《RWS：2024人机协作的未来研究报告-了解人工智能与人类智慧如何协同工作（英文版）（26页）.pdf》由会员分享，可在线阅读，更多相关《RWS：2024人机协作的未来研究报告-了解人工智能与人类智慧如何协同工作（英文版）（26页）.pdf（26页珍藏版）》请在三个皮匠报告上搜索。

1、THE FUTURE OFGENUINE INTELLIGENCETMHUMAN-MACHINECOLLABORATIONA new era of content transformationNo turning backOur world stands at an AI inflection point1.Generative AI (Gen AI)especially Large Language Models(LLMs)have hit the mainstream,giving people and organizations a power they could only dream

2、 of a year ago.As society considers where this path will lead,the debate is buzzing with the noise of so many unanswered questions.If a new superintelligence emerges,will it be friend or foe?Will AI replace human cognitive and creative capabilities,destroying jobs and incomes or create them?Could AI

3、 go rogue,enslaving humanity or even wiping out the planet in pursuit of more paperclips2?All too often,these debates present a binary,zero-sum,either-or proposition:its either Artificial Intelligence or Human Intelligence.This approach isnt just unhelpful.Its wrong.Unhelpful because of the negative

4、 social impacts,spreading a sense of precariousness and distrust in AI systems.Wrong because it fails to recognize that AI is a profoundly human achievement:a powerful example of whats possible when we combine human intelligence and creativity with the design and engineering skills it takes to build

5、 machine intelligence.History shows that enterprise innovation has always been driven by the essential partnership between humans and technology.We use technology to create tools to perform tasks for us or to help us perform tasks more effectively,efficiently and safely.We use pilot projects and pro

6、ofs of concept to test and improve how they work.Then we integrate them into new production processes.Despite what many Gen AI headlines suggest,it takes time for any innovative technology to find its useful place in the enterprise.This was as true of electricity at the beginning of the 20th century

7、 as it is of information communication technologies and AI in the 21st:it takes human insight to unlock a new technologys potential.IntroductionGENUINE INTELLIGENCE21 AI is at an inflection point2 AI and the paperclip problem“Genuine Intelligence:two intellects,one mind As we settle into the fifth i

8、ndustrial revolution the emerging age of AI this historic partnership has a new name.Genuine Intelligence embodies the unique strengths of both artificial intelligence and human intelligence,recognizing that the two are more powerful combined than either are alone.Genuine Intelligence is alive to bo

9、th the weaknesses of artificial intelligence and the limits of human capabilities.Our previous report,Unlocking Global Understanding,explored how people around the world feel conflicted about technology.They love the benefits,but they worry about the pace of change,the pace of life,and the loss of h

10、uman connections and traditions.Genuine Intelligence allows us to hold these tensions with sensitivity and empathy.Our world is being reshaped by an explosion of LLM-fuelled AI solutions from start-ups and Big Tech alike,each racing to outdo the last.The inflection point highlighted by renowned AI r

11、esearcher,Fei Fei Li3,signals how important the choices we make today are in terms of AI responsibility and safety,as well as the challenges of democratizing the technology for public good.While the emergence of Gen AI is an exciting opportunity,Melanie Peterson,Programme Director,TrainAI,is“concern

12、ed that enterprises are so keen to jump aboard the Gen AI train that they will expose themselves to huge risks”.For enterprises that want to get it right,Genuine Intelligence positions Gen AI and LLMs within the evolutionary sweep of information communication technologies,pragmatically evaluating th

13、eir potential,reliability,and risks,without the hype of inflated expectations or the gloom of negativity.The key to content confidence The explosion of Gen AI start-ups,combined with the radically simple chat interface pioneered by OpenAI and the emergence of open-source models,democratizes access t

14、o AI.But this also brings new opportunities for bad actors to exploit the technologys capabilities.Already were seeing rapid growth in fake content whether audio,image,text or video.Then there are the significant challenges for business and society.Gen AI enables content to be produced with unpreced

15、ented variety,velocity and volume(the three core dimensions of Big Data)while questions surround value and veracity in this new content universe.As with Big Data mining,the challenge for all content consumers will be to separate the signal from the noise to identify whats true and valuable from what

16、s false,fraudulent,inaccurate or misleading.For Bart Maczynski,VP Machine Learning,Language Weaver,RWS,“In a world of Gen AI content,the noise-to-signal ratio may become unbearable.”Language is a fundamental feature of human intelligence.While we may take it for granted,its incredibly complex and ha

