《蘭德公司：2023-2024學年美國K-12公立學校AI工具使用情況及應用差異研究報告（英文版）（24頁）.pdf》由會員分享，可在線閱讀，更多相關《蘭德公司：2023-2024學年美國K-12公立學校AI工具使用情況及應用差異研究報告（英文版）（24頁）.pdf（24頁珍藏版）》請在三個皮匠報告上搜索。

1、JULIA H.KAUFMAN,ASHLEY WOO,JOSHUA EAGAN,SABRINA LEE,EMMA B.KASSANUneven Adoption of Artificial Intelligence Tools Among U.S.Teachers and Principals in the 20232024 School YearAccording to a RAND American Teacher Panel survey of K12 public school teachers in fall 2023,nearly one in five teachers was

2、using artificial intelligence(AI)tools for teach-ing,with even higher proportions of English language arts(ELA)or social studies teachers doing so(Diliberti et al.,2024).Another survey from Impact Research(2023)suggested that the use of AI among teachers could be even higher,but mea-surement challen

3、ges make it dif-ficult to know for sure.1 Further-more,upward of 90 percent of a large sample of school district superintendents who were sur-veyed in the 20232024 school year(EAB,2024)agreed that AI“will transform the way we edu-cate students within five years.”That said,a little more than one-thir

4、d of those superintendents indicated that their districts had plans for supporting or training teachers on their use of AI.In this report,we follow up on Diliberti et al.(2024)byagainasking teachers about their use of AI tools for Research ReportKEY FINDINGS Comparable with previous surveys of the A

5、merican Teacher Panel,the results indicate that one-quarter of ELA,math,and science teachers used AI tools for instructional planning or teaching in the 20232024 school year.Nearly 60 percent of surveyed American School Leader Panel principals also reported using AI tools for their work in 20232024.

6、Although one-quarter of teachers overall reported using AI tools,the authors observed variation in use by subject taught and some school characteristics.For example,almost 40 percent of ELA or science teachers reported using AI compared with 20 percent of general elementary education or math teacher

7、s.Teachers and prin-cipals in higher-poverty schools were less likely to report using AI tools relative to those in lower-poverty schools.Eighteen percent of principals reported that their schools or dis-tricts provided guidance on the use of AI by staff,teachers,or stu-dents.Yet,principals in the h

8、ighest-poverty schools were about half as likely as principals in the lowest-poverty schools to report that guidance was provided(13 percent and 25 percent,respectively).2AbbreviationsAIartificial intelligenceAIRSAmerican Instructional Resources SurveysAMESAmerican Mathematics Educator StudyASLPAmer

9、ican School Leader PanelATPAmerican Teacher PanelCCSSOCouncil of Chief State School OfficersELAEnglish language artsFRPLfree or reduced-price lunchIEPIndividualized Education PlanLLMlarge language modeltheir instruction.This time,we drew on American Instructional Resources Surveys(AIRS)data from spr

10、ing 2024 to ask about use of AI tools during the 20232024 school year.The AIRS have been monitoring teachers use of instructional materialsincluding published curriculum materials and online instructional productssince 2019.The AIRS are distinct from the K12 public school teacher survey fielded in 2

11、023 for Diliberti et al.(2024),in that the surveys have a larger nationally representative sample of K12 public school teachers who teach any of three subjects:ELA,mathematics,and science.Thus,the 2024 AIRS allow us to draw more distinctions among teachers in terms of how they used AI for different

12、subject areas.We also examine the extent to which principals are using AI tools and for what purpose by drawing on nationally representative K12 public school principal data from a survey fielded in spring 2024 as part of the American Mathematics Educa-tor Study.Last,we use that principal data to be

13、tter understand how much guidance schools and districts provided to teachers on the use of AI tools and prod-ucts during the 20232024 school year.We explored the following research questions:How much and for what purposes were K12 public school teachers using AI tools for their ELA,mathematics,or sc

14、ience instruction during the 20232024 school year?How much and for what purposes were K12 public school principals using AI tools for their work in the 20232024 school year?To what extent were schools and districts setting policies for use of AI tools during the 20232024 school year?Our investigatio

15、n of these questions provides useful information for those who support teachers at the school,district,and state levels,as well as provid-ers and developers of AI tools for use by teachers and students.Our data also provide guidance to research-ers on critical questions to explore related to teacher

