《Raconteur：2023年數據和人工智能的未來報告（英文版）（11頁）.pdf》由會員分享，可在線閱讀，更多相關《Raconteur：2023年數據和人工智能的未來報告（英文版）（11頁）.pdf（11頁珍藏版）》請在三個皮匠報告上搜索。

1、22/03/2023INDEPENDEN T P U B L I C AT I O N BY#0861R AC O N T EU R.NE TMake moments ofhuman connection more FUTURE OF DATA&AICHATGPTS PLACE IN THE FUTURE OF WORKIS PREDICTIVE AI FUNDAMENTALLY LIMITED?1206THE NEW FRONTIER FOR DIVERSITY IN AI14R A C O N T E U R.N E TF U T U R E O F D ATA&A I0302Report

2、s editorIan Deeringre you scared yet,human?Artificial intelligence has proliferated with trans-formative effects in sectors from au-tonomous vehicles to personalised shopping.But the latest use of AI to generate content such as text,imag-es or audio has caused quite a stir.ChatGPT is a particularly

3、superior language model,even passing the US medical speciality exam.Thats not to say there havent been some bloopers.The OpenAI release has delivered inaccurate information and even abuse.It also self-warns that it could generate bias and harmful instructions.Even before tools such as ChatGPT,Bard a

4、nd Dall-E 2 won such wide at-tention,there have been concerns that discrimination and bias are baked into algorithms.This phase of newer,more accessible AI could greatly impact users trust in the technology.What,then,for busi-nesses that have rushed to adopt these latest forms of AI should they coun

5、t on its longevity and if so,is it possible to embed AI ethics,so that mistrust doesnt hurt their rep-utation and bottom line?Despite some opinions that gener-ative AI is a fad,most people think its here to stay.Still,as a Morning Consult survey of 10,000 US adults revealed,only 10%of the public fin

6、d generative AI“very trustworthy”.Drilling down further,that level of trust wavers between demographic groups,with younger cohorts,pri-marily male,more trustful and will-ing to adopt early than older generations,who are generally hesi-tant to pick up new technology.It isnt only the end users who har

7、-bour doubts,though.High-profile gaffes such as when Googles Bard circulated false facts in search re-sults have perhaps been a caution-ary tale.Apple Inc also delayed approving updates to its email app with AI-powered language tools,over concerns that it might show in-appropriate content to childre

8、n.Thats not to say that generative AI isnt hugely useful.It is applied across business functions,from marketing and sales to IT and engi-neering.Its applications range from crafting text to cutting through dense material to aid understanding and answer complex questions.“Companies are investing a lo

9、t in data and tech,”says Karl Weaver,SVP consulting,EMEA,at Media-Link.But he cautions that there is a general acceptance of analytics.Added to which,“Businesses see what their competitors are doing and think they should perhaps be doing the same or risk missing out.All that could cause a misstep an

10、d subsequently a trust problem.”That isnt to say that businesses should or even can avoid the wave.But CEOs and boards should step back and think,in an informed way,about why they might be using AI tools,including these most recent iterations.If its a genuine desire to improve customer experience,it

11、 should be set up to ensure they are serving that area.“Were developing a unified and common understanding of the big risks,”says Robert Grosvenor,a managing director at Alvarez&Mar-sal.“But there is a long way to go to translate high-level,principle-based objectives into codified require-ments,stan

12、dards and controls.”The scope and scale of AIs appli-cations span different industries and sectors and the possibilities of harm and degree of risk vary accord-ingly;impacts could even differ within the same organisation and be unforeseeable.As a result,individu-al business functions could have uniq

13、ue workflows for their use of AI and the data they need,instead of relying on analytics or compliance functions to dictate a cookie-cutter approach to rules about using AI.Andrew Strait is associate director of emerging technology and indus-try practice at the Ada Lovelace In-stitute,which researche

14、s the impact of data and AI on people and society.He says that distrust in some AI technologies has meant people want to see more regulation.Consumers can be confident that the food they buy in a supermarket is relatively safe,for example.But the same level of regulatory oversight,and thus consumer

15、trust,doesnt exist for AI the technology has developed too quickly for the regulation to keep up.People want transparency about the data practices involved in AI and individual privacy.But Strait ob-serves that there is a misconception that simply telling people what you are doing is enough to build

16、 trust.“That lacks a deep understanding of the context in which someone is experiencing your product,”Strait says.He would like people to partic-ipate in the governance of AI.Data cooperatives could be the means to that end.A representative collective of people in a data set de-cides who accesses th

17、e data.This is in action in Spanish healthcare,where the cooperative Salus Coop gives citizens control over their data for research purposes.Despite the general regulatory lag,the EU is tackling the problem.The Regulation Laying Down Harmo-nised Rules on Artificial Intelli-gence,the so-called AI Act

18、,is under discussion.One of the regulations proposals to address ethical dilem-mas and safeguards would assign risk levels to AI uses,while enabling its benefits.Generative AI tools for use in sensitive areas such as re-cruitment would be classed as a high risk designation.This would trigger conform

19、ity assessments to check that certain standards were met,hopefully reassuring end-users.Human apprehension about AI,or a hesitancy to fully trust it,persists.But while we attribute human-like characteristics to AI,it is machinery.As actual humans,we can still check fact from fiction and set expectat

20、ions around acceptable uses of AI and businesses can lead that charge.Can distrust in AI impact your business?FUTURE OF DATA&AIThe hype around generative AI has further influenced public trust in the technology.Businesses can use this as a guide to how they use it and the ethics they applyDistribute

21、d inSophia AkramAlthough this publication is funded through advertising and sponsorship,all editorial is without bias and sponsored features are clearly labelled.For an upcoming schedule,partnership inquiries or feedback,please call+44(0)20 3877 3800 or email Raconteur is a leading publisher of spec

22、ial-interest content and research.Its publications and articles cover a wide range of topics,including business,finance,sustainability,healthcare,lifestyle and technology.Raconteur special reports are published exclusively in The Times and The Sunday Times as well as online at The information contai

23、ned in this publication has been obtained from sources the Proprietors believe to be correct.However,no legal liability can be accepted for any errors.No part of this publication may be reproduced without the prior consent of the Publisher.Raconteur MediaIBM,2022AI says theres a threat.Data bias cre

24、ep is hard to avoid.Thats why we do more to address it.Pairing the right people with the right skills to your projectsperforming bias detection and removal at every stage in the process,from worker selection and compensation,to task design,completion and validation,and final dataset output.Giving yo

25、u targeted data you can rely on for your machine learning.On time,every time.For fair and inclusive data you can depend on,choose TrainAI by RWS your responsible AI partner.Get started at R U S TSophia AkramJournalist who specialises in foreign policy,human rights,global development and their crosso

26、ver with business and lifestyle.Natasha SerafimovskaEdTech SaaS expert and freelance writer,who specialises in the future ofwork,digital transformation and workplace learning.Jon AxworthyJournalist,specialising in health,tech,science and the future,with work published in T3,Wareable and The Ambient.

27、Mark WalshNew York-based freelance writer covering business,technology and media.He has contributed to TheS and The Guardian,among others.Alison ColemanWriter and editor,working as a senior contributor at Forbes,with articles published inThe Guardian and Quarterly,among others.Emma WoollacottJournal

28、ist writing on business,technology and science.She is a regular contributor to the BBC News website,Forbes and Private Eye.Bernard MarrRaconteurs columnist,author and one of the worlds most successful social media influencers at the intersection of business and technology.ContributorsAraconteurracon

29、teur.storiesraconteur-media/future-data-ai-EditorSarah VizardSub-editorChristina RyderChief sub-editorNeil ColeCommercial content editorsLaura BithellJoy PersaudDeputy reports editorJames SuttonDesign/production assistant Louis NassDesignKellie JerrardHarry Lewis-IrlamColm McDermottSean Wyatt-Livesl

30、eyDesign directorTim WhitlockIllustrationCelina LuceySamuele MottaCampaign managerEthan WredeHead of productionJustyna OConnellAssociate commercial editorPhoebe BorwellDiffusionBeeWHATS STANDING IN THE WAY OF TRUST?Barriers to developing trustworthy AI in organisations worldwide Lack of skills and t

31、raining to develop and manage trustworthy AIAI governance and management tools that dont work across all data environmentsLack of an AI strategyAI outcomes that arent explainableLack of company guidelines to develop trustworthy,ethical AIAI vendors who dont include explainability featuresLack of reg

32、ulatory guidance from governments or industryBuilding models on data that has inherent bias63%60%59%57%57%57%56%56%R A C O N T E U R.N E TF U T U R E O F D ATA&A I0504THE COMPETITIVE EDGEMotivations of product managers and marketers for conducting competitive intelligencehe business landscape is str

33、ewn with the corpses of corporations that failed to notice what competitors were up to.Think of Blockbusters dismissive attitude to Netflix in the noughties and then even the latters lack of fore-sight in seeing the streaming wars on the horizon.This is why organisations need robust competitive inte

34、lligence(CI)strategies:to ensure they are not blind-sided by competitors or dis-ruptive business models.But gathering data as part of CI can be daunting because there is so much information available.Industry ex-perts blogs,financial reports,news media items,public data sources and more are all ther

35、e for harvesting.The good news is that AI and ma-chine learning can streamline and accelerate this task.SaaS platforms use AI to track and collect historical and real-time data insights,which al-lows businesses to use information from competitors digital footprints.“There is exponentially more infor

36、-mation online in the digital footprint of every business relative to a decade ago,”says Jonah Lopin,founder and CEO of Crayon,a competitive intelli-gence software platform.“If you have the tools to aggregate and analyse this information you can Theres a tremendous amount of competitor data in the p

37、ublic domain but its often dispersed and opaque.AI-enabled competitive intelligence can help companies gather it,sort it and draw insights from itGather it,sort it,use itunderstand a competitors product life cycle,pricing and more.Those in-sights can help a firm anticipate that competitors moves.”Au