17、rd to authentically replicate.Its culturally and contextually nuanced,ever-evolving,subtle,multi-layered and enhanced by non-verbal signals.As such,language holds the key to identifying authentic content.As a technology-led company focused on language,content and unlocking global understanding,we se

18、e this as the heart of the Gen AI challenge and the opportunity.Success will be measured by the levels of confidence and trust that content producers and consumers place in our outputs.IntroductionGenuine Intelligence is the key to delivering this content confidence at scale and speed.This report sh

19、ows you why.GENUINE INTELLIGENCE33 AI is an inflection pointWHAT IS GENUINEINTELLIGENCE?Genuine IntelligenceGENUINE INTELLIGENCE4“Genuine Intelligence enables us to work confidently with AI,to mitigate the risks of nave AI implementation,and to unlock the true potential of AI for business and societ

20、y.In contrast to the either-or approach,we start with a both-and approach to AI innovation.The pace of growth in the volume and variety of enterprise content demand makes it possible and necessary to integrate AI with human intelligence.As Matthew Hardy,SVP Linguistic AI,RWS,argues,“Were seeing expo

21、nential growth in content production thats not matched by growth in human or financial resources.Therefore tech and AI are essential.”In this approach,AI does not replace human linguists and translators.Instead,AI complements specialized human skills vastly increasing the speed and volume of deliver

22、y,and making it possible to manage todays enterprise content demands.For Bart Maczynski,“Human intelligence is too precious and unique a resource to be wasted on banal tasks.By focusing our machines on the tedious and repetitive,we allow humans to shine.”Genuine Intelligence represents the unique re

23、ality of our business at RWS.Our in-house team of 1,700+language specialists works closely with a team of 600 AI research and development specialists to optimize the way we transform content and achieve business objectives.We build our approach to Genuine Intelligence on three pillars:InclusiveIn th

24、e words of Melanie Peterson,Programme Director,TrainAI,RWS,Genuine Intelligence represents an authentic partnership between humans and machines.Its a commitment to high-value,high-quality,human-in-the-loop processes.Developing specialized language and Gen AI models takes specialized domain expertise

25、.The TrainAI community at RWS provides a deep pool of such expertise,including niche domains such as opera singing and professional athletics.TrainAI offers clients access to this TrainAI community at RWS an extensive,active pool of 100,000+annotators and linguists,who provide skilled AI data traini

26、ng services in 400+language variants across 175+countries.The community creates,collects,annotates and validates all types of AI data from text,audio,images and videos to multilingual and synthetic data.This ethical,fair and diverse sourcing of human talent,integrated into the production process,is

27、a world away from the ancillary role typically played by humans in the AI supply chain.Too often in a low-value,commoditized market,less responsible tech companies have been implicated in global supply chain abuses enabled by gig economy platforms and employment dynamics.Today,were pushing open the

28、boundaries of inclusivity because we recognize there are still too many barriers to bringing diverse human intelligence into the content transformation market.Genuine Intelligence is the seamless fusion of Artificial Intelligence and Human Intelligence to optimize the way content and data are transf

29、ormed for audiences worldwide.Genuine IntelligenceGENUINE INTELLIGENCE5InclusiveResponsibleCreative“Our Trados translation platform has already achieved the highest WCAG(Web Content Accessibility Guidelines)accessibility standard.But were not stopping there.Thats why were doing a comprehensive acces

30、sibility review of the entire translator journey from application to operation.This is being completed by a 100%blind translator as a PhD project and will involve updating the source code for the entire platform.Our objective is to ensure accessibility for all levels and types of visual impairment.W

31、ere also committed to working with 1,000 visually-impaired translators in 2024.Responsible“There is too little acknowledgement that so much is still unknown about these very new systems.We need a much greater sense of responsibility with a focus on governance,guardrails and ethics.”Vasagi Kothandapa

32、ni,President,Enterprise Services,RWSOur unique approach to sourcing diverse human intelligence reflects the need to emphasize ethics and governance as the field of AI develops.Maria Schnell,Chief Language Officer,RWS,stresses how important it is to“establish basic values that wrap around our approac

33、h to AI.In contrast to the explosive growth of Gen AI start-ups,we need to step back and recognize that just because we can do something,it doesnt mean that we should do it”.Any enterprise looking to adopt technology responsibly must have a non-negotiable commitment to information security and data