16、s and principals use of AI.BackgroundRoughly a decade ago,even before we fielded the AIRS,we were gathering data through the American Educator Panels on which instructional materials U.S.teachers were using regularly for their classroom instruction.We learned that online supplemental materials had b

17、ecome the go-to resource for many teachers during the 20142015 school year(Opfer,Kaufman,and Thompson,2016).According to Opfer,Kaufman,and Thompson(2016),25 percent to 30 percent of teachers might use some of the most popular comprehensive textbooks,such as Harcourt Reading or Go Math.But,in additio

18、n,large numbers of U.S.elementary teachers(86 percent)reported using Teachers Pay Teachers at least once a week;53 percent of mathematics and ELA teachers in grades 612 also indicated using Teachers Pay Teach-ers.The authors also found that most of the second-ary ELA and mathematics teachers were co

19、nsulting a variety of online resources for their instruction,such as Pinterest and Khan Academy,and roughly one-half or more of elementary teachers reported using Readworks and ReadWriteThink(websites that sup-port students reading comprehension).Findings from Opfer,Kaufman,and Thompson(2016)indicat

20、e that the internet has been deeply 3influencing what teachers taught.At this time,many statesparticularly Louisianaand the Council of Chief State School Officers(CCSSO)had been put-ting in place reforms to encourage school district adoption of high-quality,standards-aligned instruc-tional materials

21、 that teachers would ideally use con-sistently every day for their instruction(Kaufman,Cannon,et al.,2018;CCSSO,undated).These reforms were based on the hypothesis that instruc-tional material reforms were the best way to improve teachers instruction at scale.In contrast,teachers high use of online

22、materials of unknown or dubi-ous quality was likely increasing variation in their instruction:Some teachers might use those materials in ways that align with academic standards,challenge students,and build knowledge over time,while other teachers may use those materials in scattershot ways that are

23、not aligned with academic standards and do not lead to marked increases in student learning.Tools that employ generative AI,such as ChatGPT,are the latest technology advancement that is begin-ning to influence how public educators lead and teach.Yetjust as with such internet resources as Teachers Pa

24、y Teachers or Pinterestthe likelihood that genera-tive AI tools are leading to measurable improvements in teaching and learning is low.As noted by Kleiman and Gallagher(2024)in a recent article about state policy and AI,“The heart of the matter is that gen-erative AI does not distinguish constructiv

25、e,accurate,appropriate outputs from destructive,misleading,or inappropriate ones.”AI may indeed transform lesson planning and teaching,as those other online lesson repositories have,but not necessarily as hoped.Just as concerning as the variable quality of AI-driven instruction,better and more thoug

26、ht-ful uses of AI may be reserved for districts with more resources to train and support teachers.As Diliberti et al.(2024)note,somewhat greater pro-portions of superintendents in suburban and low-poverty districtsas well as those with more White studentsreported in fall 2023 that they had already p

27、rovided training to teachers on AI use or planned to do so in the 20232024 school year.These find-ings have concerning parallels with our research on the use of online resources,in which we observed that,although use of Teachers Pay Teachers and other online resources increased overall from 2015 to

28、2017,use shot up particularly among teachers in schools with more students who are eligible for free or reduced-price lunch(FRPL),English language learners,or students with Individualized Education Plans(IEPs)(Kaufman,Opfer,et al.,2018).Kaufman,Opfer,et al.(2018)posits that teachers working with the

29、se less-advantaged student populations were using such sites as Teachers Pay Teachers because they lacked the instructional guidance and resources they needed to differentiate their instruction to support diverse student needs.The U.S.Department of Educations Office of Educational Technology release

30、d a toolkit in 2024 to help educational leaders guide the integration of AI into teaching and learning in school systems.The toolkit encourages district and school leaders to build a strategic planning process for use of AI that is informed by evidence on effective use,builds AI literacy among educa

31、tors,and depends on them listening to and responding to their communities(Office of Educational Technology,U.S.Department of Education,2024).Yet,those educational leaders who are best poised to take advantage of such toolkits are likely those in schools with more resources and capacity to do so.As A

32、I tools and products for educational pur-poses become more prevalent,studies should track their use among educators.Researchers could iden-tify the particular needs AI is addressing in schools andpotentiallyguide the development of AI products that better meet those needs.In addition,data on educato

33、r use of AI could help policymakers and practitioners consider disparities in that use and implications for equitable,high-quality instruction across the United States.For all these reasons,this report presents results from nationally representative surveys on the use of AI tools among K12 public sc