38、tomating the platforms process-es ensures the intelligence is timely enough for a business to take prompt action.Drawing on the guidance and recommendations of AI at execu-tive-level meetings could also en-hance agile thinking when critical actions are needed.The evidence suggests that more business

39、es are using the technology.More than 10,000 roles for competi-tive intelligence are advertised on LinkedIn,for instance.The global market size of CI tools is projected to hit$82bn(68bn)by 2027,according to Fortune Business Insights.A significant proportion of this spend is likely to come from corpo

40、ra-tions whose deeper pockets allow them to deploy state-of-the-art large language models(LLMs)like GPT-3,which can provide more sophisticat-ed data insights.GPT-3 is one of the most straightforward models for building CI.But it needs infrastruc-ture,skills and software to work ef-fectively,which do

41、nt come cheap.Regardless of the computational power of the AI model,it needs tar-geted and verified data to function and benefit a companys CI strategy.That data needs to be drawn from multiple sources in order to prevent its intellect from becoming circular and repetitive.AI models are only as inte

42、lligent as the training data they are given.The more data theyre trained on,the more accurate,versatile and useful they are to a business.This means there must be skilled oversight of the quality of the data.“Training with low-quality data,like social-media posts or blogs,can introduce more noise in

43、to the learn-ing process,which can confuse the model and decrease the quality of the outputs,”explains Kalyan Veera-machaneni,a principal research sci-entist at the MIT Institute for Data,Systems and Society.Data bias can also be a consequence of working with low-quality data.For example,if data is

44、limited to text from a particular group it can skew the AIs function,lower overall per-formance and reduce the reliability of the intelligence.But by monitoring data quality from sources with editorial filters or filtered web content and not relying too heavily on user-generated con-tent,companies c

45、an increase the quality of the inputs,which will also raise the level of intelligence that they are receiving,in terms of relia-bility and usefulness.Whether you have decided to base your CI strategy on insights which are provided by an advanced LLM or a more cookie-cutter SaaS platform,you will sti

46、ll need to involve people.Then,you have the combination of matching competitive intelligence with creative intelligence from in-house teams to maximise any advantages and make the business stand out from the crowd.“AI can be an asset for brands to stay ahead of the game in competi-tive markets.It ca

47、n be used to help to increase efficiency and insight within marketing teams.And it can reduce time-consuming,laborious processes,”says Anthony Lamy,vice-president,client partnerships at VidMob.“But without human input to give context to the data that AI generates,brands wont be able to use the creat

48、ive intelligence they have at their fingertips.”Businesses can therefore use the technology to discover a competitive edge not by relinquishing complete control to AI,but by using it in tan-dem with human creative teams to increase their creative intelligence.It seems,then,that while there was eithe

49、r huge distrust or blind faith in what the technology could achieve for an organisations CI strategy,there is now a realisation that if you arent seriously considering invest-ment in or scaling up AI-driven CI,its very likely that your competition will be.Without human input to give context to the d

50、ata AI can generate,brands will be unable to utilise the creative intelligence they have at their fingertipsJon AxworthyC O M P E T I T I V E I N T E L L I G E N C ETLaurence Dutton via Getty ImagesCommercial featureWhy web scraping is the future of data-led innovationAccess to actionable data is be

51、coming vital for businesses.Whether youre fuelling AI solutions or looking for insights into consumer behaviour,web scraping can help ensure reliable access to the data decision-makers needor many years,web scraping has been a central concept in the tech world and has impacted our lives in ways many

52、 of us are unaware of.“Even regular inter-net users constantly run into busi-nesses that could only be possible through web scraping,”says Juras Jurnas,the chief operating officer of Oxylabs,a web-intelligence-acqui-sition solution and premium-proxy provider.Data scraped from the internet using auto

53、mation provides the backbone of everything from search engines to travel-fare aggre-gators,price-comparison websites and many other services.But what has long been an unsung,central part of our digital lives is now entering the limelight.The rise of AI has captured the attention of even the most tec

54、hnophobic executives,who are now seeking ways to integrate it into their business models.Yet behind every AI model is a powerful corpus of training data most of which is scraped from the web.“We are moving in a new direction,”says Jurnas.“One where AI namely machine-learning models are becoming ubiq

55、uitous.ChatGPT,Bing Chat,Google Bard are all based on the same principles.”Oxylabs recently surveyed more than 1,000 senior ecommerce industry data decision-makers in the UK and US.Nine in 10 of them said they thought web scraping would become a more impor-tant part of their focus in the coming year

56、s.That focus is even more important now that weve entered the age of AI,where large language models(LLMs),chatbots and image generating models all have a voracious appetite for training data,which is needed to offer iterative improvements in how busi-nesses are run.However,its difficult to get to a

57、point where AI can reliably help you.“Machine-learning models are hungry for data,as they need millions,some-times even billions,of data points to provide high levels of accuracy and predictive power,”says Jurnas.“Web scraping can provide companies devel-oping machine-learning models and AI with the

58、 data they need.”How to solve the data quality problemYet its not simply a case of setting a web scraper going and overhauling the way your business works.Theres a maxim in the world of AI:garbage in,garbage out.A model is only ever as good as the data its trained on,which means the level and scale

59、of high-quality data that is needed to realise the full potential of the future is greater than ever before.“Enormous volumes of data will be fed into models that will often func-tion as black boxes,”says Jurnas.“If a large part of the data is faulty,the results will be unpredictable and could cause

60、 damage.”The risks of low-quality data infiltrat-ing an AI system and then poisoning the well has real-world ramifications.Google saw$100bn wiped off its value within a day of unveiling its own AI search tool,Bard,which was shown to answer simple questions incorrectly.Having high-quality web scrapin

61、g is an essential part of solving the prob-lem,without which it would be exceed-ingly easy to run into issues.Businesses must dedicate proper attention and resources to this area in order to remain competitive.It will soon become normal for businesses to use AI in their day-to-day operations,meaning

62、 those that dont adopt the technology could quickly fall behind.Walking the tightropeWhether your business benefits from the AI revolution is likely to come down to the quality of data youre working with.From good data you can get good AI-generated insights,which can help you innovate and overhaul y

63、our busi-ness for the future.“Web scraping is the way to enable the creation of advanced AI through the provision of high-quality data,”says Jurnas.“On the other hand,the data has to be carefully managed and acquired from reputable and ethical solution provid-ers.A single misstep can be exception-al

64、ly costly.”Those missteps can be simple errors like Googles,which if implemented into a business could lead you into a new market or a new contract that ends up harming,rather than helping your business.Another common mis-step is training a model on biased data,which can come back to harm your reput

65、ation.This was the case for Microsofts Tay chatbot in 2016,which shortly after release began expressing racist sentiments.Finding a good partner who can help you remain on the cusp of innovation,while doing so responsibly and ethically,can be a challenge.“At Oxylabs,we pro-vide web-scraping solution

66、s for all busi-nesses that need public web data,”says Jurnas.“Our services may be used to gather data for optimisation,to build machine-learning models,or even combine both purposes.”Trusted,truthful partnershipOxylabs is trusted by companies big and small to conduct web scraping that can help build

67、 AI models that work.“We solve the greatest pain point for any data-driven venture:data acquisi-tion,”says Jurnas.“Through our numerous scraper APIs,companies can extract as much publicly available data as they need and have it delivered in real time.”Among the list of clients trusted by Oxylabs to

68、build the foundations of their tech stack through web scraping is the Lithuanian government.Oxylabs created a solution that scans the Lithuanian IP address space on the internet and detects illegal sexual and child-abuse imagery.Evidence is then forwarded to specialists for review.It is evidence of

69、the level of trust placed in the company but it is far from the only use of the AI it helps enable.“We firmly believe in an AI-led future,but we understand that it will be data-hungry,”says Jurnas.“Our goal is to enable businesses of all sizes to get the web intelligence they need,build machine-lear

70、ning models and optimise their business processes.If data is the foundation of your business,Oxylabs will help you collect it reliably,safely and ethically.”For more information visit oxylabs.ioEnormous volumes of data will be fed into models that will often function as black boxes.If a large part o

71、f the data is faulty,the results will be unpredictable and could cause damageFOxylabs,2022WEB SCRAPING HELPS UNLOCK THE POWER OF EXTERNAL DATAShare of technology business leaders expecting to shift towards web scraping in the next 12 monthsMost common uses of external data in organisations worldwide

72、Market research UK USA AllYesNoForecasting consumer demandCompetitor monitoringDevelopment of products and servicesDynamic pricingOptimising customer serviceExternal data is not used in my organisationAll91%9%UK86%14%USA4%96%90%of senior ecommerce industry data decision-makers in the UK and US think

73、 web scraping will become a more important part of their focus in the coming yearsOxylabs,202269%68%49%45%44%32%1%78%65%65%68%55%48%0.5%73%66%57%56%50%40%1%Understanding how our company compares to the competitionPinpointing ways to differentiate our companyGetting a feel for the direction competito

74、rs are heading inMaking sure we stay ahead of the competitionClosing the gap on the leaders in our industryFinding new ideas to help our company innovateReplicating and matching rivals to negate their advantage84%75%75%65%40%33%19%R A C O N T E U R.N E TF U T U R E O F D ATA&A I0706or the past few m

75、onths,it seems everybody has been going AI crazy.Futurists have long been predicting that it will revolutionise our lives and its plain to see it already is.Mostly,though,this has been happening under the hood quietly powering tools we use every day such as Google,Netf-lix and Uber in a way that is(

76、by de-sign)invisible to the user.ChatGPT and related tools and ap-plications such as Bing and the soon-to-be-released Bard,on the other hand,are in-your-face AI.Millions,who have now had the chance to see them in action,have been left in no doubt that this is something truly new,genuinely revolution

77、ary and a little(perhaps alot)scary.News is moving quickly.At the time of writing,Microsoft is thought to have scaled back and limited the ChatGPT functionalities it recently integrated into its Bing search engine.This comes following reports of users who were cheeked and chastised by the feisty cha

