《Contentful：2024生成式AI（GenAI）與職場：認知及使用情況調研報告（英文版）（40頁）.pdf》由會員分享，可在線閱讀，更多相關《Contentful：2024生成式AI（GenAI）與職場：認知及使用情況調研報告（英文版）（40頁）.pdf（40頁珍藏版）》請在三個皮匠報告上搜索。

1、Generative AI Professional Usage and Perception SurveyAmid an emerging gap,knowledge is powerGenerative AI has made quite an impact in the year since OpenAIs ChatGPT gained prominence.But,to paraphrase Jim Collins in Good to Great,if generative AI is the bus,business leaders still need to make sure

2、the right people are on it,people are in the right seats,and leaders know where to take it.At Contentful,our goal is to put our customers in the best possible position to make use of digital content and any technology that helps them engage and communicate with the audiences they care about.Its clea

3、r that generative AI(or genAI,as well refer to it)falls into that category.As we continue to anticipate its capabilities and build them into our own product roadmap,we want to better understand the context and attitudes that shape our customers priorities and usage.In our first survey of 820 profess

4、ionals in a range of technical and non-technical roles around the world,we endeavored to understand how those would-be bus riders see the opportunities and potential drawbacks of genAI.For us,understanding how people across a business feel about genAI provides essential context.Our customer base com

5、prises a wide range of roles and departments:marketers and other business users,designers of various descriptions,digital strategists,developers,and engineers.We surveyed a diverse range of roles to understand how differences and similarities among them might influence the ways they use genAI.We sou

6、ght,in particular,the views of people who are experienced enough to have a perspective beyond just their own roles,but who are not so high-ranking that they were removed from the details of day-to-day work going on.What we found:More than geographic differences or distinctions between those in techn

7、ical versus non-technical roles,perhaps the significant characteristic that marked the most meaningful differences between respondents was the level of self-reported knowledge of genAI.Executive Summary3Contentful Generative AI Professional Usage and Perception SurveyThere is a significant gap betwe

8、en the people who consider themselves highly knowledgeable on generative AI and everyone else.Regardless of whether they work in technical or non-technical roles,the people who consider themselves highly knowledgeable about genAI are the ones working the most with it.They are experimenting and ident

9、ifying how and where it makes a productive impact on their work and target objectives.This is where the real value of genAI is being uncovered.Despite the differences,the knows and know-nots strongly agree on several subjects,including the need to disclose use of genAI,that these new tools will requ

10、ire learning new skills,and that they want the ability to turn genAI capabilities on and off.Over three-quarters of respondents have paid access to genAI tools at work.Only 24%of people in our survey dont pay to use genAI tools at work and nearly as many pay out of their own pockets(either entirely

11、or on top of what their employers fund)to do so.Most people especially those with less knowledge of genAI want more guidance on how to use it responsibly.Sticking your head in the sand and hoping genAI will go away is not an effective strategy for individuals or businesses.The ways in which genAI is

12、 already changing how many people work point to a potentially fast-growing divide between the businesses that empower their employees to make use of these tools and the ones that do not.Were on the cusp of a wave of tailored genAI tools:two-thirds of respondents indicated that their businesses eithe

13、r already have plans for some kind of tailored large language model(LLM)or are considering them.Of those who have or are considering plans,the trend is slightly more in favor of applying an existing LLM rather than training their own.Theres an inherent bias in our data set people who feel they under

14、stand a subject are more likely to respond to a survey about it but its clear that most of our survey respondents feel they have some level of expertise in genAI.Already!Nearly a fifth(18%)rated themselves a“5”in their knowledge(on a scale of 1 to 5)of generative AI and 71%gave themselves a 3 or 4.T

15、echnical respondents are more likely than non-technical respondents to have rated themselves a 4 or 5 in terms of genAI expertise.(Also,Americans and males were more likely to do so in line with general trends across all kinds of surveys.)Everyones a genAI expertSelf-assessed knowledge of genAITotal

16、 n=820,Tech n=390,Non-tech n=430Experience,attitudes,and perceptions5Contentful Generative AI Professional Usage and Perception SurveyFrequency of professional genAI useFrequency of professional genAI useFrequency of personal genAI useFrequency of personal genAI useExperience,attitudes,and perceptio

17、nsPerhaps unsurprisingly,those who consider themselves extremely knowledgeable about genAI also tend to use it the most often.Similarly,those who profess to have little knowledge of genAI are very likely to say they have either never used it or have only tried it a few times.By roleBy knowledgeTechn

18、ical respondents are more likely to use genAI tools more frequently,specifically on a daily basis,both professionally and personally.Non-technical respondents are more likely not to have used genAI,to have only tried it a few times for personal use,and to use it several times a month professionally.