34、hool principals and teachers and the guidance being provided by schools to support that use.Overview of the Data Sources and Methods in This ReportThis report draws on data from two different sources:(1)teacher survey data collected from the spring 2024 AIRS,which were administered to a national 4sa

35、mple of K12 ELA,mathematics,and science public school teachers in the RAND American Teacher Panel(ATP);and(2)principal survey data collected from the spring 2024 American Mathematics Edu-cator Study(AMES),which was administered to a national sample of K12 public school principals who are part of the

36、 RAND American School Leader Panel(ASLP).The 2024 AIRS teacher survey was com-pleted by 9,126 teachers,and the 2024 AMES princi-pal survey was completed by 3,631 principals.In both datasets,responses were weighted to be nationally representative.In the 2024 AIRS teacher survey,we defined AI as inclu

37、ding“chatbots(e.g.,ChatGPT),adaptive learning systems(e.g.,Khan Academy),and virtual assistants(e.g.,Amazon Alexa).”2 In the 2024 AMES principal survey,we left AI undefined.We asked survey items in the following topic areas in the 2024 AIRS teacher survey:whether teachers used AI tools and products

38、for their instructional planning or teaching in the 20232024 school year and,if so,whether they used AI tools or products to plan their instruction or to introduce AI to their students the frequency with which teachers used AI for their instructional planning or teaching an open-ended question to te

39、achers about how they have used AI tools or products to plan their instruction.We asked survey items in the following topic areas in the 2024 AMES principal survey:whether a principals school or district has provided any policy or guidance on use of AI products or tools in the 20232024 school year w

40、hether principals used AI tools or products in their job as principal and,if so,an example of how they have used those tools or products in the 20232024 school year whom principals have sought guidance from about whether or how to use AI in their jobs what concerns they have about using AI in their

41、job as principal.Throughout this report,we present samplewide and subgroup-specific means and proportions of vari-ables of interest.All reported means and proportions are weighted to be nationally representative of K12 public school ELA,mathematics,and science teach-ers(including elementary teachers

42、 of all subjects)and K12 public school principals.Additional method-ological details for the teacher sample are described in the 2024 AIRS technical documentation(Doan et al.,2024);further methodological details for the school leader sample are described in the 2024 AMES techni-cal documentation(Sch

43、weig et al.,2024).To compare responses for teachers and principals in schools with different demographic profiles,we matched their responses to school-level data from the 20222023 Common Core of Data(National Center for Education Statistics,undated).We compared responses of teachers and principals i

44、n key demo-graphic subgroups(e.g.,educator experience,grades and subjects taught),along with school demograph-ics(e.g.,the percentage of students who are eligible for FRPL,the percentage of students who identify as people of color,and locale).To investigate differences among schools with varied perc

45、entages of students who are eligible for FRPL,we specifically compared responses of principals and teachers in schools where 0-24 percent of students receive FRPL to princi-pals and teachers in schools where 25-49 percent,50-74 percent,and 75-100 percent of students receive FRPL.We refer to lowest-p

46、overty schools as those schools in which 0 percent to 24 percent of students are receiving FRPL.We refer to highest-poverty schools as those in which 75 percent to 100 percent of students are receiving FRPL.Unless otherwise noted,we reference only dif-ferences among educator subgroups that are stati

47、sti-cally significant(p 0.05).We did not make statisti-cal adjustments for multiple comparisons because the intent of this report is to provide exploratory,descriptive information rather than to test specific hypotheses or causal relationships.5As noted above,our report includes findings based on an

48、alyses of teacher and principal survey responses to two open-ended questions,which were answered by a subsample of those respondents:In the 2024 AIRS,we asked teachers to“briefly describe how youve typically used AI tools or products to plan your instruction.”Only teachers who responded that they us

49、ed AI tools or products to plan their math,ELA,or science instruction in a prior close-ended survey question received this open-ended question.Thus,1,883 teachers,or about 21 percent of the whole sample of surveyed teachers,were presented this question,and 1,878 teachers chose to respond.In the 2024

50、 AMES,we asked principals to“provide one example of a specific Artificial Intelligence(AI)product or tool you have used,and how you have used it,in your job as a principal this school year(20232024).”Only principals who responded that they used AI products or tools in their job as a principal in a p

51、rior close-ended survey question received this open-ended question.Thus,1,960 principals,or 55 percent of the whole sample of surveyed principals,were presented this question,and 1,955 principals chose to respond.In analyzing our qualitative data,a team of researchers coded each of the 1,878 teacher