34、protection.Weve been managing content transformation for enterprise clients for over 50 years,including the work of our Regulated Industries and Intellectual Property specialist business areas.We understand the complexities of enterprise needs and the critical importance of confidentiality,privacy a

35、nd security.Thats why our in-house technology is backed up by a security-first culture and our ISO 27001 accreditation.This is especially important with the risks surrounding Gen AIs conversational interfaces,such as prompt injection attacks,and also at the upstream level of fine-tuning models throu

36、gh integrating new training data.A responsible approach to AI means being able to identify and mitigate the risks arising from new technologies and techniques.Proprietary foundation models have come under scrutiny because the training data they use compound the biases,prejudices and injustices of th

37、e world we live in(in as much as they are manifest across the internet).Responsible practitioners,like RWS,know better than to blindly disseminate such biases.They work to understand the limitations of individual datasets and models and find solutions to these problems.As explored in our 2021 digita

38、l content trends report,rogue algorithms mustnt be allowed to rule if we are to build and sustain trust in technology and in content.Rigorous and robust processes are a key defence.In the words of Ben Middleton,Principal Architect,RWS,“too much trust in AI creates significant risks.Robust processes

39、mitigate risks and ensure quality”.This is even more important today when many Large Language Models(LLMs)are effectively inscrutable,and hallucinations are recognized as a recurring feature of Gen AI.Genuine IntelligenceGENUINE INTELLIGENCE6The diversity of our 100,000+TrainAI community,our in-hous

40、e team of 1,700+language specialists and our global network of 30,000+linguists,together with established processes for testing,identifying and mitigating bias,are critical to addressing these risks successfully.“Creative“Theres an opportunity here for brands to stand out by standing up for quality

41、information and investing in originality and creativity.”Frank Closset,Head of Product,Tridion,RWSGenuine Intelligence affords brands the opportunity to stand out in a content market that will be increasingly dominated by Gen AI.Theres a very real distinction between human creativity and machine gen

42、eration.And there are very real fears that the growth of Gen AI will mean that the content universe rapidly becomes“flat,standardized,derivative and safe”.The authentic human-machine partnership underpinning Genuine Intelligence means we can guard against the risk,highlighted by Marina Pantcheva,Dir

43、ector,Linguistic AI Services,RWS,“that we will lose our creativity in light of Gen AI”.This active partnership is also vital to make sure“people dont stop thinking and simply default to the machine”with all the negative implications that brings.Its important that brands retain the capacity to produc

44、e original and compelling content to truly engage audiences.This requires the ability to segment content,to appreciate that not all content is created equally,and to target scarce resources accordingly4.Thinking this way about high-quality content production can also help brands overcome current cha

45、llenges with customer experience across the world.Many consumers are turned off by an over-reliance on automation and machine content.Theyre craving human connections and empathy and they want to feel the love from brands5.With budgets under pressure,this is a huge challenge for brands and enterpris

46、es of all types.Fortunately,integrating LLMs into content authoring and transformation processes equips them for success.Take our approach to eLearning6,for example.Recognizing that one size doesnt fit all,our eLearning practice tailors the learning experience around individual learners,their prefer

47、red formats and styles,and their cultural settings,to make sure the content is right for everyone,everywhere,every time.Genuine IntelligenceGENUINE INTELLIGENCE74 2024 Global Consumer Trends5 Accenture Life Trends 20246 RWS launches global eLearning practiceAI lessons from a chess world championIn t

48、he 1990s,the worlds most prominent AI developers were on a mission:to create a machine capable of beating the worlds best chess player at their own game.In 1997,they succeeded,with IBMs Deep Blue famously beating the chess world champion,Garry Kasparov.Recognizing the futility of fighting the machin

49、e on his own,Kasparov became interested in the potential for human and AI collaboration to outperform AI alone.His vision was realized in a 2005 chess competition with the human-plus-machine combination beating both a sole human and a sole machine.What made this interesting was the specific combinat

50、ion that won the tournament.Rather than a Grand Master boosting their skills with a powerful machine,the winning team combined two amateur human chess players playing with three weak computers(no supercomputers).As Kasparov himself put it:“Weak human+machine+better process was superior to a strong c

51、omputer alone and,more remarkably,superior to a strong human+machine+inferior process.”8 With the passage of time,and todays inflated expectations and existential fears surrounding Gen AI,this message has been somewhat lost.Yet for us its a North Star,guiding our approach to Genuine Intelligence.In