19、Total n=820,Tech n=390,Non-tech n=430High AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=96Total n=820,Tech n=390,Non-tech n=430High AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=966Contentful Generative AI Professional Usage and Perception SurveyOf those who have not used genAI

20、 professionally or personally,we see a mix of skepticism,concern,and lack of knowledge or opportunityRespondents who said they hadnt used genAI either professionally or personally were asked to explain why in an open-ended response.Of the 159 respondents who indicated they hadnt used genAI either pr

21、ofessionally,personally,or at all,the most common reasons include no interest or need,lack of knowledge,concern or fear,and lack of opportunity(see data on page 7).Another 12%indicated that they had either used genAI in one context or another or that they were about to start using it,typically in a

22、professional context.In this latter group,several indicated that they were waiting for their companies to develop guidelines or policies on how to use genAI.Heres one such example:“We are still in the process of finding a proper way of using AI in our work model so that it will be helpful rather tha

23、n a liability.”A similar proportion of respondents either didnt have a reason or werent sure why they hadnt yet used genAI.Other reasons included criticism of the capabilities or output,data protection or privacy issues,and lack of guidance.Experience,attitudes,and perceptions“We are still in the pr

24、ocess of finding a proper way of using AI in our work model so that it will be helpful rather than a liability.”7Contentful Generative AI Professional Usage and Perception SurveyExperience,attitudes,and perceptionsNo interest or need 29%Concern or fear 13%“Havent gotten around to it yet,lots of conc

25、erns too.”“I dont really trust it.”“I dont think its a good thing.”“I have tried it.I do not like it because it takes creativity away from people.”“I think that they need to be careful with AI.It could take over the world.”“Im concerned it is not regulated properly.”Lack of knowledge 16%“Dont know e

26、nough about it,or the accessibility of it.Would be open to learning.”“I dont really know how to use it or what its used for.”“Dont know enough about it.”“I am not familiar enough with it to use it.”Lack of opportunity 11%“I just havent found the time to consider useful applications.”“Firewall issues

27、 at the professional level.No issues using personally.”“I just havent come across it yet,I would try if it came up.”“It hasnt been applicable yet.”“Dont need it,believe people are too dependent on it.”“Havent had the need so far.”“Hasnt occurred to me yet that I wanted to use it,perhaps in the futur

28、e.”“I dont have a reason to use it as I enjoy writing and coming up with ideas on my own.”“No concerns,I just dont need it.I do things by hand.”Most frequent reasons for not using genAI8Contentful Generative AI Professional Usage and Perception SurveyOverall,enthusiasm for genAI is strongAcross the

29、entire survey population,our respondents are enthusiastic about genAI(3.79 mean and 4.0 median,SD 1.0)and perceive the enthusiasm of their employers leadership to be on par with their own(3.77 mean and 4.0 median,SD.94).These top-level results surprised us somewhat wed expected that there might be a

30、 discernible gap between how individuals feel about genAI and how they rate their companys enthusiasm,but in general there really wasnt one.We hypothesized that there might be significant variations in enthusiasm for genAI across different geographic regions,but none emerged.Across regions,there wer

31、e only minor differences between respondents levels of enthusiasm and perceptions of managements enthusiasm for genAI.Even differences between regions were quite small and could largely be attributed to variations in sample sizes,concentrations of respondent roles,and cultural effects on rating scal

32、es.Experience,attitudes,and perceptionsEnthusiasm for genAIHow enthusastic are you about genAI?How enthusastic do you believe management and/or business leaders are about genAI?Total n=820Total n=820,USA n=203,Canada n=104,Australia n=102,Europe n=360Total n=820,USA n=203,Canada n=104,Australia n=10

33、2,Europe n=3609Contentful Generative AI Professional Usage and Perception SurveyBy far,the most meaningful differences depend on levels of self-professed genAI knowledge.Those who are most knowledgeable also tend to be most enthusiastic and they are somewhat more enthusiastic themselves than they pe

34、rceive their managers or business leaders to be.That relationship flips at the other end of the spectrum.Respondents who indicated a lower level of knowledge perceive management to be more enthusiastic about genAI than they are themselves.This highlights one of the most important findings throughout

35、 this survey:there are significant differences between those who consider themselves more knowledgeable about genAI especially the highly knowledgeable and everyone else.As well examine further,this emerging gap is one businesses should identify and act on.Experience,attitudes,and perceptionsHow ent

36、husastic are you about genAI?How enthusastic do you believe management and/or business leaders are about genAI?Total n=820,High AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=96Total n=820,High AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=9610Contentful Generative AI Profession

37、al Usage and Perception SurveyIt hardly comes as a surprise that technical professionals have higher knowledge about generative AI than their non-technical peers.Their job might demand it.Their personality type might draw them to it.Things get interesting when this gap in knowledge takes them to ver

38、y different places.When they think about the future,the same technical professionals who already know a lot are the ones who think theyll need to learn even more skills in the future.It might be tempting to see this as counterintuitive:They already know so much,so the non-technical professionals sho

39、uld be the ones expecting a higher need in order to play catch-up.But that overlooks a snowball effect in the mind.One of the best predictors of future commitment is having made a smaller commitment in the past.Having knowledge fuels the desire to learn more and more.Technical professionals are a bi

40、t more enthusiastic about genAI,non-technical professionals a bit less,and everyone expects that others agree with them.The enthusiastic experts think that business leaders are enthusiastic;those with less enthusiasm believe business leaders share their lukewarm attitudes.This is a perfect example o

41、f what psychologists call nave realism:People think theres an objective world out there,that they themselves see it clearly,and that anyone else with any sense sees things the same way.This tendency makes technical and non-technical professionals think different things about other people.But whos ri