78、tbot,and others who have worked out clever ways of instructing it to adopt new,not entirely helpful personali-ties.Some researchers have even claimed that the machine learn-ing-powered algorithms have been telling them that they are sentient and want to be alive.Its fair to say that weve all had a l

79、ot of fun and its thrown up some in-teresting ethical and philosophical debates.But is it set to be as revolu-tionary as it seems when it comes to changing the way we work?Or is it a flash-in-the-pan that will be forgot-ten about when we eventually real-ise it still isnt quite good enough to let loo

80、se on really important tasks?Although it is impressive technol-ogy,anyone who has used genera-tive AI for some time will have bumped up against some of its limi-tations.The most glaring is proba-bly the fact that it isnt capable of original thought.ChatGPT(and other applications that will follow sho

81、rtly)draws all of the knowledge that goes into its output from its training data.In simplified terms,it constructs responses to questions and queries by analysing millions of words of text that have previously been writ-ten and applying probability to determine the best thing to say next.It is a lan

82、guage model that under-stands the structure and context of sentences and therefore is capable of creating its own.What it cant do,though,is come up with an origi-nal idea or an answer to a question that has never been correctly answered before.For most use cases,this is proba-bly fine.No one expects

83、 to use it to ask,for example,the secret of gener-ating perpetual energy.(Although it can certainly summarise a large amount of the corpus of existing human knowledge on the issue).Where it is likely to be useful is in au-tomating routine parts of our work.This limitation is the main reason that nat

84、ural language technology is not(yet)simply going to replace hu-mans.There will,for the foreseeable future,be a need for humans to oversee and steer AI,providing the big picture direction and the origi-nal thought needed for truly useful or valuable endeavour.This is why,when speculating about how th

85、is technology is likely to impact our working lives,it makes sense to look at the particu-lar abilities and skills that it can augment,rather than at specific jobs or professions that may or(more likely)may not be in danger of being automated out of existence.Any such list has to start with writ-ing

86、.On the face of it,this is ChatGPTs main function to pro-duce text.If youre going to use it to write,though,its important to re-member that it wont generate any-thing new or original.Where it can be helpful is with suggesting ideas about how to structure an essay,ar-ticle,blog or social media post,o

87、r generating a list of the most impor-tant points that need to be covered.Just be wary that if your audience comes to you for specialist knowl-edge,expert opinion or simply be-cause they like your personality,then AI-generated content is likely to leave them cold.Another useful capability is gener-a

88、ting code.Not limited to human languages,ChatGPT can write code in several popular languages,in-cluding C+,JavaScript and Python.It can also error-check existing code.By taking advantage of this,just about anyone can become ca-pable of quickly creating simple computer programs to auto-mate routine e

89、lements of their work.Developing this skill is likely to be increasingly impor-tant in many professions.It can also be a great tool for research.It can be more useful than a search engine but its out-put can often include errors or omis-sions.Therefore,the ability to review and verify the informatio

90、n it churns out is still essential.Others may find that its most ef-fective use cases involve data analy-sis.It can interpret information,dissect text and numeric data,and even create charts.Combining this with its ability to generate code,it can be used for data analytics.Finally,it has tremendous

91、po-tential to assist with planning and project management.It can provide a step-by-step guide,including the tools and skills needed,the processes to put in place,and how to ana-lyse and assess your results.If you work in an area heavily dependent on one(or more)of these skills,you wouldnt be alone i

92、n wor-rying if youre likely to be soon re-placed by a machine.But no one should be immediately looking to move into something that will never be automated if such a thing ex-ists.Instead,it would be more re-warding to look at how you can use AI,and specifically natural lan-guage technology,to augmen

93、t your skills in the relevant areas.Writers should use AI to become more thor-ough and informed in their writing.Programmers can become more pro-ductive and efficient at creating code.Data analysts can use AI to find ways to look at their informa-tion and to process bigger datasets more quickly and

94、efficiently.But thinking beyond that,writers can become data analysts,to create copy thats more informed by facts and statistics.Data analysts can become writers,presenting their findings more engagingly and com-pletely.Programmers can become project managers,bringing together different skills to cr

95、eate more useful applications and the list goes on.ChatGPT,Bard or some future iteration of language-based AI will change many things about how we work.It might not happen right away,but anyone who wants to be part of this future has all the tools they need today to start taking steps in the right d

96、irection.In the same way as the internet,or the mechani-sation brought about by the industrial revolution it isnt going to go away.Raconteurs columnist Bernard Marr is a world-renowned futurist,influencer and thought leader on business and technology.He sets out the uses of ChatGPT and natural langu

97、age processing and why they wont make humans surplus to requirementsBernard MarrIf your audience comes to you for specialist knowledge,expert opinion or just because they likeyour personality AI-generated content is likely to leave them coldChatGPT could make you better at your jobWhere it is likely

98、 tobe useful is in automating some routine elements of our workA N A LY S I SF23%of businesses worldwide have adopted AI for natural language speech understanding as of December 2022have adopted AI for natural language generation18%OnlyMcKinsey,2022R A C O N T E U R.N E TF U T U R E O F D ATA&A I090

99、8Commercial featureCommercial featureOpen warfare:will data sharing win the fi ght against cybercriminals?Security teams are battening down the hatches against a barrage of coordinated cyberattacks.But without transparency and collaboration,are corporations fi ghting an uphill battle?rom battlegroun

100、ds to sport-ing fi elds,its often noted that the best defence is a good offence.The strategic weight of this well-worn adage holds fi rm for businesses looking to reinforce their cybersecurity.In the attack landscape,cybercrim-inals often join forces to disseminate sensitive information,share sophis

101、ti-cated tactics and expose corporate vulnerabilities.While intelligence can be harvested by attackers globally and weaponised against any sector at any scale,CTOs and CISOs are left putting out fi res individually rather than working together to proactively prevent them.Tony Meehan,vice president o

102、f engi-neering for security solutions at Elastic,believes that democratising data in the same way that cybercriminals do will keep businesses a step ahead.“Dont get me wrong,confi dentiality is still really important,”says Meehan.“Im not asking for every security team on planet Earth to go and post

103、all their detections on GitHub tomorrow.But we do need to fi nd ways to collaborate more openly and share knowledge,techniques,and best practices.”As infi ltrations become more pro-lifi c,coordinated and commod-itised,organisations cant afford to let cybersecurity skills gaps or out-dated defence st

104、rategies hamper their responses.“The attack surface has become way bigger.I dont know if we can make a dent in this problem with the same approach of the last 20 years,”Meehan continues.Meehan,who worked at the United States National Security Agency(NSA)for a decade on programs to col-lect foreign i

105、ntelligence,outlines three main problems facing todays defensive teams when defending their organisations.The fi rst is the speed of digi-tal transformation post-pandemic,which opened up holes due to busi-nesses accelerated transition to the cloud.The second is the growth of nation-state attacks,som

106、ething that wasnt a concern 10 or 15 years ago.And the third is talent scarcity in the security space which makes it harder for individual teams to keep up with new and emerging threats.“The goal of a good defence is to make the adversary work harder.I think the journey to achieving that really need

107、s to be built around an open community,”says Meehan.Elastics own search-powered solu-tions are built on this premise of open-ness,regardless of whether data lives on a single or multiple cloud setup or on-premise.The company has helped the likes of Adobe,BMW and Zurich Insurance fi nd what they need

108、 faster while keeping mission-critical appli-cations running smoothly and pro-tecting against cyber threats.Meehan appreciates C-suites may feel fear or scepticism over sharing sometimes sensitive information.But to fi ght off sophisticated attacks designed by malicious collectives,organisations mus

109、t achieve the same level of transparency as those trying to get in through the backdoor.Removing organisational data silos is one answer to deliver greater vis-ibility of what information is where when it is attacked.However,cor-porations must actively pursue new routes for collaboration if they wan

110、t to transform the preeminent cybersecurity culture.According to“When confronting the trends of the last couple of years,its para-mount to really understand what your products are doing for you,”he says.Additionally,data sharing to a far greater extent detailing threats,foiled attacks,and successful

111、 infi ltra-tions will empower teams.Meehan likens this to the success-ful sharing of YARA signatures,com-monly used to identify and detect malware.“Openness enables knowl-edge sharing,which will help elevate your team.You can even share spe-cifi c detection methods in smaller groups without ever exp

112、osing them to the world,”he says.Increased openness means the entire security community learns and grows.Meanwhile,the attack surface shrinks as it becomes harder for mali-cious actors to fi nd bypasses across multiple companies.Meehan explains:“All of our detections are in the open.Meehan,this clos

113、ed-off culture that prioritises privacy at all costs means companies rarely understand how their purchased vendor security products work;they just accept that they will.Thats an excellent starting point.Even if youre not using our product,you can still go and use our detections.”But security has tra

114、ditionally adopted a very closed culture,mean-ing potential vulnerabilities can go unexamined.At the same time,attackers could spend every day for months searching for gaps.Meehan accepts that few security vendors want to poke around in their own products because they dont want to be confronted by t

115、he holes they might discover.But this reluc-tance is evidence enough that the system is broken.“Its very hard to get people to spend that much time looking for these things,so we have to have a conversation around doing more things in the open,”he says.“If every-one is being a little more transpar-e

116、nt about their security controls,threat logic and detection rules,that becomes a force multiplier for all teams best practices.Not everyone has to start from scratch.”The combined efforts of partners and volunteers as part of the Shields Up initiative following Russias inva-sion of Ukraine is a prim

117、e exam-ple of collective defence in action.Openly sharing vital information has helped Ukraine become a cyber-security heavyweight.“Supporting one another is a natural reaction.In Ukraine,it was the obvious thing for us to do,”Meehan explains.While companies do share data,the practice is primarily r

118、elation-ship-driven and isnt as formalised or progressed as it should be.If the majority of organisations are solving the same problems at the same time,the system needs to be revised.“As an industry,we shouldnt be embarrassed about the fl aws we fi nd,but how long it takes us to fi x them and the l