52、this partnership,the hard part is the and how do we efficiently and effectively integrate human and machine intelligence into our processes while playing to the strengths of each?Frank Closset,echoing the thoughts of Kasparov,believes that in the not-too-distant future,across most domains and aspect

53、s of life,“humans without AI will be redundant”.Genuine IntelligenceWhile Genuine Intelligence offers a fresh approach to human-machine partnership for enterprises today,its rooted in a back story7 decades before its time.GENUINE INTELLIGENCE87 The Future Of Communication In The Age Of Artificial In

54、telligence8 How To Become A CentaurWhy Genuine Intelligence?Genuine IntelligenceWhy is Genuine Intelligence so vital to sustained success in the content transformation market?Public attitudes toward technology,AI and LLMs hold some of the answers.For example,many people today see AI as a threat to h

55、umanity9,and only 50%of global respondents believe that AI benefits outweigh the risks.10These sentiments could be a direct consequence of popular coverage over many years framing AI as a substitute for human intelligence.Such fears have only been stoked by the debate surrounding the emergence of Ge

56、n AI,including forecasts about AI taking away peoples jobs.By contrast,recent research from Capgemini11 shows that consumer confidence in Gen AI is relatively high.The headline finding is that 73%of global consumers trust content generated by AI.This trust is evidenced by consumer willingness to eng

57、age with Gen AI systems in highly personal aspects of their lives.Capgemini also found that awareness of the risks related to Gen AI systems is relatively low.These trends make the current operating environment a challenging one.Negativity can stifle AI innovation while nave adoption will expose con

58、sumers,like businesses,to new risks and dangers.Genuine Intelligence,built on years of experience with new technologies and AI,can chart a middle way between these poles.Heres how.GENUINE INTELLIGENCE99 AI threatens humanitys future10 Building trust in artificial intelligence11 73%of consumers globa

59、lly say they trust content created by generative AIContent confidenceNo shortcutsBetter togetherInnovate like you mean itContent confidenceThe Misinformation Machine*AI-generated content comes loaded with a profound question:how do you know whats real?As the use of Gen AI and LLMs explodes exponenti

60、ally,so too does the challenge of misinformation.Now anyone can use Gen AI to produce fake text,image,audio or video content.Many examples might seem like harmless fun today,but they seed the potential for more serious fraudulent impersonations,especially as the technology continues to develop.For e

61、xample,there are very real and immediate concerns about the possible impact on key global elections in 2024,including in India the UK and the USA.Surrounding this is a general climate of distrust.As we explored in our 2023 Unlocking Global Understanding report,findings from the long-running Edelman

62、Trust Barometer suggest that distrust is now societys default position.Edelman argues that the way to break this cycle of distrust12 is for brands,enterprises and governments to focus on providing quality information.Misinformation may not be deliberate,but an inherent issue with LLMs is the problem

63、 of hallucinations:presenting false or illogical information as fact.Content producers must not over-rely on AI and risk inadvertently fuelling distrust in content.Instead,they should be confident that their content will stand apart as authentic,accurate and reliable.Content confidence Genuine Intel

64、ligence is fundamentally about providing content producers with the assurance that content transformation will not compromise the accuracy,reliability and authenticity of their communications.This is the foundation of content confidence.Delivering content confidence at the scale and speed required b

65、y global enterprises is a complex challenge.It requires deep experience and expertise with different technologies.Our 47 technology patents and 100+published peer-reviewed articles demonstrate our deep commitment to this.Weve built proven,enterprise-grade products in Trados,Tridion and Language Weav

66、er.Were investing tens of millions in R&D,including a significant share in developing Machine Translation Quality Estimation(MTQE)standards.MTQE is a vital tool in this mix.As we integrate LLMs into our processes and workflows,were increasingly using AI to target human expertise as efficiently as po

67、ssible.In simple terms,were controlling the human impulse to review all machine content.Instead were learning to trust MTQE to prioritize and sustain efficiencies throughout the transformation lifecycle.Another key method for delivering content confidence is Retrieval Augmented Generation(RAG)13.RAG

68、 is a defence against hallucinations and inaccurate outputs that Tom Burkert,Solutions Consultant,TrainAI,RWS,says will be critical to ensuring AI outputs are grounded in objective reality and are verifiable.These are just some of the reasons were respected as a responsible player in the industry,wo