42、ght?The evidence favors the technical professionals.For proof,look no further than the fact that two-thirds of businesses have a vision for an LLM at their company in the works.We could congratulate the technical professionals for their edge in prognosticating.But wed be better served as managers in

43、 getting the non-technical professionals up to speed.Failing to anticipate the future accurately could find them left behind.Which would be a shame,because new research suggests that theyre the ones best poised to reap the benefits of generative AI at work.Expertise:The rift between haves and have-n

44、ots Perspective from the Prof:Sam Maglios takeSam Maglio Professor of Marketing and Psychology,University of Toronto Scarborough11Contentful Generative AI Professional Usage and Perception SurveyThe higher the knowledge,the greater the time saved with genAI Thirty-eight percent of respondents say th

45、ey save from one to almost five hours of time a week using genAI tools.An impressive 37%save between five and 10 hours per week and 11%save more than 10 hours per week.Experience,attitudes,and perceptionsNumber of hours saved by using genAITotal n=82012Contentful Generative AI Professional Usage and

46、 Perception SurveyRespondents with higher levels of genAI knowledge are more likely to save a greater amount of time per week using these tools,with 21%of the most knowledgeable saving more than 10 hours per week.By contrast,those who have the least knowledge save the least time.Levels of genAI know

47、ledge were more significant in determining who saves more time using genAI than technical vs.non-technical job roles,though there are some differences there as well.Notably,respondents in technical roles are more likely to save between five and 10 hours per week and those in non-technical roles less

48、 than an hour.Experience,attitudes,and perceptionsNumber of hours saved by using genAINumber of hours saved by using genAIBy roleBy knowledgeHigh AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=96Total n=820,Tech n=390,Non-tech=43013Contentful Generative AI Professional Usage and Perception

49、 SurveyPositive views outnumber cautious,mixed,or negative viewsWhen given the opportunity to share their perspectives on genAI or AI more broadly,most of our respondents demurred,but the overall sentiment ranged from cautious optimism to strong enthusiasm.Fifty-seven percent provided no response,sa

50、id no comment,told us they didnt know,or provided an incoherent answer.Of the 43%,or 356,who did comment,the majority(61%)were positive about genAI.Twenty-eight percent expressed concerns or had mixed views,with an overall sense that genAI is happening and is going to make a significant impact regar

51、dless of potential drawbacks.Another 8%expressed neutral views.Only 3%were explicitly negative in their views,ranging from it not being necessary in their organizations to genAI being dangerous or outright bad.We classified the responses into positive views and benefits(16%),cautionary views and con

52、cerns(10%),future outlook and evolution(8%),practical applications and utility(5%),uncertainty and lack of knowledge(3%),and concern about jobs(2%).We also received two responses that had clearly been produced by ChatGPT.Its tough to say whats more interesting:the fact that two respondents took this

53、 approach,or the fact that we could so easily recognize thats what they did.Experience,attitudes,and perceptionsPerspectives on genAIWhen it comes to using genAI professionally,the vast majority of our respondents willingly pay to play(sometimes out of their own pockets)But the most interesting grou

54、p is the 18%of respondents who pay out of their own pockets without expensing it to use genAI tools for work.Add the other 5%who pay out of pocket for additional professional use beyond what is company-funded and nearly a quarter of all respondents find these tools so valuable in a work context that

55、 they seem happy to put in their own money to access them.We struggle to identify a comparable technology development that individuals have been so eager to access that theyve been willing to fund professional usage themselves.Mobile phones might come close but most of us with firsthand experience c

56、arried a work phone and a personal phone until employers started rolling out bring-your-own-device policies.Cloud computing may have had some parallels in the early days,but most of that use was billed on credit cards and expensed back to employers.Only 24%of respondents dont pay to use genAI tools

57、at work.For the largest portion of our respondents,37%,their companies pay for their professional use.Another 16%pay for it themselves and expense it.Access and appeal15Contentful Generative AI Professional Usage and Perception SurveyAs in other parts of our survey,patterns vary depending on respond

58、ent segments.Respondents in technical roles are more likely to say their companies pay for their genAI usage;non-technical respondents are more likely to say they dont pay for usage.But theres no significant difference between the groups when it comes to paying and expensing their use,paying themsel

59、ves,or paying for additional usage.The more meaningful distinctions here seem to be determined by the level of genAI expertise.The people who rated their knowledge a 4 or 5 are the most likely to pay their own money to use these tools at work,whether or not they expense it,with the“5s”more likely to

60、 fund over and above employer-paid access.Given that these are the same groups that are most likely to be saving the greatest amount of time per week,it stands to reason that they would also be more willing to pay their own money for access to genAI tools.The proportion of respondents who fund their

61、 own access to genAI tools for work purposes raises several questions:Do employers know that employees are using these tools?Is this use sanctioned?Does it follow established corporate guidelines or policies?These answers are beyond the scope of our survey but something businesses should seek out fo

62、r themselves.Access and appealModes for paying for genAITotal n=749,Tech n=368,Non-tech n=381Total n=749,Tech n=368,Non-tech n=381High AI Knowledge n=147,4 n=326,3 n=229,Little AI Knowledge n=4716Contentful Generative AI Professional Usage and Perception SurveyHow much are people buying the hype aro