119、ack of investment in fi nding more.We should want to fi nd fl aws,”says Meehan.The more comfortable companies are with internal scrutiny,the harder it becomes for outsiders to game the system.Appreciating the need for greater sharing will make a tangible differ-ence to the people and products that s

120、ystems are designed to protect.A new cybersecurity culture that pro-motes open information and close ranks will set organisations on a path to victory.For more information visitelastic.co/explore/security-without-limitsAs an industry,we shouldnt be embarrassed about the fl aws we fi nd,but how long

121、it takes us to fi x them and the lack of investment in fi nding more.We should want to fi nd fl awsFHow are data silos creating security challenges fororganisations?Since Covid-19,businesses have exploded into the cloud much faster than expected.In the rush to support remote working,companies began

122、pulling data from more appli-cations and sources than ever,which opened them up to exploitation.Its also much harder now to break down information and identify which por-tions are most critical to operations.Companies are fi nding they dont have the right expertise internally to under-stand or monit

123、or it all at scale.And the sheer amount of information they need to sift through can be overwhelming.When your systems are compro-mised,its not just about spotting the intrusion;thats only half the prob-lem.Preventing someone from being inside long enough to cause damage is crucial.Minimising the dw

124、ell time of your adversary is effectively a data access challenge.You may have seen an initial alert but cant gain access to the areas you need because the designated expert is away,the data isnt availa-ble to the analyst,or the data simply doesnt exist.Analysts cant connect the dots if theyre segre

125、gated from the data they need when they need it.Whats the next step for businesses to respond effectively to cyber threats?People silos are as tricky as data silos.You might have an end-point expert,a fi rewall expert,and an email expert-but they all work in isolation.Unifi ed visibility is the fi r

126、st step towards security,and that means embracing openness.With an open schema or framework,the power goes back into the hands of the customer as opposed to the vendor.You control your data and your rules,and you can freely switch out technology vendors as new prod-ucts emerge.Your analysts need to

127、be in a position to act quickly when a security incident happens.But the more silos there are,the longer that process takes,and the business risks greater exposure.Removing those restraints for your analysts can make a big difference to the damage toll at the end of the day.How can security teams im

128、prove their decision-making?Ive had many conversations where the breached business or organisation doesnt even know what happened or what was stolen.So how can your analysts make deci-sions about business risks if they dont understand what the data looks like and where it was when it was attacked?Bu

129、ilding that understanding and improving visibility are fundamental areas to invest in before you even think about the technology.If you dont understand where your data and assets are within an environ-ment,thats a big crack in your secu-rity foundations.You need to be able to perform root cause anal

130、ysis in real time.When an adversary attacks,its not over in seconds.There is a window where they make enough noise in the envi-ronment for security teams to detect and intercept the ultimate breach.Theyre always going to get in,but as long as you recognise that,you can really prepare.If silos are br

131、oken down,and security analysts know what data is where they can react fast enough to stop data from being destroyed or stolen.What advice do you have for CISOs to get ahead of data challenges?First,ensure your security oper-ations are not viewed by the rest of the organisation as a silo or as a tea

132、m that sits behind closed doors,only emerging to tell someone they made an error.Instead,talk with your business leaders consistently and reg-ularly to get to know their processes and requirements directly.When your team interacts with the rest of the business,you gain key insights that will acceler

133、ate response actions and improve explanations for alerts and requests coming from those teams.Second,own your own data.And make sure your security vendor does not lock you into their ecosystem.By insisting on open standards for data storage,data analysis,detection engi-neering and more,your teams ca

134、n be agile in adopting new technolo-gies and vendors to suit your security needs as they evolve.The data silo dilemmaQ&AVisibility is the fi rst step towards security,and that means embracing openness,explains Mike Nichols,vice president of security product management at ElasticWhen your systems are

135、 compromised,its not just about spotting the intrusion;thats only half the problemElastic,2022The data silo Visibility is the fi rst step towards security,and that means embracing Mike Nichols,vice president of security product of all organisations are investing in upskilling cybersecurity and IT st

136、aff46%of executives think that cyber risk initiatives have not kept pace with digital transformation41%R A C O N T E U R.N E TF U T U R E O F D ATA&A I1110Nearly all CEOs recognise that AI will become a significant factor inthe success of their firms over the medium term.As the market for AI-powered

137、 business tools develops,how will these be applied in various functions?Andhow have the early adopters benefited from using them so far?AI ACROSS THE BUSINESSPERCENTAGE OF RESPONDENTS IN SELECTED FUNCTIONS REPORTING COST DECREASES AND REVENUE INCREASES RESULTING FROM THEIR USE OF AIEACH FUNCTIONS MO

138、ST POPULAR APPLICATIONS FOR AI,BY THE PERCENTAGE OF RESPONDENTS CITING ITS USEMIT Technology Review Insights,2022McKinsey,2022“AI is vital to our organisation”Widespread adoptionLimited adoptionPiloting useNot usingITSupply chain/manufacturingProduct developmentHRFinanceMarketing and advertisingSale

139、s0%40%20%60%90%10%50%80%30%70%100%EXPECTED UPTAKE OF AI BY VARIOUS FUNCTIONS ACROSS INDUSTRIES WORLDWIDE IN 2025,BY PHASE OF ADOPTIONUPTAKE OF AI BY VARIOUS FUNCTIONS ACROSS INDUSTRIES WORLDWIDE IN 2022,BY PHASE OF ADOPTIONCost decreaseRevenue increaseMore than 10%5-10%Less than 5%More than 20%10-20

140、%Less than 10%Optimisation of workforce deploymentContact-centre automationService operationsFraud and debt analyticsMarketing and salesTreasury managementProduct/service developmentCustomer segmentationHRAI-based enhancement of productsSupply chain managementLogistics network optimisationRisk manag

141、ementStrategy and corporate finance29%25%3%21%4%30%27%41%41%16%8%28%20%4%33%31%8%4%24%13%6%17%14%7%5%11%10%8%5%41%20%9%3%1%31%13%14%37%10%10%10%6%Optimisation of talent managementService operations optimisationRisk modelling and analyticsCapital allocationCustomer service analyticsCreation of new AI

142、-based productsSales and demand forecasting24%10%19%15%10%20%7%16%5%19%11%9%19%4%of business leaders believe that AI will be critical to their organisations success over the next five years94%Deloitte,2022R A C O N T E U R.N E TF U T U R E O F D ATA&A I1312hen Arvind Narayanan gave a presentation at

143、 the Massa-chusetts Institute of Tech-nology called“How to recognize AI snake oil”in 2019,he was surprised to find his academic talk going viral on Twitter,with the slide deck even-tually downloaded tens of thousands of times and numerous requests fill-ing his inbox.The overwhelming response has sin

144、ce led Narayanan,assistant pro-fessor of computer science at Prince-ton University,to expand his talk into a book that he is co-writing with graduate student Sayash Kapoor.Following the sensation caused by ChatGPT and generative AI,thesub-ject of the book is clearly more time-ly than ever.But what i

145、s AI snake oil and just how would you distinguish it from the real thing?Narayanan explains that AI is an umbrella term for a set of loosely re-lated technologies without a precise definition.To help demystify the term,he has devised a scheme that classifies AI from genuine to dubious across three c

146、ategories:AI relating to perception,AI automating human judgement and predictive AI.The first category includes technol-ogies such as the song identification app Shazam,facial recognition,and speech-to-text.The second refers to AI used for making content recom-mendations,automating content moderatio

147、n in social media,or de-tecting spam or copyright violations online.The third refers to predictive AI systems in tasks from hiring to setting bail to gauging business risk.“The third category is really where most of the snake oil is and thats about using AI to predict what a per-son might do in the

148、future,”says Narayanan.“And then use that pre-diction to make decisions about them that might,in fact,give or deny them important life opportunities.”how companies collect and use peo-ples personal information.To limit the dangers of an AI free-for-all,Narayanan says government regulation will have

149、to play a role in ensuring new AI systems perform as advertised and dont abet discrimi-nation,disinformation or other harms.Indeed,governments are scrambling to figure out how to ad-dress the proliferation of AI across all aspects of society.But Narayanan emphasises that ex-isting laws,such as those

150、 dealing with discrimination or fraud,can al-ready be applied to problems emerg-ing from the rise of AI.In that vein,the US Federal Trade Commission re-cently issued a warning to business-es about exaggerating what AI products can do or whether they use AI at all.And since business as an institu-tio

151、n enjoys a measure of public trust,Narayanan says its especially im-portant companies dont over-promise what AI can deliver.“Unfortunately,when they overhype some of these technologies and con-fuse public discourse,theyre doing everyone a big disservice,”he says.generative AI into their operations.T

152、hat means starting with the sim-plest tasks to be automated for pro-ductivity gains,“then once you have experience where you start to under-stand the limitations,gradually build up from there to try the more complex tasks”.That approach might involve cus-tomising a general model as opposed to using

153、a smaller,specialised one.“The reason people are currently ex-cited is because they feel that foun-dation models are perhaps a more general and quicker way to get to business-specific objectives than to train a model on a particular data set,”says Narayanan.The recent release of the ChatGPT API by O

154、penAI is likely to spur the rush of companies and startups har-nessing the technology to add chat-bots or other AI-powered features to applications,so as not to get caught behind the curve.Narayanan praised Google for,in contrast,taking a cautious approach,significantly delaying the public re-lease

155、of its AI chatbot amid ethical considerations and internal debate.But in the wake of ChatGPT and Mi-crosofts Bing relaunch,the search giant is playing catch-up.It an-nounced its Bard chatbot in Febru-ary and is planning to include AI in all its major products within months,according to a Bloomberg r

156、eport.Welcome to the AI arms race.“A lot of the last five years of progress in re-sponsible AI is in fact eroding at this moment,”says Narayanan,who co-authored a textbook on machine learning and fairness.He also led the Princeton Web Transparency and Accountability Project,uncovering At the same ti

157、me,he highlights some of the flaws that have recently surfaced,most notably Microsofts new AI-powered Bing becoming er-ratic and telling lies in lengthy ex-changes with journalists and early testers.That suggests to him that generative AI wont necessarily upend search overnight.Narayanan,who has a l

158、ively Twit-ter account(random_walker),has also referred to ChatGPT itself as a“bullshit generator”.That isnt a sci-entific term.“I just wanted to remind people that chatbots arent trained to be accurate,”he explains.“Theyre trained to sound convincing,but fundamentally chatbots arent built with an a

159、bility to evaluate the truth or falsehood of statements.”As such,he suggests ChatGPT and its rivals shouldnt be viewed as trusted sources in areas where accu-racy is vital,like providing health in-formation.“I dont think that problem is fundamentally insoluble.A lot of researchers are working on it,

160、but its just not there yet,”he says.Within the business realm,Nara-yanan suggests that companies should move carefully to incorporate In a forthcoming book,Princeton computer scientist Arvind Narayanan aims to offer a clear-eyed corrective to the hype around AI.But he also sees promise in generative

161、 AIUnlike using AI for something such as speech transcription or image recognition,Narayanan ex-plains that there is no ground truth data,or gold standard,to compare and evaluate results with predictive AI because the outcomes havent happened yet.“The future is funda-mentally unpredictable,”he says.