52、 responses from the 2024 AIRS and the 1,955 principal responses from the 2024 AMES.The qualitative team coded the data using a coding scheme inductively developed by the qualitative lead for this report.The coding scheme was designed such that we could cap-ture broad themes and finer-grained subthem

53、es to meaningfully describe the variety of teachers and principals responses.We used established proce-dures to ensure reliability in the process of coding the qualitative data(Bernard and Ryan,2010).At the start of the coding process,the qualitative lead met with a team of two other coders to revie

54、w the coding scheme and clarify definitions and examples for each code.Then,excerpts were divided among the three members of the qualitative team.The qualitative team met frequently throughout the coding process to ensure reliability in applying the codes,resolve potential ambiguities,and discuss re

55、visions to codes or the addition of emergent codes.For instance,when members of the qualitative team had questions about how to interpret specific open-ended excerpts,the team would collectively discuss which code(s)to apply to the excerpt and refine definitions for the codes,as appropriate.When ana

56、lyzing the coded excerpts,we excluded analysis of excerpts we deemed uninterpretable(i.e.,too vague or difficult to under-stand).This process resulted in the analysis of 1,861 interpretable teacher responses and 1,942 interpre-table principal responses.We examined the possibil-ity of substantive dif

57、ferences in teachers responses by subject matter and grade level and in principals responses by the percentage of students eligible for FRPL in their schools.Notable differences by these subgroups are described in the report.Given the broad nature of the open-ended questionsand the nonrepresentative

58、 nature of our qualitative samplereaders should keep in mind that the teachers and principals perspectives expressed through these questions may be more or less prevalent in the general population of teachers and principals than our data currently suggest.Thus,the primary goal of our qualitative ana

59、lysis was to understand key patterns and themes in the use of AI by teachers and principals rather than to determine the overall prevalence of various responses.To that end,when we discuss the open-ended data,we provide proportions and/or frequency counts(e.g.,“about 20 teachers,”“about 300 principa

60、ls,”or“one-third of responding teachers”)to provide readers with a sense of how fre-quently some themes arose in our analysis process.In some cases,we provide the number of teachers or principals who expressed sentiments that aligned with specific themes rather than proportions because our coding sc

61、heme captured subthemes that were drawn from a relatively small proportion of our total sample of teachers or principals.6How Much and for What Purposes Were K12 Public School Teachers Using AI Tools for Their ELA,Mathematics,and Science Instruction During the 20232024 School Year?One-Quarter of Tea

62、chers Reported Using AI Tools for Planning or Teaching,with Higher Use Among ELA,Science,and Secondary Teachers Than Math and Elementary TeachersThe survey asked ELA,math,and science teachers about their level of familiarity with AI tools,from never having previously heard of AI to using AI tools or

63、 products for their instructional planning or teach-ing.Figure 1 shows only the proportion of teachers who reported using AI tools for planning or teaching in the 20232024 school year.About one-quarter of teachers,overall,reported doing so.This percentage is slightly higher than the 18percent of K12

64、 public school teachers across all subjects who reported using AI tools in fall 2023(Diliberti et al.,2024);however,because the samples draw on somewhat different pop-ulations of teachers,we cannot directly compare them.Although not included in Figure 1,nearly two-thirds of teachers reported that,ev

65、en though they had FIGURE 1Percentage of K12 ELA,Mathematics,and Science Teachers Who Reported Using AI Tools or Products for Instructional Planning or Teaching01020304050All teachersPercentage of teachersTeacher characteristicsNOTE:This figure depicts response data from the following survey questio

66、n:“What statement best describes your use of AI tools and products as part of your instructional planning or teaching this school year(202324)?”Response choices were as follows:(1)“This is the first time Ive heard about AI tools and products for instructional planning or teaching”;(2)“Ive heard of A

67、I tools and products but have never used them for instruction-al planning or teaching”;(3)“Ive never used AI tools or products for my instructional planning or teaching,but have used them outside of my job”;and(4)“Ive used AI tools or products for instructional planning or teaching.”Only those who r

68、esponded with(4)Ive used AI tools or products for instructional planning or teaching are displayed.An asterisk(*)indicates that percentages for that subgroup significantly differ at the p 0.05 level from the reference group(ref.)before controlling for any teacher-or school-level characteristics.The