63、und generative AI?Enough to put their money where their mouth is,thats for sure.Close to one in five pay out of their own pocket for access to these tools.The opportunity to use these tools at work makes an employer more enticing to over 60%of workers.If companies dont already have their own LLM,odd

64、s are theyre working on one.Nothing riles people up quite like potential,even if that potential might take a while to be realized.Between now and then,we see people behaving like they always do when faced with a hazy opportunity.Theyre terrified by FOMO.They want to make sure they get in on the grou

65、nd floor.And this frenzied rush guides their investment strategy.The overwhelming share of usage for ChatGPT makes it look like an index fund,seen as a catch-all that can benefit even novices.But people are also diversifying their assets.The average person uses approximately three,but people overall

66、 sample from a dozen or more different options.This behavior dipping a toe into multiple genAI pools reveals that people want to explore,to learn,to keep their finger on the pulse of this new technology,or at least to hedge their bets.Theres still plenty left to sort out among these providers.In the

67、 meantime,people navigate this uncertainty by doing what theyve always done:keeping their options open.Give me some of that whatever it isPerspective from the Prof:Sam Maglios takeSam Maglio Professor of Marketing and Psychology,University of Toronto Scarborough17Contentful Generative AI Professiona

68、l Usage and Perception SurveyOverall,most respondents view a potential employers decision to provide access to genAI tools favorably in choosing whether or not to take a job,with far more ambivalence than any negative impact.This is even more strongly the case among those who consider themselves gen

69、AI experts.Those who rated their knowledge of genAI higher(a 4 or 5)were more likely to say it would positively influence their likelihood of taking a job(73%and 89%,respectively).By contrast,access to genAI tools doesnt seem to make much of a difference at all to respondents who dont consider thems

70、elves to be among the genAI cognoscenti.Access to genAI tools at work is a net positiveAccess and appealInfluence of access to genAI tools on choice to work for an employeerImpact on the likelihood to choose to work for an employeer that uses genAI toolsTotal n=820High AI knowledge n=148,4 n=332,3 n

71、=244,Little AI Knowledge n=9618Contentful Generative AI Professional Usage and Perception SurveyOver two-thirds of all respondents,67%,rated the degree to which theyll have to develop new skills as a 4 or 5.A scant 2%said not at all.Technical respondents are more firmly convinced than non-technical

72、respondents that they will need to learn new skills as a result of genAI.Here too,the level of genAI knowledge determines the degree to which respondents anticipate the need to develop new skills.Those who consider themselves the most knowledgeable are overwhelmingly likely to think that genAI will

73、require learning a significant amount of new skills.Even respondents who didnt claim to know much about genAI seem to have a good inkling that theyll need to learn new skills,but they perhaps dont yet know to what extent.Most anticipate that genAI will require them to develop new skillsAccess and ap

74、pealDegree to which genAI requires developing new skillsDegree to which genAI requires developing new skillsTotal n=820,Tech n=390,Non-tech n=430On average,those using genAI tools,whether professionally or personally,are using more than one regardless of role type or level of genAI knowledgeOverall,

75、respondents who use genAI tools professionally use slightly more tools than those who use them personally.As with other areas of our analysis,the average number of tools increases with levels of genAI knowledge.This“more than one tool”statistic and among all but the least knowledgeable,more than two

76、 supports the idea that people across all manner of job roles are experimenting and trying out various options to find what genAI capabilities work best for them.Tools,use,and guidanceWhich genAI tools are you using?Total n=1409,High AI Knowledge n=291,4 n=629,3 n=413,Little AI Knowledge n=7620Conte

77、ntful Generative AI Professional Usage and Perception SurveyNo surprise here ChatGPT is the most well known and the free version is readily accessible to all.Bing makes a respectable showing as the second most frequently used tool,with over a third of respondents indicating professional or personal

78、use.More interesting is how many of the other paid,often more specialist,genAI tools have a significant percentage of users,notably GitHub Copilot,Adobe Firefly,Jasper,and Writer.Theres a significant subsegment of users for whom these tools are becoming mainstream.The fact that close to a seventh of

79、 respondents indicate using Falcon and Metas LLaMA,in one capacity or another,hints at the experimentation being done in building custom genAI tools.Among the fairly small set of“other”tools mentioned,we see on one side genAI capabilities that are integrated into other commonly used tools like Canva

80、 and Wix and on the other highly technical investments like proprietary genAI tools.This suggests that were moving quickly toward a“something for everyone”approach,spanning experts and novices across a range of functional uses.By far,the most commonly used genAI tool is ChatGPT,for both professional

81、 and personal useTools,use,and guidanceGenAI tools usedProfessionally n=699,Personally n=71021Contentful Generative AI Professional Usage and Perception SurveyPeople anticipate using genAI for a wide range of use cases,not just creating content.And their organizations already are.Does that mean that

82、 the“tsunami of crap”that many expect genAI to produce might not be so overwhelming after all?Time will tell,but these results give hope that useful output will result regardless.We asked survey respondents two questions to understand more about what challenges they anticipated genAI would solve for

83、 them and where in a professional capacity these tools are currently being used.First,we asked what types of challenges and needs they anticipated generative or other forms of AI to solve for them.Second,we followed up by asking respondents to tell us where they or,to the best of their knowledge,oth

