945,winning most battles but not all their wars.30Strategic competition in the age of AI:Emerging risks and opportunities from mili

316、tary use of artificial intelligence Third,the degree to which AI offers decisive first-mover advantage.Historically,the first actor to field a disruptive new technology is often not the one to perfect and most benefit from it.Others quickly learn lessons from the first-movers mistakes and/or develop

317、 better concepts of employment or countermeasures.Furthermore,AI is a set of GPTs,and thus inherently more dual-use than most previous disruptive military technologies,with innovation driven by the private sector and thus much more proliferated and democratised.63 As such,there may be different dyna

318、mics for bespoke military AI(e.g.battle management systems).First-mover advantage may be more enduring here as the algorithms remain classified and hard to discern from external observation,as compared to dual-use AI systems,where technologies and concepts of operations can be quickly emulated by ot

319、hers(as,for example,with Russian and Ukrainian emulation of each others tactics for commercial drones on the battlefield since February 2022).Fourth,the extent to which differing ethical stances will generate long-term advantage when it comes to AI and,relatedly,autonomous systems.While it is often

320、presented as a truism that Western actors approach deployment of military AI with one hand held behind their back,due to 63 Anonymous,interview by authors,22 March 2024.64 Kenneth Payne,interview with the authors,15 March 2024.greater ethical,policy and legal restrictions than found in authoritarian

321、 regimes,it is not clear a)that adversaries necessarily will deploy such systems in fundamentally different ways,b)that lower ethical standards necessarily bring a decisive advantage(especially in the long term,where demonstrating responsible use of military AI is important to maintain political leg

322、itimacy both at home and abroad)or c)that the adversarys less ethical uses of AI cannot be countered by other asymmetric means(e.g.counter-C4ISTAR capabilities).64As such,while this chapter has demonstrated that military AI is likely to create a wide range of strategic risks and opportunities,it is

323、unclear how decisive or long-lasting the resultant advantages will be,given the fluidity of strategic competition and the impact of factors besides AI.This merits more analysis.3.4.SummaryAs will be explored further in Chapter 4,the interplay of these different AI-related impacts at the national lev

324、el contributes to developments at the global level,and vice versa.This creates feedback loops that can either accentuate risks(i.e.vicious cycles)or further enhance strategic opportunities(i.e.virtuous cycles).Figure 3.3 below provides illustrative(and therefore simplified)examples of such potential

325、 feedback loops.31Figure 3.3 Example of virtuous or vicious cycles emerging from strategic impacts of military AISource:RAND Europe analysis(2024).65 Representatives of Adarga,interview with the authors,3 April 2024.66 Johnson(2021b).67 Anonymous,interview with the authors,15 March 2024.The evidence

326、 collected for this short exploratory study suggests that there are significant AI-related opportunities to be had from bolstering governance systems to the benefit of wider society,the economy and public services,not only the military instrument and from use of AI decision support and productivity

327、tools to enhance the efficiency of Defence in delivering its tasks with the finite resources available.65 Conversely,there are significant risks associated with Defence going too fast and making poor choices about the rollout of AI that then create unintended consequences and backlash.Or,alternative

328、ly,going too slow and being left behind by more agile competitors.Responsible development of military AI thus means balancing the need to go fast to secure an advantage over hostile actors who would threaten international peace and security,with the need to take time building governance and safeguar

329、ds.Crucially,literature and interviews emphasise that,while there are differences between military use of AI and applications in other sectors,it is hard to disentangle the question of AIs impacts on Defence from broader questions over its effects on innovation,skills,policy,ethics,law,regulation an

330、d the transformation of governance systems,economies and their underlying societies.66 Defence will be shaped profoundly not only by military-specific AI systems be they friendly or hostile but also by the ways and success with which AI is applied to dealing with other pressing global challenges.The

331、se include(re)building social cohesion,economic growth,trust in institutions,public services and a collective response to climate change.67 With this in mind,the next chapter considers how these impacts from AI at the national level might aggregate at the international level.Poor economic growth con

332、trains UK spending on AI programmesAuthoritarian states gain upper hand in AI,with fewer ethical constraintsHostile actors weaponise AI for disinformation campaigns against UK institutions and marketsClimate change drives conflict and migration,disrupting supply chainsUK leads on development and rol

333、lout of responsible AI for military useAI-enabled strategic decision makes more efficient use of defence resourcesUK military is better able to deter threats and meet MACA/HADR demandsIncreased global security and stability enables period of prosperityUK tech sector attracts capital and talent given growth opportunitiesViciouscycleVirtuouscycleSocial and political volitility frustrates efforts to