162、Whether screening job candidates,predicting recidivism or the risk of a motor vehicle accident,Narayanans research has found purported AI tools fare little better than flipping a coin.Certainly,they are not as effective as long-established statis-tical analysis methods such as regression analysis.Wh

163、ere,then,does leave ChatGPT?Narayanan views generative AI,in-cluding ChatGPT,as an outgrowth of perception-related AI,going be-yond just perceiving and classifying content to being able to generate im-ages or text on request.Through such progress,he believes genera-tive AI holds more promise than as

164、 a substitute for human judgement or discerning the future.“The potential is clearly there but a lot of work still lies ahead to figure out which applications are even the right ones,”he says.In that vein,Narayanan points to AI tools he uses himself,such as GitHub Copilot,which can turn natural lang

165、uage prompts into code and translate code between programming languages.Mark WalshChatbots arent built with the ability to evaluate the truth or falsehood of statementsThe potential is clear but a lot of work lies ahead to figure out which applications are even the right onesHow far can wereally go

166、with AI?I N T E R V I E WCommercial featureCreating a sat nav for your dataBusinesses are creating valuable data but all too often it lies undiscovered,meaning it cannot be connected to other systems or used to drive insight,with this process duplicated time and again.Aiimis AI-powered Insight Engin

167、e helps by discovering and interconnecting information that informs business decisionsor a financial services regula-tor,trying to stop fraud can feel like a game of whack-a-mole.The moment a dubious operator is prosecuted,they often reappear under a new name and go back to breaking laws.But what if

168、 AI and machine learning made it possible to follow actual indi-viduals rather than the paper record?That,instead of a row in a spreadsheet,offenders could be traced by their online behaviour?By applying machine learning-based entity recognition,the financial regu-lator can see who really profits fr

169、om a business.Instead of bad actors being able to simply relaunch companies under a new name,the regulator can use graph technology to show com-monality between companies and individuals by studying transactions,flow of money or people who work for them.This exposes whole networks of potential breac

170、hes,says Paul Maker,CTO at Aiimi,an AI firm that helps busi-nesses create a data mesh that ena-bles factors to be linked and watched in real time.The system does not require exper-tise from staff,meaning it can help businesses find commonality within their systems,for example by spotting manufacturi

171、ng issues before they arise.Applying machine learning and natural language technology(NLT)can identify patterns,topics or words within a system;cluster and classify types of information;differentiate between quirky synonyms and meta-phors;or mark sensitive information as secret,for instance.“A busin

172、ess might have five safety incidents caused by weak components,part failures or human error,all of a similar nature.With NLT and graph technology,this problem can be made visible,exposing common part failings and identifying problem parts or spe-cific issues with that equipment or staff.Armed with t

173、his knowledge,a business can also anticipate wear and tear in advance.It finds themes and future learnings with every piece of data you input.”Aiimis Insight Engine gives workers of all abilities a heads-up display,offer-ing information and help from past learnings to enable them to do their job,say

174、s Steve Salvin,Aiimis chief executive.“Just as Google Maps sends cars around the world,taking images of every street and mapping them out,we create a satellite navigation system for your data and business.”Building a data meshAiimi describes this system as a data mesh,where departments that may not

175、work together can still see informa-tion in real time because data learnings are constantly made available,depend-ing on need.One customer,a major global man-ufacturer,accrued half a billion indi-vidual pieces of data.But too much data can become a problem,for example when designing a complex asset,

176、elements can be created in isolation that,when married together,become incompatible.In one example,insight from earlier design discussions was missed,leading to a product recall.The key to avoiding this is the ability to surface this infor-mation to the right people during the design phase.Graph tec

177、hnology allows Aiimi to find and connect subject matter experts,and build knowledge networks of topics,phrases and people that avoids re-work and speeds up time to market,says Salvin.“These pieces of information are locked away within departmental data systems,such as product design,test-ing,manufac

178、turing and sales,leading to an inability to ask questions about the data and get the answers needed,”he adds.“These data silos are labelled differently and have different permissions structures,which make it impossible to build an interconnected view of the enterprise.“With a data mesh approach wher

179、e all departmental data systems have been labelled consistently,we can interpret and connect disparate terms,allowing staff to get to their answers and be right the first time.”This wholly transparent system,which constantly flags up commonali-ties and explores existing resources,can also be applied

180、 outside a firms ecosystem.For example,a major sup-plier of UK passenger and freight rolling stock relies on knowing intricate infor-mation about the national rail network and future planning.Machine learning gives the firm an index of all the public,open-source meetings of all the councils across t

181、he UK as they happen,explains Maker.“AI can quickly find all the people responsible for budgeting,closing,reopening or commissioning any train station across the UK and the out-comes of any meetings as they happen,”he says.More interestingly,in doing this,it is possible to create a knowledge net-wor

182、k connecting figures and solutions that go beyond one individual,he says:“Knowledge networks and contact books exist in peoples heads.But when a person retires or moves to another job,they take that knowledge with them.Who is the best person to sell to?To advise on a specific issue?You can persist a

183、nd repopulate that knowledge network and reuse it even after people are gone.”Using augmented reality to apply rigourAny traveller with a smartphone will be automatically offered their boarding pass when they enter an airport,yet we do not apply this same accessibility to the world of work,says Salv

184、in.For example,it would be useful for an engi-neer to be able to see a passport of a machines history when they come to fix it,which could include accident history and the current temperature of the machine,as well as on-site or open-source details on how that part is per-forming inside or outside t

185、he business.“An engineer might be standing in a field about to fix a part,”says Maker.“But with a pair of smart glasses,they have access to a whole new dimension.You can deliver the right information for that location,the entire history of the part,recent maintenance and also know their daily calend

186、ar.Then you also have all the geotagging and geolocation input,so that days work is recorded for future use automatically.”In an Aiimi business study across one of the UKs largest water compa-nies,it was discovered that around 75%of the data created by staff was never used again.This showed that peo

187、ple were not able to find and reuse information that already exists,and were forced to recreate and relearn corporate knowledge.People use a sat nav,even for short and familiar journeys,because it con-stantly investigates to find improve-ments to the route,for example by finding traffic congestion a

188、nd road closures,says Salvin.The Aiimi Insight Engine similarly makes getting to an answer easier and gives business more confidence.For more information visit Just as Google Maps sends cars around the world,taking images of every street and mapping them out,we create a satellite navigation system f

189、or your data and businessFWPrinceton University,2022329the number of academic papers on machine learning across 17 fields that Narayanan and his team have found to suffer from reproducibility failures or pitfalls in ML-based scienceR A C O N T E U R.N E TF U T U R E O F D ATA&A I1514With no one step

190、ping in to lead on AI and its ethics,this has in essence become an arms raceIts widely acknowledged that there are very human biases baked into many machine learning models.Butwhat are those with the power to solve the problem doing about it?and peer reviews,in which teams with no prior knowledge of

191、 the data analyse the code and the results.“Machine learning models that banks are using in areas such as fraud detection have too often been based on relatively small samples,”says Ian Liddicoat,CTO and head of data science at Adludio,an AI-powered advertising platform.“To achieve more accurate res

192、ults,syn-thetic data can be used to mimic and augment the original data sets.This data also seeks to even out the distributions for factors such as gender to ensure that they reflect society more accurately.”Data engineers are also using more advanced random-sampling methods that create data sets fo

193、r modelling,where each record has an equal chance of selection.Sup-ervised learning methods,mean-while,can be applied to neural networks,so that real-world distri-butions for factors such as race or gender are enforced on the model.“Another effective solution”,adds Liddicoat,“is to use sub-teams to

194、conduct detailed reviews of the input data,the machine learning method,the training results and the operational outcomes.”Beyond the technical challenges of removing bias from the systems and the data theyre based on lies the bigger issue of improving diversity in the AI sector.The only truly eff-ec

195、tive way to achieve this is for its employers to recruit from a wider talent pool.The ongoing lack of diversity in academia,particularly in STEM subjects,is troublesome for the future of AI.Until that fun-damental problem is solved,AI may always pose a threat to diversity,equity and inclusivity.for

196、system development and dep-loyment that often feature commit-ments to improving inclusivity.Despite this,some experts argue that little coordinated progress hasbeen achieved on establishing ethical norms for AI,particularly in relation to diversity.Justin Geldof,technology direc-tor at the Newton Eu

197、rope consul-tancy,is one of them.He argues that“governments have abdicated responsibility for this to the tech in-dustry and its consumers.There is no formal watchdog and no general agreement on norms in AI ethics.With no one stepping in to lead on AI and its ethics,this has in essence become an arm

198、s race.”High-profile breakthroughs in the field of generative AI recently have also done little to address concerns about discrimination.In fact,theres a risk that the poten-tial harms of generative systems have been forgotten in all the media hype surrounding the power of OpenAIs ChatGPT chatbot an