84、ers in their organization were already using genAI.Among the top challenges or use cases respondents see genAI solving are indeed some that are content-related:researching a content or technical topic,creating a draft,or creating an outline.But there are several important areas not directly related

85、to content,per se,like testing applications,writing code,or cleaning up data.GenAI isnt just for writing blog posts(but yes,that too)Tools,use,and guidanceThis validated a hunch we had:for all the focus in the media and elsewhere on genAI“producing content,”its actually being used in many different

86、ways,often as part of a broader process,and not necessarily to produce“final product”content.Use cases for genAITotal n=82022Contentful Generative AI Professional Usage and Perception SurveyTools,use,and guidance72%Code69%Blog posts70%Photo-realistic images68%SEO content77%Graphics and charts65%Audi

87、o72%Marketing banners80%Technical documentation80%Product descriptions68%Video68%User interfaces65%SEO metadata68%PR or other headlinesRespondents indicated that they and their colleagues are already using genAI in a wide variety of specific areas,either directly by respondents or,to their knowledge

88、,by others in their organizations.Producing technical documentation and product descriptions rank highest based on respondents themselves or others in their organizations using genAI in the process.Graphics and charts,marketing banners,and code round out the top five current uses.Current usage areas

89、 for genAITotal n=82023Contentful Generative AI Professional Usage and Perception SurveyIn comparing where respondents say theyre using genAI tools and where others in their organization are,we see clear indications that most people think that others are using these tools more widely than they are.W

90、hether hype or that universally felt fear-of-missing-out,most people seem to think their own usage is behind the curve.Tools,use,and guidanceCurrent AI usage areas(Me)Current AI usage areas(Others in my organization)Total n=820Total n=82024Contentful Generative AI Professional Usage and Perception S

91、urveyWhile theres a fair amount of overlap,the areas in which technical and non-technical users indicate that they themselves are using genAI reflect different priorities and objectives,but still have significant overlaps across both more technical tasks(i.e.,coding)and less technical tasks(i.e.,blo

92、g posts).Tools,use,and guidanceCurrent AI usage areasTotal n=820,Tech n=390,Non-tech n=43025Contentful Generative AI Professional Usage and Perception SurveyUncertainty.Lets take a minute to talk about how much people hate it.Consider a lottery that costs$1 to play.Theres a 60%chance that it pays$2,

93、leaving a 40%chance of winning nothing.Thats a pretty good bet!With a 60%chance of$2,you can expect to walk away with$1.20 for every$1 you play.But lots of people dont think like that.Theres still a chance of losing the dollar,so they refuse to play.Throughout the results were seeing,respondents are

94、 telling us how much uncertainty they feel about genAI.Based on how far things have come in the short year since the release of ChatGPT,using these tools seems,similar to the gamble above,like a good bet.But its also a risky one.That explains why people are dipping their toes in the genAI water and

95、not doing a sprinting cannonball.Ask people what they think of genAI and its mostly positive,plus a hefty dose of“I dont know,”“Im not sure,”and“no comment.”Those unknowns take some of the shine off the otherwise glistening,positive promise of genAI.Appreciating the benefits it might bring,our respo

96、ndents want to know how to dial down the uncertainty.Nowhere is this on clearer display than when you ask people about how they use generative AI and compare it to how they think others use the same tools.For nearly every type of task,and for the technical and the non-technical alike,professionals t

97、hink that theyre using generative AI some but that others are using it more.What are those other people doing?Yes,theyre presumed to be using it more to make marketing banners and improve SEO.But what theyre really doing,according to our respondents,is“other.”That is,people are thinking that theres

98、this awesome tool in genAI and that everyone else is finding ways to use it that they themselves cant fully appreciate.Wait,wait do tell mePerspective from the Prof:Sam Maglios takeSam Maglio Professor of Marketing and Psychology,University of Toronto Scarborough26Contentful Generative AI Profession

99、al Usage and Perception SurveyAlthough 36%say they have been given a sufficient amount of guidance from their organization on how to use genAI responsibly,more than half of respondents,51%,would like more.Forty percent say they already have some guidance but want more and an additional 11%have none

100、but would like some.Only those who rated their knowledge a 5 are likely to say they have a sufficient amount of guidance.The self-professed experts out there may feel like they know what theyre doing with genAI,but just about everyone else would like more reassurance.Importantly,the people who consi

101、der themselves the least knowledgeable are most likely,on one hand,to say that they dont have but would like clear guidance and,on the other,to say they dont need it presumably because they arent using genAI anyway.Despite their enthusiasm,people overwhelmingly want more guidanceTools,use,and guidan

102、ceAmount of guidance provided on use of genAITotal n=820High AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=9627Contentful Generative AI Professional Usage and Perception SurveyThe overwhelming majority of those in our survey,76%,believe that the use of genAI should be disclosed,whether in

103、ternally or to customers.Only 12%say that it should not.This is one area in which we expected to find some noteworthy geographic differences.To our surprise,these views are highly consistent across regions,with no significant differences.Once again,the most meaningful contrasts emerged among levels

104、of knowledge.If anything,those who are more knowledgeable about genAI are more likely to be in favor of disclosure.We asked respondents to explain in an open-ended response why they felt the way they did about disclosure requirements.Reasonings varied and were often insightful.Consistent themes incl