69、black bars represent 95-percent confidence intervals.The universe of respondents was all teachers in the sample.ref.=reference group.N=8,814.SchoolcharacteristicELA(ref.)ScienceMathElementary educationElementary(ref.)Secondary05 years610 years1120 years21 or more years(ref.)024%(ref.)2549%5074%75100

70、%Main subjectGrade levelYears of experiencePercentage ofstudents eligiblefor FRPL25393721*18*1834*29*29*25*22302724*23*23*7never used AI for instructional planning or teaching,they had either at least heard of AI tools(46 percent)or used AI tools outside the classroom(15 percent).These data suggest

71、that,even though use of AI for planning or teaching may not be highly prevalent now,most teachers are familiar with the concept of AI,which could lay the foundation for future use.Certain categories of teachers were much more likely to use AI tools for planning and teaching com-pared with others.Fig

72、ure 1 shows the following:Almost 40 percent of teachers whose main sub-ject was ELA or science(including teachers at the elementary and secondary levels)reported using AI tools,in comparison with about 20 percent of teachers who reported their main subject as elementary education or science.3 Second

73、ary teachers(those who are teaching grades 612)were also about twice as likely as elementary teachers to report using AI tools for teaching and planning(34 percent com-pared with 18 percent).Teachers with fewer years of experience were also more likely to use AI tools for teaching and planning than

74、veteran teachers with 21 or more years of experience.Last,teachers in the lowest-poverty schools(i.e.,those where less than 25 percent of stu-dents qualified for FRPL)were more likely than their counterparts in higher-poverty schools to report that they used AI tools for planning or teaching.As we w

75、ill discuss further later in this report,these differences may reflect teachers varying access to knowledge or resources to engage with AI tools to enhance their instruction and guidance on AI use.Among Teachers Using AI Tools or Products,A Vast Majority Reported Using Them for Instructional Plannin

76、g;Only 36 Percent Reported Introducing AI Tools to StudentsThe survey asked teachers whether they“used AI tools or products to plan”their ELA,math,and/or science instruction,as well as whether they“introduced AI tools or products to students”in ELA,math,and/or science class.Among the one-quarter of

77、teachers nationally who reported using AI tools,64 percent said that they used them for instructional planning only,whether for their ELA,math,or science instruction;only 11 percent said that they introduced them to stu-dents but did not do instructional planning with them;and 25 percent said that t

78、hey did both(see Figure 2).Elementary teachers were more likely to use AI tools for instructional planning and to introduce AI tools to students for their ELA instruction than their math or science instruction.Among secondary teach-ers,we observed a slightly different pattern.Among secondary teacher

79、s who used AI tools,92 percent of both ELA and science teachers reported using AI tools for instructional planning versus 83 percent of math teachers.About 35 percent to 37 percent of secondary ELA,math,and science teachers reported introducing their students to AI tools,although sec-ondary math tea

80、chers were slightly more likely than their counterparts to use AI tools only with students versus using AI tools for instructional planning.We did not ask teachers to elaborate on what it meant to them to“introduce”AI tools or products to their students.Thus,we lack information about the variety of

81、ways teachers could be doing so.A Little More Than One-Half of Teachers Who Used AI Tools for Instructional Planning Reported Using Them Once a Month or Less,Although Teachers in High-Poverty Schools Used Them More Frequently Than Their Counterparts in Low-Poverty SchoolsGiven the predominant use of

82、 AI for instructional planning,we examined trends among teachers who used AI for this purpose.Among teachers who reported using AI in their instructional planning or teaching(i.e.,one-quarter of teachers),53 per-cent said that they used AI once a month or less frequently,and 29 percent said that the

83、y used AI a few times per month(see Figure 3).Only 19 per-cent of teachers who used AI for instructional planning4 percent of all teachers nationallysaid that they used AI at least weekly.8At the same time,among teachers who reported using AI for instructional planning,25 percent of teachers in scho

84、ols with the highest levels of poverty said that they used it at least weekly,in comparison with just 16 percent of teachers in the lowest-poverty schools.Thus,when teachers at high-poverty schools did use AI,they used it more than their peers at lower-poverty schools.Although these findings may sou

85、nd contradictory,they suggest that teachers in higher-poverty schools may initially face more barriers in using AI(e.g.,may not have the technology to access AI readily,may be focused on other obstacles)but also that they may find that AI addresses their needs in useful ways when they do use it.In a