69、rking with many of the largest global technology companies to develop and improve their AI approaches and processes.Increasingly,we provide high-value consultancy services to our clients and partners,helping them develop their understanding of AI technologies and building bespoke AI solutions around

70、 their specific needs.As Ben Middleton puts it,“The RWS heritage is such that we can advise and educate clients as to the best way to work with AI.Our history insulates us from the hype,froth and excitable start-ups on the market.”Content confidenceNo shortcutsBetter togetherInnovate like you mean i

71、tGENUINE INTELLIGENCE1112 Breaking the vicious cycle of distrust13 What is retrieval-augmented generation?*As first referenced in The Coming Wave by Mustafa SuleymanKey components for delivering content confidence on a global scale:Accuracy in translationCultural sensitivity and relevanceConsistency

72、 across languagesExpertise in local nuancesCompliance with local regulationsTargeted communications for diverse audiencesQuality control and fact checkingAdapting to dynamic linguistic trendsContent confidenceNo shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE12Innovate like you

73、 mean itContent confidenceNo shortcutsBetter togetherInnovate like you mean it“Hyper-acceleration So much of todays tech rhetoric is focused on urgency and speed to market.The hype-cycle surrounding new technologies is accelerating as start-ups and solutions continue to emerge.This can lead organiza

74、tions to adopt new technologies in an uncoordinated way,in many cases simply because they want to be seen as a modern,progressive organization thats active in these new spaces.But when enterprise functions are moving at different speeds and enterprise architecture is not keeping pace with developmen

75、ts,this exposes organizations to new risks.We understand the market dynamics the need for vendors to promote new products and solutions to generate returns on often significant investments.But we also understand the implications of bloated technology stacks,obsolete applications and fractured data f

76、lows leading directly to disappointing customer experiences(reflected in research from Accenture,for example14).The launch of ChatGPT and subsequent influx of proprietary and open-source LLMs has been surrounded by huge hype,creating expectations and pressures for enterprise adoption that are yet to

77、 be fulfilled.No-one can afford to stand still today.But without a strategic approach to Gen AI,many enterprises are exposed to the risks of shadow(undeclared and unauthorized)use of these new systems and of exposing their customers to immature solutions.Purposeful innovationBy contrast,purposeful i

78、nnovation flows from a clear understanding of the technology space,an openness to new technologies,and a rigorous approach to testing and evaluation.Our own innovation journey spans decades from developing electronic dictionaries and translation memory and management systems,to first launching stati

79、stical machine translation in 2003,to later developing the worlds most advanced neural machine translation platform.Our patents and AI papers highlight the ways our language specialists and elite scientists continue to redefine the limits of machine translation.Each wave of technological innovation

80、builds on the last,creating a ladder effect.Every step on the ladder benefits from robust testing and development before integration.In this way,were today integrating LLMs into processes that have already been optimized for statistical and then neural machine translation.This technology maturity is

81、 vital to our AI approach.For us,innovation is not for a performance its for a purpose.As Ben Middleton says,in contrast to the hyper-acceleration problem,“Its really important that were not using AI for AIs sake.We need to be very clear on the best use cases and applications for AI.”Were also purpo

82、seful about the way our AI experts work together with our language specialists,with the latter able to inform and influence the direction of our technologies.This integrated,iterative approach,is a fundamental part of the Genuine Intelligence DNA and how we enrich innovation through human-machine co

83、llaboration.Content confidenceNo shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE1414 Accenture Life Trends 2024We announced the beta launch of Evolve in November 2023.A pioneering linguistic AI innovation by Language Weaver,Evolve harnesses the power of secure neural machine tr

84、anslation,linguist-verified quality estimation and LLMs.This unique combination of human and artificial intelligence has the potential to revolutionize translation processes by significantly reducing the time it takes to achieve near human-like quality output.The patent-pending capability of Evolve

85、first analyses Language Weavers initial translation of the content,checks the quality,and suggests areas where the translation can be further improved.It then uses an LLM,fine-tuned by Language Weavers linguistic experts,to target those parts of the content that require further edits and automatical

86、ly deliver a better translation of that content in near real-time.Designed with security in mind,Language Weaver ensures that client content and information are protected throughout the process.Our in-house language specialists and data scientists have been involved in every step of Evolves developm

87、ent.By reducing the time required for a human-in-the-loop during the translation process,Evolve allows language specialists to focus their skills and cultural expertise on the content that most needs review.Content confidenceNo shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE15L