105、ude,most commonly,the need for transparency and honesty(23%of respondents).The second most common response(11%)expressed uncertainty people werent sure or couldnt explain why they felt the way they did.(Six percent of respondents provided no answer or an incoherent one,but well chalk that up to surv

106、ey fatigue.)The vast majority support requirements to disclose the use of genAITools,use,and guidanceNeed for genAI disclosureTotal n=820,USA n=203,Canada n=104,Australia n=102,Europe n=360High AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=9628Contentful Generative AI Professional Usage a

107、nd Perception SurveyAlthough 5%of respondents pointed in some way to fear of what genAI might do,most responses articulated a more nuanced view.They touched on themes around the potential of genAI to improve work or operations(10%),the need to improve awareness or understanding(8%),the evolving capa

108、bilities of genAI(6%),the ethical obligations or requirements of disclosure(6%),and the need for regulation or policy guidelines(6%).The majority of respondents who believed use of genAI should be disclosed clearly indicated that,for one reason or another,its better to make known where these tools a

109、re used.Of those who dont believe there should be disclosure requirements,many indicated a concern that use of genAI may provide some material advantage that businesses shouldnt have to make public.Others felt that since theres already some level of human quality control or involvement,disclosure is

110、nt necessary.A small but distinct group doesnt think disclosure is necessary because they dont think genAI will amount to much or make a meaningful impact.And then there are responses like this,encapsulating a tech-elitist view that,though far from predominant among our survey population,certainly e

111、xists:“I feel like it isnt a hard concept to grasp,if you need to be explained and/or taught about everything regarding AI,you shouldnt be using it.”(Considering this view came from a male in the 18-24 age group,we may well classify this as the ignorance and certainty of youth!)Tools,use,and guidanc

112、eMost intriguing to us,respondents who had differing views on the need to disclose use of genAI often had very similar reasonings.For example,the idea that“it depends”came from respondents who variously said they thought use of genAI should be disclosed,shouldnt be disclosed,or werent sure.One respo

113、ndent who felt the use of genAI should be disclosed said:“For our industry,I do not think disclosure of using AI is necessary but in other industries,senior leaders need to be transparent with customers if they use AI in their systems,e.g.,health sector.”One who did not believe there should be requi

114、rements to disclose the use of genAI had a similar view:“If its just a part of the process and its not fully controlled by the AI,I dont think its necessary.”Most of the“it depends”camp werent sure about the need for disclosure,with rationales like this:“It depends on the type of content and how it

115、is used.For example,code generated by AI would not need to be disclosed,but images or other art that is published might.”Across the range of comments,there is a palpable sense of cautious optimism.Most of our survey respondents see the potential and possibilities of genAI and want to encourage posit

116、ive development while avoiding potential harm or misuse.29Contentful Generative AI Professional Usage and Perception SurveyAnd action!No wonder everyone wants more transparency and disclosures around usage.On some level,theyre worried that genAI,if not regulated properly,will put others at an advant

117、age over them.As a result,people strive to stay current(paying for genAI out of pocket,playing with more than just ChatGPT,prioritizing employers who use these tools).If theres an advantage to be had,they call dibs.On the risky chance that others(say,someone competing for the same promotion)might ge

118、t an unfair leg up,they want guardrails to level the playing field.People may well have an existential fear that,in 20 years,genAI will take their job or wipe out humanity.In the shorter term,were seeing another kind of AI-nxiety:concerns that genAI will help others get ahead of them.At the institut

119、ional level,this means that businesses need clear internal policies around how genAI can and cannot be used in the workplace.You could imagine a near future where companies not only have HR professionals(to govern matters among employees)but also AI professionals(to govern matters between employees

120、and the technology they use).Even with an internal issue like this settled,businesses also need to consider outward-facing matters where their AI usage meets their customers.We see from our respondents what people have been saying for a long time:The more we trust genAI,the more well embrace it.They

121、ve also,for almost as long a time,been telling scientists like me how to make AI more trustworthy:by making it less uncertain.People fear the unknown,but a little bit of help goes a long way to demystify AI.They might see genAI as a black box but,if you let users put their own tweak on the algorithm

122、,they trust it more.The same thing happens,according to my research,when people watch an algorithm make a mistake and then learn from it.People might not totally get how genAI works,but people do get people.An easy route to making algorithms less uncertain and scary is to make them more human-like.P

123、eople are begging for ways to make the future of generative AI more predictable,regulated,and equitable.As in so many other areas of life,its best to give them what they want.Perspective from the Prof:Sam Maglios takeSam Maglio Professor of Marketing and Psychology,University of Toronto ScarboroughM

124、ore than half,56%,want genAI to be integrated.A sizable minority,29%,prefer them to be standalone.Most people,72%,want the ability to turn integrated genAI capabilities off and on.Though the majority of respondents(56%)favor integrating AI capabilities into existing tools,those with greater genAI kn

125、owledge are more likely to prefer these capabilities be integrated(70%of 5s and 62%of 4s).Those with the least knowledge are fairly evenly split between integrated,standalone,and no preference(see data on following page).Most people prefer genAI capabilities to be integrated into other tools they(al