199、d its ilk,according to Will Williams,vice-president of machine learn-ing at Speechmatics.Meta took Galactica,its large lan-guage model,offline in November only three days after launch,for ny AI system,however so-phisticated,is only as good as the data on which it is trained.Any bias in its outputs w

200、ill result from distortions inthe mate-rial that humans have chosen to feed into its algorithms.While such biases are unintentional,the AI field has a predominantly white male workforce creating products that will inevitably reflect that de-mographics particular prejudices.Facial recognition systems

201、,for instance,could be inadvertently trained to recognise white people more easily than Black people be-cause physical data on the former tends to be used more often in the training process.The results can put groups that have traditionally faced marginalisation at a further disadvantage,heightening

202、 barriers to diversity,equity and inclusivity in areas ranging from healthcare provision to recruitment.The good news is that the prob-lem has been widely acknowledged in business,academia and gov-ernment.Efforts are being made to make AI more open,accessible and balanced.There is also a new ethical

203、 focus in the tech industry,with giants including Microsoft and Google establishing principles Alison ColemanIn simple terms,generative AI models average the opinion of the whole internet and then fine-tune that via a process known as rein-forcement learning from human preferences.They then present

204、that view as the truth in an overly con-fident way.“This might feel like safe AI if you are one of the humans fine-tuning and editing that model,”Williams adds.“But,if your voice isnt represented in the editing room,youll start noticing how your opinion varies dramatically from those of ChatGPT,Clau

205、de and Bard.Your truth might be some distance from the truth they present.”The race to produce a winner in the generative stakes has given new urgency to addressing bias andhighlighting the importance of responsible AI.Emer Dolan is president of enter-prise internationalisation at RWS Group,a provid

206、er of technology-enabled language services.She saysthat,while the detection and removal of bias is“not a perfect sci-ence,many companies are tackling this challenge using an iterative process of sourcing targeted data to address specific biases over time.As an industry,its our duty to edu-cate peopl

207、e about how their data is being used to train generative AI.The responsibility lies not only with the firms that build the models but also with those that supply the data on which theyre trained.”Wider technical,analytical and academic communities are applying several methods to reduce or even remov

208、e bias.These include super-vised learning;synthetic data sets that contain computer-generated ma t erial instead of real-world data;Discrimination game:time to scrap the skew instance,amid fears about its inac-curacy and potentially dangerous impact on society.Williams says:“The truth is that the in

209、herent bias within models such as ChatGPT,Anthropics Claude and Googles Bard means that they cannot be deployed in any business where accuracy and trust matter.In reality,the commercial applications for these new tech-nologies are few and far between.”NEARLY THREE-QUARTERS OF FIRMS ARE FAILING TO RE

210、DUCE BIASES IN THEIR AI SOLUTIONSShare of organisations that are falling foul of AI safeguarding in the following ways IMB,2022I N C L U S I V I T YACommercial featurehen football commentator Alan Smiths words echoed out,“Messi shoots-its a goal!”as Argentinas star player slot-ted home a penalty aga

211、inst France in Qatar last December,fans listening in on Veritones YouTube channel may have assumed Smith was reacting to the match live.What listeners actually heard was Alans AI-generated voice clone pro-viding real-time commentary from the live matchs analytical data.The pro-ject,developed by Stat

212、s Perform and powered by Veritones AI technology,lets fans listen to live game updates from a professional commentator on any device in local languages.Its hard to fathom,but this is just one of the many ways that generative AI is helping media,entertainment and other industries create more engaging

213、 content that can connect with audi-ences anywhere in the world.Generative AI also allows companies to more effectively scale content to engage increasingly diverse,multina-tional audiences.For those looking to further monetise content distribution,its a significant opportunity.Powering the creative

214、 engine Voice cloning offers several applications customer solutions at Veritone.“If they are deceased,you need permission from their estate and the IP owner of the voice training data to ensure that it has the rights to reproduce the voice.”There are two main ways that AI can generate cloned speech

215、:text-to-speech or speech-to-speech.“Using a speech-to-speech model where you speak in,and their voice comes out,youre capturing the intonation,speed,emotion.Its frighteningly accurate,”says Candler.Voice cloning can even allow com-panies to create new advertisements by brand ambassadors in multiple

216、 lan-guages without them having to partic-ipate in additional recording sessions.“You can transcribe a campaign,translate it into multiple languages and create additional ads and assets at a scale that we cant do on our own as humans,”says Bailey.“So,it goes way beyond having generative AI help you

217、with ad copy to actually creating campaigns.”Scaling audience reach and revenue growth Given the vast content archives that most companies possess,AI helps them discover and enhance their exist-ing assets to expand both audience acquisition and revenue growth.Content continues to grow expo-nentially

218、.More content is uploaded to digital platforms in 30 days than what the major US TV networks broadcasted over the previous 30 years.In this cli-mate,organisations that dont adopt AI to assist in optimising their extensive archives could be leaving bigger audi-ences(and money)behind.According to Acce

219、nture,nearly a third of AI pilot initiatives are scaled beyond their initial scope to deliver outcomes across the business.42%of those surveyed determined that the return on their AI initiatives exceeded their expectations last year.Savvy investments now can bring recurring value,giving businesses m

220、ore mileage out of their content and streamlining their operations by har-nessing AI to replace time-consuming manual tasks.“For starters,media companies can use AI voice models to produce for-eign language versions of their back catalogues,helping expand their dis-tribution footprint globally,”Cand

221、ler explains.And AI algorithms can also help them search through their archives faster,making it easier to find monitisable content.“A media and entertainment busi-ness may have millions of assets that it can potentially monetise,but unless you can find them,you cant activate them,”says Candler.“Wit

222、h Veritones AI-based archive licensing service,for example,we can very quickly lift spe-cific moments from vast archives and then sell them to documentary film-makers,agencies or networks.It takes far too long to manually sort through the content to find what you are looking for.”Working side-by-sid

223、e with human creators Using generative AI,a large language model like ChatGPT can be built for a specific brand,which can then deliver relevant content and ideas.“All of the generative AI components would be very specific to their domain,”says Bailey.“It can take all the data of a content rights hol

224、der and provide content recommendations and cam-paign recommendations on the back of that.”Although some content producers may be fearful of AI replacing them,these innovations are designed to support humans in their roles.With generative AI content,humans will still need to go in and edit what has

225、been produced.“For our voice work,we partner very closely with the voiceover commu-nity.Its not about taking jobs awaywe augment human ability,”says Bailey.“It helps content creators keep up with content demand,allowing them to connect with global audiences in a more authentic,personalised way.”As p

226、eople and machines collabo-rate more effectively,generative AI can create operational efficiencies,improve audience experiences,and make new revenue streams available at scale.The message is clear:AI doesnt make better content;it makes content better.For more information visit content creators.It ca

227、n be used to narrate scenes or correct audio in post-production without the actor coming into a studio to record it themselves.It also has the poten-tial to transform the dubbing and translation industries.“You can use it for dubbing in an actors unique voice in different lan-guages,”says Ashley Bai

228、ley,direc-tor of product marketing for AI voice at Veritone.“You could have Kevin Costners voice,the way he speaks,his inflection,his tone,but in Spanish,Italian,French or whatever language to bring a more authentic experience to audiences.”The same can be said for podcasts,where AI can open up new

229、markets by transposing a hosts voice into differ-ent languages.While AI can clone any voice,doing it ethically is critical to the mediums success.A Danish production com-pany recently worked with Veritone on a documentary about a reality TV star that had passed away.They wanted his voice to narrate

230、the programme,high-lighting the legal issues around this application of AI.“You absolutely need to get sign-off from the person youre cloning,”says David Candler,senior director of How AI-generated content raises revenues and connects audiences Artificial intelligence can complete many tasks,from cl

231、oning voices to generating full-scale ad campaigns that boost revenue and engagement.But for creatives,is it an ally or adversary?A media and entertainment business may have millions of assets that it can potentially monetise,but unless you can find them,you cant activate themWNot reducing unintende

232、d biasNot tracking performance variations and model driftNot ensuring explainability of AI-powered decisionsNot developing ethical AI policies internallyNot tracking data provenance,changes in data and model variationsNot guarding against adversarial threats and potential incursions to keep systems

233、healthyNot monitoring AI across cloud and AI environmentsNot safeguarding data privacy through the entire life cycle74%61%60%55%68%60%59%52%Mads Perch via Getty ImagesR A C O N T E U R.N E TF U T U R E O F D ATA&A I1716AI is helping pharmaceutical firms to drastically reduce the time and cost they i

234、ncur in the lab on evaluating the medicinal potential of hundreds and thousands of chemicalseveloping pharmaceuticals is a notoriously lengthy and expensive business.The median cost of bringing a new drug to market is 810m,according to research led by the London School of Economics,and its not uncom

235、mon for the process to take more than a decade from start to finish.Whats more,the failure rate is about 96%.Much of this attrition occurs towards the final phases of the R&D process during safety testing nine out of 10 drugs in development dont make it through human clinical tri-als.But even the ea

236、rly stages of iden-tifying a potential therapy thats worthy of investigation are highly complex and time-consuming.So-called drug discovery involves first identifying a molecular path-way or genetic variation thats linked to a particular disease and then finding a molecule that can interact with it

237、to halt,or at least delay,the progression of that dis-ease.Until recently,this was a painstaking process of trial and error,with firms typically creating and screening several hundred thousand compounds in vitro to whittle them down to a couple of options deemed suitable for clinical testing.But in

238、recent years the industry has started using machine learning to model the likely efficacy of these chemicals in silico before doing any work on them in the lab.The technology can evaluate the potential of candidate molecules against several hundred criteria simultaneously,sifting out those that show

239、 little promise while highlighting those worthy of further attention.“End-to-end artificial intelligence allows us to discover and deliver better medicines faster than humans can alone,”says professor Andrew Hopkins,CEO of biotech firm Exscientia.“AI platforms make predictions based on their analyse

240、s of thousands of parameters in paral-lel,exploring a computational space far beyond our cognitive powers.”Some of the first AI-recommended chemicals to enter human clinical trials were discovered in a process that took just over a year on average,he adds.Their development involved fewer than 10%of