88、anguage Weaver EvolveNo shortcutsContent confidenceNo shortcutsBetter togetherInnovate like you mean itAI is eating itselfThe launch of ChatGPT was met with excitement across the world.The radical simplicity of the chat interface led to ChatGPT becoming the fastest consumer application to reach 100

89、million users in only two months.Since then,the scale of the proprietary LLMs being launched onto the market by the likes of OpenAI,Anthropic,Meta and Google has sent the world hurtling towards a new problem.The open web is already flooded with AI-generated content.The challenge now is that new larg

90、e language models are being trained on data generated not by humans but by other AI systems.Another aspect is the human supply chain used by some tech companies today with low-paid,crowd-sourced workers increasingly using LLMs to fulfil their tasks,including primary research.This means that“the raw

91、material of human insights is being replaced by counterfeit human intelligence”15.Ultimately,these issues will lead to a degrading quality of outputs,even more derivative content,and greater risks of hallucinations.With the fast pace of LLM development the demand for AI data training continues to ri

92、se.A consideration for enterprises is how to do this to deliver outputs that provide content confidence.Content confidenceNo shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE1715 The AI is eating itselfResponsible AI The responsible,rigorous and ethical approach to AI development

93、 we build through Genuine Intelligence provides far more solid AI foundations.For Melanie Peterson,a responsible approach to AI development starts with the design of training data,ensuring that data is only sourced from the right places,lawfully and compliantly,and with appropriate protections aroun

94、d personal data.The next step is working with the best domain experts to work on specialized projects.There are no shortcuts to responsible AI.The calibre of the TrainAI community ensures that none are taken when annotating and labelling training data.With a rigorous recruitment and on-boarding proc

95、ess,ethical working practices and fair remuneration,and a growing community dynamic,the TrainAI community is a highly engaged adjunct to our in-house language and technology specialists.Transparency over training data is essential to understanding the strengths and weaknesses of different language m

96、odels and to mitigating the risks of bias and hallucinations.For Melanie,“Most problems with AI come down to the lack of explainability of many opaque methodologies.”Thats why we make our AI as explainable as possible,developing best practice as we go and working with the data engineers to calibrate

97、 outputs.Were a long way from a trust the computer mindset.Quite the opposite:were always iterating upwards on standards.TrainAIs rigorous processes and guidelines to evaluate language models begin with the use of high-quality domain experts.Theres a lot of work on prompt validation and quality assu

98、rance.There are clear guidelines on testing,identifying and mitigating bias(while trying not to introduce our own biases).There are extensive safety checks and adversarial red team exercises to try to break models.All of this is delivered with constant client collaboration and feedback.Importantly,a

99、ll the same rigorous processes are used to test synthetic data as would be used with an organic data set.Synthetic data can play an important role in preventing AI from eating itself and in addressing gaps in datasets but only if were confident its high quality and fit for purpose.Content confidence

100、No shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE18Key components for delivering responsible AI on a global scale:Content confidenceNo shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE19The quality of training dataClean and accurate data annotationEthical d

101、ata sourcingCustomized datasetsBias identification and mitigationScalability and efficiencyLegal and regulatory adherenceOngoing monitoring,maintenance and updatesBetter togetherContent confidenceNo shortcutsBetter togetherInnovate like you mean it“Autonomous AI Weve seen some of the worlds largest

102、tech companies fast-track solutions to compete in the AI race.Weve heard the stories of AI chatbots going rogue.Weve felt the hype and fear around the potential for AI to replace human intelligence and work.While these examples may be extreme,there are serious issues with protecting language models

103、from prompt injection attacks.Academic researchers have shown they can navigate the defences of LLMs,manipulating them to bypass the guardrails in place to protect against inappropriate outputs.Then theres the issue of hallucinations and inconsistent outputs from LLM systems,together with the challe

104、nge of keeping performance levels consistent.And with end users willing to use AI chatbots for highly personal and sensitive purposes,the risks arising from rogue outputs only increase.These are not wholly new challenges.In recent years,the specialized field of MLOps(Machine Learning DevOps)has deve

105、loped to keep a watchful eye on all aspects of machine learning development and deployment ensuring models behave as intended.This includes actively monitoring machine learning systems in production,recognizing that models can drift and performance degrades in unexpected ways over time.So,how do we

106、steer the future of AI away from unchecked independence and all the risks that brings?Authentic partnership Genuine Intelligence offers a positive alternative to the zero-sum scenario,building an authentic and mutually beneficial partnership between AI systems and human intelligence.“We could let th