126、ready)use,but want the ability to turn it offLookingforwardPreference for integrated vs.standalone AI toolsAbility to control use of genAI capabilitiesTotal n=820Total n=82031Contentful Generative AI Professional Usage and Perception SurveyThe majority of all respondents(72%),regardless of knowledge

127、 level,favor the ability to turn genAI capabilities off and on if theyre integrated.The least knowledgeable are more likely to have no preference.Looking forwardPreference for integrated vs.standalone AI toolsHigh AI Knowledge n=148,4 n=332,3 n=244,Little AI Knowledge n=9632Contentful Generative AI

128、Professional Usage and Perception SurveyWhere experts and non-experts agreeIt makes sense that experts and non-experts would diverge in their thinking about generative AI.Similarities in how they think the psychology behind their judgments about generative AI lead to differences in what they think.S

129、o it really jumps out at you when these two groups agree.We found this in two important areas.First,both self-described experts and non-experts want the ability to flip off the genAI switch.This makes it sound like algorithm aversion is alive and well.People can be hesitant when it comes to taking a

130、dvice from an algorithm,especially in certain domains.Theyll let an algorithm tell them which tax prep software to buy but not what kind of clothes to buy with their refund.Theyll trust an algorithm on how best to drive to a movie theater but not what to see once they get there.Of course,despite all

131、 the enthusiasm for this new technology,users still have reasonable reservations about it.The fact that everyone still wants the option to pump the brakes on genAI tells me that people will always want a blend of human and machine.Second,experts and non-experts want to make sure everyone knows whos

132、doing the work.Three-fourths of both groups insist that users of genAI disclose that theyve used it.Some respondents answered the survey from the perspective of being a diligent employee,like the person who said,“Employers should know who is doing the work employees or AI.”But these respondents also

133、 make decisions as consumers,like the person who said,“Consumers need to know if recommended action is from a human.”This sense of wearing both hats was put well by the respondent who said,“One should work transparently both within the company and with customers.”Disclosure around generative AI is t

134、he new frontier in business ethics and corporate social responsibility.Consumers have always prioritized and will always prioritize these values.Companies that meet the moment with openness stand to benefit the most.Perspective from the Prof:Sam Maglios takeSam Maglio Professor of Marketing and Psyc

135、hology,University of Toronto Scarborough33Contentful Generative AI Professional Usage and Perception SurveyJust 31%of our survey respondents said they were unaware of any such plans in their organizations,18%already have plans and a small,but forward-thinking 6%have projects underway.Of those organi

136、zations with projects or plans,49%are applying an existing LLM and 42%are training their own.We seem to be on the cusp of a major wave of tailored genAI use.These results validate our view that,for all the philosophical or academic discussions of“artificial general intelligence”emerging at some poin

137、t in the future,the significant near-term value of genAI for most organizations is in a tailored approach.GenAI tools that reference vetted,validated,and approved inputs are more likely to produce useful outputs,whatever the objective or context.One of the most pointed questions these data raise is

138、which approach to building a tailored LLM will be the fastest and most effective.Training a proprietary LLM using,say,open-source tools can deliver substantial value,but is not for the faint of heart.This approach requires clearly understood use cases,specialist engineering capabilities,and signific

139、ant resources.The ability to fine-tune or apply retrieval-augmented generation(RAG)to existing LLMs may present faster and less expensive(or less demanding)alternatives.In all of these cases,the volumes of proprietary content businesses already have become valuable assets.We look forward to examinin

140、g this topic more thoroughly in future research.More than two-thirds of organizations are considering plans either to apply an existing LLM to their own content or to train a proprietary LLMLooking forwardPlans to train LLM on proprietary contentPlans to use LLMTotal n=820Total n=568The knowledge ga

141、p in genAI has major implications for businesses and individuals alike.Those most knowledgeable about genAI are far more enthusiastic about this new technology than others.They are more efficient because they save more time using it and they anticipate a much larger need to learn new skills.Exposure

142、 and access do not necessarily mean unfettered use.There are many well-founded reasons to be selective about how and where genAI tools are used.However,finding opportunities to use and experiment with them,for all employees regardless of job role,is the most likely avenue to identifying useful appli

143、cations as well as potentially problematic or harmful outcomes.In this case,knowledge really is power.Given the rapid rise of genAI and the demonstrable enthusiasm among those who are most knowledgeable about it,businesses have much to gain by making sure all of their employees have access to these

144、tools and the guidance they need to work with them appropriately.People who dont have the opportunity to work with genAI on the job may seek out other opportunities to do so,either outside work or in other roles elsewhere.Recommendations35Contentful Generative AI Professional Usage and Perception Su

145、rveyBusiness leaders must take actionRecommendationsEncourage everyone throughout the organization to experiment with genAI and give them access to tools Even in businesses that anticipate little benefit from genAI,ignorance could prove exceedingly risky.Knowledge should not be compartmentalized wit

146、hin only a few teams.The potential impact is far-reaching and firsthand experience should be as well.As Prof.Maglio points out,current research suggests that non-technical users may be the most likely to gain from its use.Even negligible benefits may yield valuable lessons(see our comments on survey

147、 methodology).Provide a clear set of guidelines and encourage experimentation Our survey results clearly indicate that employees across a range of roles are eager to see what genAI can do for them,but theyre concerned about doing something harmful or unethical.Business leaders must provide the guard