241、the pro-totype compounds that would have to discover novel molecular mecha-nisms that we believe can cause ALS.We achieved all this in half of the industrys standard time.”There have been other break-throughs elsewhere in this field.For instance,computer scientists at the University of Sheffield rec

242、ently worked with AstraZeneca to develop a new AI platform that,they say,provides enriched information about the interactions between drugs and their protein targets through a so-called bilinear atten-tion network.The team has made the source code freely available to other researchers.The lead resea

243、rcher on this project is Haiping Lu,professor of machine learning at Sheffield.He explains that“our AI,called DrugBAN,learns the multiple pairwise interactions between the substructures of drugs and their protein targets to provide How in silico testing is accelerating drug R&DEmma Woollacottbeen te

244、sted using the traditional approach to drug discovery.In another recent first,a new AI-aided drug designed to treat amyotrophic lateral sclerosis(ALS),a motor neurone disease for which there is no known cure,has entered clinical trials.The therapy going under the name VRG50635 at this stage was disc

245、overed by Verge Genomics using Converge,an AI platform it has developed.“We used Converge to build an ALS disease signature based on more than 11 million data points,sourced from almost 1,000 human tissue samples,”explains the firms chief business officer,Dr Jane Rhodes.“Our signature comprises more

246、 than 200 genes that are dysregulated compared with neurons from healthy brains and spinal cords.This gave us insights into the com-plexity of the disease and enabled us End-to-end artificial intelligence allows us to discover and deliver better medicines faster than humans can aloneP H A R M A C E

247、U T I C A L Suseful biological insights.These help researchers to understand whats happening at a molecular level.DrugBAN contrasts with most drug-prediction AI,which learns from whole representations of drugs and proteins rather than their substructures.”The use of AI as a drug discovery tool incre

248、ased by 40%in 2022,according to research published in Nature in February,and uptake is only likely to grow over the next few years.McKinsey estimates that almost 270 firms are already working on AI-powered drug discovery,mostly in the US,and predicts that many of them will seek to establish partners

249、hips with well established biopharma companies.“The next challenge we face is to scale up this approach,”Rhodes says.“We need to systematically apply the use of AI across the entire drug discovery pipeline and embed it effectively into business processes.By using large quanti-ties of high-quality hu

250、man data,we can reduce the need for animal experimentation and move towards a system of predictive modelling that can save both time and cost while increasing the probability of technical success.”Some organisations are introduc-ing AI into other aspects of the drug R&D process.For instance,pharma c

251、ompanies are using natural-lan-guage processing to sift through the vast archive of scientific literature,including academic papers and already-identified gene sequences,to help them detect and analyse pat-terns that could enable them to identify potential drug targets that might otherwise be missed

252、.And other AI-based tech,when fed with the right data,can not only help users to identify candidate drugs;it can also help them to predict which patients are likely to benefit most from such medicines.This will potentially lead to better-targeted,and there-fore more effective,therapies.In 2021,for i

253、nstance,Exscientia demonstrated for the first time that an AI-driven precision medicine platform could propose which drugs would be most beneficial for people with late-stage blood cancers.Such techniques can also help to optimise dosages for individual patients and even establish the best times of

254、the day for them to take their medicine.“In the future,such AI techniques will enable the development of per-sonalised medicines,including vac-cines,”Lu predicts.“By incorporat-ing multimodal data including a patients genetic information and medical history in AI modelling,we can reduce the risk of

255、adverse reac-tions and increase the likelihood of a successful treatment.”The future success of AI-based drug development obviously depends greatly on both the quality and the quantity of data that is available to be crunched.But the increasing use of electronic medical records by health services is

256、 helping in this respect.“We believe that using AI to design and develop better drugs faster will eventually become the model to create new medicines,”says Hopkins.“Within a dec-ade,most drugs will be discovered and developed this way.”Morgan Stanley Research,2022AI IN R&DAI spend as a share of biop

257、harmaceutical R&D Using AI to design and develop better drugs faster will eventually become the model for creating new medicinesCommercial featuref data is now the lifeblood of any company,it must flow seamlessly and coherently to keep a business healthy and thriving.From helping with future projec-

258、tions to delivering on C-suite ambi-tions,good-quality and real-time data creates a solid foundation for deci-sion-making.However,siloed data sets,many of which are often stale and incomplete,will slow down the oppor-tunity for making critical choices fast.Steve Mulholland,regional vice-pres-ident E

259、MEA at Fivetran,believes auto-mated data movement offers one answer.“CEOs and executive business leaders will have a greater guarantee that only the freshest data is being used and that its being delivered to the right people at the right time with-out cumbersome manual processes,”he argues.This is

260、important because according to Fivetrans own 2022 research,about eight in 10 organisations admitted to making decisions based on stale data.And inaccurate insights can be costly.Mulholland warns:“Businesses are losing an average of 5%of global annual revenue due to underperforming AI programmes buil

261、t on bad data.”Saving time,money and talentFivetran sits within the data stack turn-ing raw data into usable insights so leaders have confidence in the deci-sion-making.One company using its solution generated more than 300,000 new leads and saved 200 hours per month.Another tackled its disparate da

262、ta silos,which were forcing teams to manually deduplicate and refresh data Automating your data for better decision-makingInstead of relying on inaccurate,messy and low-quality data,business leaders can instead drive innovation forward by turning to automated data management,creating a clean and ste

263、ady data flowmultiple times a day.The technology reduced data-refresh times dramati-cally,from 53 minutes to just six.Mulholland points out that main-taining data pipelines manually is time consuming:“It has a knock-on effect throughout the organisation.Data scientists are spending as much as three-

264、quarters of their time preparing data,rather than building AI models.”And,he emphasises,this critical resource issue cannot be tackled by hiring given the current skills and talent shortage.Mulholland cites Fivetran research showing 87%of businesses acknowledge data scientists within their organisat

265、ions are not being used to their full potential.“The immedi-ate solution must be to outsource and automate mundane tasks,”he adds.A single source of truthEstablishing a“single source of truth”through data should now be top of the C-suite agenda,argues Mulholland.“It holds the key to operational effi

266、-ciency,customer success,employee satisfaction and business growth”and allows users across the business to“self-serve and innovate on behalf of the end customer.”Building trust is also key,and this means ensuring what is fed into the algorithms for automation is clean data checked for accuracy and q

267、uality.Mulholland highlights how concerns about under-performing AI models are usually due to the use of inaccurate or low-quality data.Robust,automated data flows can underpin these models to instil greater confidence in AI-driven decision making across the entire business.Many leaders though may s

268、till be sceptical about the cost and time involved,but Mulholland notes that Fivetran is plug and play,with more than 300 pre-built data connec-tors giving immediate solutions for resource-intensive tasks.What once took internal teams months to build is now available in minutes,with-out costly maint

269、enance charges and with enhanced security controls and in-built compliance for data legislation.“Businesses will only reap the rewards from their AI capabilities when internal operations are geared towards making the most of data and making data accessible,”says MulhollandThis,he explains,is achieve

270、d by cen-tralising in the cloud.“Decision makers can examine the total business impact of product and marketing changes,seeing trends that might not be obvi-ous from simply looking at profit and loss statements.”“Fivetrans business ethos is to make access to data as simple and reliable as electricit

271、y;its like flicking a switch on innovation,”he adds.Learn more about how automated data movement can propel your business forward by visiting INTELLIGENCE MAY BE THE FUTURE,BUT ORGANISATIONS HAVE A LONG WAY TO GOINEFFICIENT DATA PROCESSES CURTAIL ORGANISATIONS AI PROGRESSDATA SCIENTISTS TALENT IS BE

272、ING WASTED87%87%69%86%70%71%90%73%of organisations say Al is vital for business survival of organisations say they are not utilising data scientists to their full potentialof organisations struggle to access the right data at the right timewould struggle to fully trust Al to make all business decisi

273、ons without human interventionof organisations continue to rely on manual data processes struggle to translate data insights into practical advice for decision-makers struggle to access all the data needed to run Al programmes,workloads and modelsthe average share of time that data scientists spend

274、preparing data instead of building Al modelsVanson Bourne,2022DAndrew Brookes via Getty Images0%0.5%1.0%1.5%2.0%2.5%3.0%3.5%202120222023202420252027202820292030R A C O N T E U R.N E TF U T U R E O F D ATA&A I1918People are happy when they get what they want.But theyre happier when its a human that g

275、ave them the news and not an algorithmCompanies are rapidly adopting AI to predict short-term consumer behaviour and maximise profits.But businesses could use the technology to take a long-term view of behavioural analytics and set customer-centric goals As AI takes over more of our busi-ness intera

276、ctions and decision-mak-ing,its inevitable to think about the ethics around its use.In the PwC survey,98%of respondents said they plan to make their AI responsi-ble but fewer than half have planned to take specific actions.David Wright is a partner in Deloittes intelligent automation team and works

277、with clients in the private sector.He thinks the rule of governance will be key as business-es rely on data to anticipate consum-er behaviour.He points out that the“rule of governance is incredibly im-portant,particularly if you start to automate off the back of AI predic-tions or recommendations”.W

278、hile internal watchdogs are already present in highly regulated industries such as banking,insur-ance and healthcare,theyre less so in hospitality and retail.Loeshelle thinks this is about to change.“Were seeing it right now primar-ily in the States but its starting to bubble up in the UK companies

279、insist on putting our technology through model risk management reviews,”she explains.“This is an internal AI governance board that is responsible for evaluating the integ-rity and the purpose of each AI-re-lated tech that the company is pur-chasing or building.”Theres no doubt that AI will play a ke

280、y role in how businesses and con-sumers interact in the future.But if companies are attempting to predict consumer behaviour with AI,its important to understand the impli-cations for consumer experience and the safeguards that must be in place from the outset.This is just the starting point of har-n

281、essing the power of AI to drive reve-nue.A proliferation of generative-AI tools already process customer feed-back in real time and adjust their mes-saging to elicit the best response.As the technology develops,how can businesses make the best use of AI without inadvertently hurting their business o