107、e robots take over,but we would all be dumber.”Maria Schnell,Chief Language Officer,RWSOur culture of continuous process and performance improvement plays a big part in this.We collaborate closely with clients to ensure the best possible performance and quality outputs.Delivery at scale and speed re

108、quires a constant process of verification and validation.We enable this through direct conversation with client reviewers to make sure were hitting the right quality indicators across the content requirements.Understanding that not all content is created equally and that different quality standards

109、apply to different outputs is critical to success.We optimize for business outcomes rather than abstract linguistic precision,working with clients to audit and segment their content against two axes:value and persistence.For example,low-value,ephemeral content might include social media or email mes

110、saging,while high-value,persistent content might include hero marketing content or annual reports.As Vasagi Kothandapani,President,Enterprise Services,RWS,puts it,“TrainAI can adapt to various quality standards and many different requirements,giving our clients maximum flexibility for their AI resea

111、rch and productization.”Here,the close working relationship between our linguists,technologists and AI experts fuels success continually refining processes as the technology develops and using human intelligence to optimize outputs.Even beyond this in-house collaboration and direct client engagement

112、,AI is a team game.The idea of autonomous AI systems often misses the wide range of stakeholders and collaborators involved in language models and AI platforms.A key partner in our own drive to accelerate and scale LLM integration is Amazon Web Services(AWS).Language Weaver is using AWS machine lear

113、ning and AI services,including Amazon SageMaker,to build new features and scale its Large Language Model capabilities in a private,secure and protected environment.“AWS is the right technology for us when it comes to scaling highly complex and emerging AI technologies like LLMs rapidly and securely,

114、”said Thomas Labarthe,President,Language and Content Technology,RWS.“By using AWS,weve been able to achieve our innovation roadmap faster bringing products to market ahead of competitors16.”Content confidenceNo shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE2116 Language Weaver

115、 scales development of generative AI technologiesKey components for sustaining Genuine Intelligence on a global scale:Content confidenceNo shortcutsBetter togetherInnovate like you mean itGENUINE INTELLIGENCE22Quality control and error correctionEthical considerations and responsibilitiesTraining an

116、d improvement of AI modelsAddressing ambiguity,subtlety and nuanceBuilding and sustaining trust and confidenceAdaptability and flexibilityLegal and regulatory requirementsDelivering Genuine IntelligenceGenuine IntelligenceIts not just our technologies that make Genuine Intelligence a reality:its our

117、 people.From our 100,000-strong TrainAI community,our global network of 30,000+linguists,through to our pioneering AI experts and technologists,its the seamless fusion of all this human intelligence with artificial intelligence that gives our clients the edge.Our talented linguists and translators a

118、re the heart of our business.Theyre passionate about their profession and proud of their expertise and insight.Theyve proven,repeatedly,to be flexible,adaptive and capable of taking on new skills to complement their core specialism.For decades,weve been integrating technology,automation and,most rec

119、ently,machine learning and LLMs into our workflows.Given the niche nature of our industry,its been a quiet revolution in the translation market continually asking new questions of our linguists and translators,demanding new ways of working,and building an enhanced consultative relationship with clie

120、nts and their stakeholders.Everything has changed for todays linguists and translators.Except for one thing:their job title.Weve asked a lot of them throughout the quiet revolution.Now were recognizing that journey.Language Specialist:a new role for a new world The RWS Language Specialist is an expe

121、rt in all aspects of localization blending translation precision with a creativity enriched by cultural insight and nuance.These blended skills deliver relevant and engaging communications across global markets.Our language specialists are technologically adept,experienced in working with machine tr

122、anslation,and expert in digital content workflows and production.To meet the needs of todays clients,our language specialists deeply understand the client organization,brand positioning and tone of voice.Authenticity and consistency in brand communications matter in every market.Our language special

123、ists also deliver unrivalled domain expertise for regulated industries.They make sure every detail is correctly captured and communicated to meet the expectations of the most demanding regulator.And they make sure every prospect,customer and stakeholder has all the right information,communicated in

124、the right way,to support informed decision making.Whatever the clients communications needs,our language specialists will work with them every step of the way,guiding and supporting them as they grow their global presence,providing advice and expertise in every dimension of global digital communicat

125、ions.AI is our first languageOur language specialists are already working expertly with LLMs testing and iterating new approaches to co-working,developing deep skills in prompting and questioning these systems,and informing our approach to developing AI-centred products and solutions.Were developing