148、rails that allow employees to experiment without fear of inadvertent missteps.The narrow or broad guidelines should be determined by the type of business and even the work of particular teams.Build a plan for custom genAI toolsTailored LLMs,whether trained from scratch or leveraging existing models,

149、are poised to be the next step-change in genAI.The ability to leverage existing,proprietary content to drive reliable,relevant outputs has broad potential benefits.So does the use of a range of additional,specialized tools that employ genAI.We expect businesses to see the biggest benefits from these

150、 investments.Above all,recognize that genAI tools,for all their potential,are a means to an end.36Contentful Generative AI Professional Usage and Perception SurveyIndividual professionals also have agencyRecommendationsSeek out opportunities to learn,whether at work or elsewhereA growing number of b

151、usinesses are actively encouraging and facilitating experimentation with genAI tools among their employees,but this isnt universally the case.For many reasons(well justified or not),some businesses may restrict or prohibit use of genAI tools.Gaining experience with and building understanding of genA

152、I capabilities can come from any sphere,including personal use and outside courses.Be prepared for change GenAI has been widely available for just over a year.We are only at the start of using it and understanding what it can do.At the same time,capabilities are evolving even more rapidly than exper

153、tise.We can only anticipate how and where it will provide the biggest benefits and make the most impact.Those most knowledgeable about genAI are also the most likely to say there is a lot to learn.Growth mindsets and lifelong learners rejoice.Any survey will have some bias in its responses,whether b

154、ecause of the subjects it covers,the respondents sought,who responds,or the way questions are posed.Ours is no exception.With a subject like genAI in particular,its fair to say that people who feel they know more are more likely to respond.Thats certainly reflected in our survey population.It also m

155、eans that although we met our survey objectives for the number of responses in 10 countries and four regions,an approximate balance between technical and non-technical A few words about our survey sampleroles,a reasonable mix of industry representation and company size,and a mix of age ranges,the ge

156、nder balance among our respondents skewed somewhat more male than female.The full demographics of the survey are available in the demographics section.Overall,we feel confident that our survey population presents a good approximation of the diversity among our customer base,though knowledge of or ex

157、perience with a content management system(CMS)was in no way a qualification to participate in this survey.Appendix:Methodology and demographics38Contentful Generative AI Professional Usage and Perception SurveyThe survey for this research was developed by Contentful,then validated and fielded by Pur

158、eSpectrum on behalf of Contentful during December 2023.Respondents were part of voluntary research panels and contacted via email to complete an online survey.We set response quotas by country and had soft targets for a roughly even number of technical and non-technical respondents in each,job level

159、s that covered mid-level responsibilities and seniority,a reasonable distribution across industry sectors,and a range of company sizes.PureSpectrum and Contentful jointly analyzed the data using Decipher.Additional translation and classification of open-ended responses was conducted by Contentful us

160、ing ChatGPT 3.5(also known as the free version).We found ChatGPTs translations to be consistently high quality,something we validated by requesting a one-for-one output of the original response and the translation.We did a significant amount of spot-checking among these translations and were satisfi

161、ed with the results.We also asked ChatGPT,based on the original questions posed,to identify important themes and then categorize the responses within those themes,counting how many fell into each category.Here is where ChatGPT 3.5 produced some less impressive results.The tool was fairly good at ide

162、ntifying themes among the open-ended responses.It was far less good at consistently categorizing individual responses accurately.In some cases,ChatGPT only provided counts of the responses within each category,without identifying which theme it had classified individual responses within.When asked t

163、o provide each response and the category within which it fell in a table,ChatGPT 3.5 had difficulty with long lists of responses.Any more than about 100 responses at a time caused the system to break down and either stop providing responses or provide nonsense responses.Even with providing 100 respo

164、nses at a time,the tool couldnt get through the entire list of responses for some questions.(Perhaps we should have paid for access to GPT 4!)The other major difficulty the tool had was accurately classifying responses,especially when nuance or informal language was involved.Considering how challeng

165、ing this can be even for experienced humans,we werent particularly surprised.As a result,we took significant time to manually review or add classifications to all responses,and in some cases,to the list of themes we used to classify them.Despite this considerable manual effort,we estimate that ChatG

166、PT 3.5 saved at least five hours of work over several partial days of effort(perhaps 12 hours of manual work).We probably learned at least as much as ChatGPT did in the process.Prof.Sam Maglio at the University of Torontos Rotman School of Business joined the Contentful team in the substantive analy

167、sis of the survey data.His unique perspectives as a psychologist examining perceptions and attitudes toward machine learning and artificial intelligence have provided invaluable contributions to the findings we share in this report.MethodologyAppendix:Methodology and demographics39Contentful Generat

168、ive AI Professional Usage and Perception SurveyCountry breakdownRegional breakdownRespondent job level and roleWhat best describes the level of your position at your organization?What best describes the job function that you work in?Survey demographicsTotal n=820Total n=820Total n=82040Contentful Generative AI Professional Usage and Perception SurveySurvey demographicsCompany sizeGenderAgeHow big is your company?Which of the following best reflects the industry you work in?IndustryTotal n=820Total n=820