282、r their customers?Sulabh Soral is chief AI officer at Deloitte Consulting and leads the Deloitte AI Institute in the UK.He makes a clear distinction between AI and what most corporates are using today:“The broadest defini-tion of AI is anything that can create rules that mimic things that we would i

283、dentify as cognitive capabili-ties.But,if you look strictly at the question of behavioural analytics,a lot of companies still use classical machine learning.”This may sound like a distinction without a difference,but its key in understanding the limitations of the behavioural-analytics models in ts

284、been over a decade since Amazon filed a patent for its anticipatory shipping technology(shipping an item before the customer knows they want it).It may sound dystopian,but the ecom-merce giant isnt relying on a crystal ball or guesswork.Instead,it has looked at historical buying patterns,browsing ha

285、bits,surveys and demo-graphic data to predict what items will be in demand where.The result?Hard-to-beat prices and best-in-class delivery times.Since then,AI and predictive ana-lytics have taken centre stage with businesses that increasingly use AI to drive decision-making.The PwC 2022 AI Business

286、Survey reports that 96%of business leaders who responded said that they intend to use AI simulations to improve busi-ness performance.The AI leaders among them use AI to drive deci-sion-making on technology(74%),operations and maintenance(62%)and customer experience(61%).Natasha SerafimovskaGenerati

287、ng new revenue isnt the only way that behavioural analytics could contribute to business growth.Ellen Loeshelle,director of product management at Qualtrics,an expe-rience management platform,high-lights that behavioural analysis can be just as valuable to protect exist-ing revenue as it is to genera

288、te it.“Our product takes behavioural data from the past and projects it onto things in the future.If you can tell us that this cohort of customers filled out a survey beforehand and then they quit,we can then use that as training data to project for future surveys that come in.”These insights can th

289、en help busi-nesses take preventative measures and ensure customer concerns are addressed in time.A new tool seems to emerge every day with the enticing promise that it can help you work better,faster and smarter.But Stefano Puntoni,a pro-fessor of marketing at The Wharton School at University of Pe

290、nnsylva-nia,warns that automation isnt always the right answer to drive business performance:“Its impor-tant to understand the role that technology plays in peoples lives to make predictions about how theyre going to react to it.”For example,when the items we buy are closely linked to who we are as

291、a person,automation can back-fire.Think of a cooking aficionado who loves spending time in the kitchen.If you suddenly automate part of the cooking process,you have then created a problem,instead of a solution.Another scenario where automa-tion is less effective than human input,Puntoni says,is in i

292、nterac-tions where customers are“as-sessed”by your business(for exam-ple,loan or credit card applications).Here,companies usually worry about how automation will handle negative outcomes for the customer,but Puntoni emphasises that its the positive outcomes which require human interaction.“People ar

293、e happy when they get what they want,but theyre happier when its a human that gave them the news and not an algorithm,”he says.“When it comes to rejections,however,we dont observe that being rejected by an algorithm or a human makes a difference.”The long viewuse today.While deep learning and genera

294、tive-AI models could gener-ate insights from unstructured data like language and photographs,machine-learning models rely purely on structured data to make predictions.This potentially leaves a treasure trove of data unexplored.This explains why most businesses today focus on short-term goals like r

295、evenue growth and cost reduction.But Soral believes that,as these tech-nologies evolve,business leaders need to move away from this way of thinking and focus on long-term,customer-centric objectives instead.“Most behavioural analytics today is based on how a response should be elicited,not whether t

296、he customer needs the product.But how would you train your AI model if the goal is customer happiness or the savings they could make in a decade?”PREDICTING CUSTOMER BEHAVIOUR IS THE NUMBER-ONE USE FOR AI AMONG MARKETERSLeading reasons for marketers to use AIB E H AV I O U R A L A N A LY T I C SIMar

297、ketingCharts,2023Commercial featurehe AI era is upon us,with what seems like new advances every week,pushing the technology to new heights.Between Google,OpenAI,Microsoft and a raft of other companies,new developments that can ease the way we live and work are accessible to people more than ever bef

298、ore.Its little wonder,then,that businesses are starting to con-sider how best to integrate AI into their processes to reap the benefits.But thinking before acting is vital in such a fast-moving space.The first-mover advantage that businesses seek out can quickly be negated by the reg-ulatory risks o

299、f irresponsible use of AI.“Lots of companies talk about AI,but only a few of them can talk about responsible AI,”says Vikash Khatri,senior vice-president for artificial intelligence at Afiniti,which provides AI that pairs customers and con-tact-centre agents based on how well they are likely to inte

300、ract.“Yet,its vital that responsibility be front of mind when considering any deployment of AI the risks of not considering that are too great.”Think fast,act slowerIn part,the fast moving and competi-tive environment often places the responsible use of AI secondary to gaining market share.The histo

301、ry of AI,says Khatri,has seen companies develop tools that harness the power of AI by making use of big data sets without fully considering what impact they can have on society.Widely used AI tools are trained by trawling the internet and gleaning information from what is found online,which can ofte

302、n replicate and amplify our societal biases.Another problem with AI gener-ated content is that it is often ill-suited to the specific needs businesses may have when deploying AI.“If Im a broadband provider in the UK,as opposed to a health insurance company in the US,theres a specific way that I comm

303、unicate with my cus-tomer,”says Khatri.“With respect to the generative AI technology thats receiving so much attention,its important that the AI models being used are trained on the companys own data,rather than relying solely on generic,third-party data.That way,the organisation remains compliant w

304、ith global data regulation and the AI models generate content that aligns with the companys unique approach to its customers.”Khatri points to how a customer service chatbot trained on the way users interact with one another on social media,for instance,could quickly turn quite poisonous rather than

305、 supportive,lobbing insults rather than offering advice.“At Afiniti,we use responsible AI design to make those moments of human connection more valuable,”says Khatri.“That in turn produces better outcomes for customers,cus-tomer service agents and companies alike.One way we do this is by training ou

306、r AI models only with the data we need,and we continuously monitor them so our customers and their cus-tomers get the results they want,while being protected from bias or other dis-criminatory outcomes.”Its not just the risk of alienating cus-tomers that should be at the forefront of a business lead

307、ers mind when con-sidering how to roll out AI within their organisation and to their clients.Regulation is on the horizon for AI,and is likely to bring specific requirements for how data is fed into models that are used to give AI its brain,and how AI is used to handle customer interactions.Caution

308、avoids consequences“Before you even start to develop or deploy AI,you must be cognisant of the regulatory landscape,”says Kristin Johnston,associate general counsel for artificial intelligence,privacy and security at Afiniti.“This means examin-ing your governance structure around data compliance to

309、get your house in order first.”AI regulation is complex and con-stantly changing,and a patchwork of laws across the globe can make it hard for businesses to comply.For exam-ple,businesses operating in Europe have different requirements from those with customers in the US,while the UKs data protectio

310、n regulation is likely to soon diverge from the European Unions.The magnitude of the task in respon-sibly deploying AI is something most businesses have yet to fully wrap their heads around,fears Johnston.“A lot of companies havent built out a gov-ernance process specifically around AI,”she says.To

311、do so properly,Johnston says its important to con-sider,first,the definitions of AI and machine learning,then to identify how AI is being used within the organ-isation based on those definitions,and to construct your responsible AI pro-gramme accordingly so that all employees are aligned.AI is set t

312、o become so ubiquitous that external services that feed into your company may use AI as well.For instance,Google has now introduced generative AI-powered aids to develop documents and slide decks in its cloud-software suite that your employ-ees could soon find themselves inad-vertently using without

313、 knowing it.And if people in your company arent sure what AI is or even if theyre using it you cant be confident your approach to AI is responsible.Root and branch reformJohnston stresses that a clearly understood definition of AI within your company is the basis of any AI governance programme.She r

314、ecom-mends considering the definition of AI systems in the artificial intelligence risk management framework pub-lished by the National Institute of Standards and Technology(NIST)in the US as a working definition.“Making sure everyone is aligned is critical,because you want to check for any use of A

315、I throughout your organisa-tion,”she says.“Any protocol worth its salt needs to be able to categorically define who is using AI tools,when theyre using them,what data theyre using and what the limitations of the tools are.Its also important to ensure AI tools are being used in a way that respects pr

316、ivacy and intellectual prop-erty,given the mounting legal actions against some generative AI tools by those who believe their data was used to train the models that power such platforms.”Doing this work in making sure responsibility is front and centre of any AI deployment is vital because it will a

317、void headaches in the long run.Not only can the irresponsible use of AI lead to trouble,but generative AIs ten-dency to hallucinate content in other words,generate untrue responses could lead to even bigger trouble in the court of public opinion for spreading disinformation.Yet fewer than 20%of exec

318、utives say their organ-isations actions around AI ethics live up to their stated principles on AI.By putting in place a robust responsible AI programme,companies can avoid the pitfalls that come with leaping head-first into the promise of AI without con-sidering its drawbacks.“Were very mindful abou

319、t ethical and responsible use of data,”says Johnston.“Responsible AI should be a priority for organisations globally.”Responsibly transform your business with AI at Think first:why responsibility needs to be forefront when deploying AIAs the AI race heats up,no business wants to be left behind and d

320、oing things properly will yield even bigger benefits TAt Afiniti,we use responsible AI design to make those moments of human connection more valuable60%47%45%40%37%37%36%32%31%29%28%26%26%23%Predict customer behaviour and needsUncover frequent customer journeysImprove omnichannel messagingPersonalis

321、ation/hyper-personalisationCustomise contentDynamic customer segmentation and targetingRecommend products and servicesIdentify root cause of CX problemsEnabling self-serviceIdentify and head off churnCreate contentImprove MQLsReal-time decision-makingSeamless CX through customer journey stages1AI software to detect and prevent money launderingStop fi nancial criminals in their tracks with the NetReveal-Sensa suite Enterprise-grade AI at scale netreveal.aiProtect your business with SymphonyAI