126、 an organic relationship between linguists and technology making the use of AI as natural as the use of any other business tool in getting the job done.At RWS,our people dont have to worry about being replaced by AI.They know the realities of the global content market that only linguists with AI and

127、 technology skills will stay relevant in the new working world.In turn,we know how vital their human skills are to develop the right AI solutions and deliver the type of quality content that breeds confidence and trust.Genuine IntelligenceGENUINE INTELLIGENCE24Bring on tomorrowThe launch of ChatGPT

128、unleashed the first wave of new LLMs onto the market,both proprietary and open-source.The next wave of development is signalled by the launch of the Gemini model by Google Deep Mind.Now well see truly multi-modal Gen AI platforms incorporating code,image,text and video into a single model and AI sys

129、tem.Are we reaching the limits of scale in the LLM market?It takes intense energy consumption to run such systems.Looking ahead,we can expect to see smaller,specialized,efficient language models being developed to meet the needs of individual industries,enterprises and use cases.The technology will

130、continue to accelerate as more organizations and enterprises seek to unlock the potential of Gen AI.Analysis by McKinsey17 shows that Gen AI could generate trillions of dollars in aggregate enterprise value,and even more once its fully integrated into job roles.Climbing the innovation ladderYet with

131、 all the noise surrounding Gen AI and LLMs,they have a particular place within the broader sweep of machine learning and AI innovation.McKinsey17 emphasizes that the value of Gen AI is incremental relative to pre-existing expectations of other forms of machine learning and AI.Weve experienced this o

132、urselves when building LLM capabilities into production processes already optimized for statistical and then neural machine translation.This integration of different forms of machine intelligence will continue to develop,creating tighter,more efficient processes and ever-faster and higher-quality ou

133、tputs.This constant iteration is the only way to keep up with the demands of enterprise content transformation.Responsible brands and enterprises will continue to grow their content production exponentially,driven by a more inclusive approach to global markets,intense regulation across global region

134、s,demand for deeply personalized content and growing use of virtual assistants.Misleading and fraudulent content will become a growing problem,while brands will find it harder to stand out in a content market saturated by automated content.Artificial intelligence applications will continue to improv

135、e,learning from their experiences and improving their outputs accordingly.But human intelligence will also level up,as people learn better ways to partner with these AI systems through direct experience.In search of truthWith the accelerated pace of content generation today,we believe our human-mach

136、ine collaboration philosophy,Genuine Intelligence,is the only path towards a responsible future.It assures both global organizations and consumers that the content that theyre creating and engaging with is authentic,accurate,and trustworthy.When trust is tough to earn,we need to know whats right and

137、 whats real.Thats why a world of Genuine Intelligence is one we can all believe in.Its been an extraordinary 12 months in the technology,content and language industries.Genuine IntelligenceGENUINE INTELLIGENCE2517 Generative AI is poised to unleash the next wave of productivity.About RWSRWS Holdings

138、 plc is a unique,world-leading provider of technology-enabled language,content and intellectual property services.Through content transformation and multilingual data analysis,our combination of AI-enabled technology and human expertise helps our clients to grow by ensuring they are understood anywh

139、ere,in any language.Our purpose is unlocking global understanding.By combining cultural understanding,client understanding and technical understanding,our services and technology assist our clients to acquire and retain customers,deliver engaging user experiences,maintain compliance and gain actiona

140、ble insights into their data and content.Over the past 20 years weve been evolving our own AI solutions as well as helping clients to explore,build and use multilingual AI applications.With 40+AI-related patents and more than 100 peer-reviewed papers,we have the experience and expertise to support c

141、lients on their AI journey.We work with over 80%of the worlds top 100 brands,more than three-quarters of Fortunes 20 Most Admired Companies and almost all of the top pharmaceutical companies,investment banks,law firms and patent filers.Our client base spans Europe,Asia Pacific and North and South Am

142、erica.Our 65+global locations across five continents service clients in the automotive,chemical,financial,legal,medical,pharmaceutical,technology and telecommunications sectors.Founded in 1958,RWS is headquartered in the UK and publicly listed on AIM,the London Stock Exchange regulated market(RWS.L).For further information,please visit: 2024 All rights reserved.Information contained herein is deemed confidential and the proprietary information of RWS Group*.*RWS Group shall mean RWS Holdings plc for and on behalf of its affiliates and subsidiaries.