86、ddition to asking teachers about their use of AI tools broadly,we also asked all teachers in the full nationally representative sample about their use of a variety of digital supplemental instructional materials for their ELA,math,and science instruction,includ-ing ChatGPT,alongside a variety of oth

87、er resources(e.g.,Teachers Pay Teachers,YouTube).Across sub-jects,the most common online resources reported by U.S.K12 teachers were YouTube(used at least once a week by 38 percent of teachers for ELA instruction,25 percent for mathematics instruction,and 53 per-cent for science instruction)and thos

88、e from Teachers Pay Teachers(used at least once a week by 48 per-cent of teachers for ELA instruction,43 percent for mathematics instruction,and 48 percent for science instruction).By comparison,only a small percentage of teachers reported using ChatGPT weekly or more often(between 5 percent and 7 p

89、ercent,depending FIGURE 2Percentage of Teachers Who Reported Using AI with Students and/or in Instructional Planning for ELA,Math,or Science020406080100Percentage of teachersNOTE:This figure depicts response data from the following survey question:”What best describes how you currently use AI tools

90、or products for your instructional planning or teaching this school year(202324)?”Teachers could select all that apply from the following response choices:“Ive used AI tools or products to plan my ELA/mathematics/science instruction”and“Ive introduced AI tools or products to students in ELA/mathe-ma

91、tics/science classes.”For all teachers and all general elementary teachers,we display the proportion of teachers who reported using AI for planning for any subject and who reported introducing students to AI for any subject.The proportion of elementary teachers using AI for ELA,math or science is le

92、ss than 100 percent because those teachers may have reported using AI for one subject but not others.We also display the responses of general elementary teachers,who often teach multiple subjects,for ELA,math,and science separately.For secondary teachers,we display the responses of secondary ELA tea

93、chers for the survey items about instructional planning or teaching for ELA instruction,the responses of secondary math teachers for the survey items about instructional planning or teaching for math instruction,and the responses of secondary science teachers for the survey items about instructional

94、 planning or teaching for science instruction.The universe of respondents was all teachers who reported that they used AI in their instructional planning or teaching.N=2,138.All teachersusing AIGeneral elementary teachersAll generalelementaryteachersELAMathScienceELAMathScienceSecondary teachers6468

95、56312562636425201211830202811127747157493775Teachers who only introducedAI to their students BothTeachers who only used AI to plantheir instruction 9on what subject they teach).That said,other online resources we asked about,such as Khan Academy,Kahoot!,and i-Ready,do integrate elements that can be

96、considered AI.For instance,i-Ready and other adaptive learning programs can use AI to assess students learning progress and provide them with lessons,feedback,and assessments that are tailored to their skills and needs(Ray,2023).As another example,Kahoot!now uses AI to propose questions to users or

97、to read text aloud(Kahoot!,undated).Teachers Most Commonly Reported Using AI Tools to Generate Lesson Materials,Assess Students,and Differentiate InstructionGiven the various ways that teachers could define AI and their use of it,we also asked teachers to describe how they typically use AI tools or

98、products to plan their instruction via an open-ended survey item(see textbox).Only teachers who said that they used AI tools or products to plan their ELA,math,or science instructionabout one-fifth of teachers in the whole samplewere presented with this open-ended ques-tion.Among the teachers who we

99、re asked this ques-tion,1,861 teachers provided an interpretable response.We identified five major categories of AI tool use by teachers.In order of prevalence,these categories were to(1)generate lesson materials;(2)assess students;(3)differentiate instruction;(4)complete administra-tive tasks;(5)he

100、lp students use AI.We describe theseuses in more detail in the remainder of this section.Teachers use of AI tools to generate lesson materials.Altogether,about three-quarters of the 1,861 teachers who provided an interpretable response to the open-ended question reported using AI to FIGURE 3Percenta

101、ge of Teachers Reporting How Frequently They Use AI Tools for Instructional Planning,Overall and According to School Poverty Status(Based on the Percentage of Students Receiving FRPL)NOTE:This figure depicts response data from the following survey question:“How frequently have you used AI tools or p

102、roducts to plan your instruction this school year(202324)?”The figure only includes responses from teachers who reported using AI tools or products in instructional planning.An asterisk(*)indicates that percentages for that subgroup significantly differ at the p 0.05 level from the reference group b

103、efore controlling for any teacher-or school-level characteristics.Percentages within the bars do not always add up to 100 percent because of rounding.ref.=reference group.N=1,882.All teachers usingAI for instructionalplanning024%(ref.)2549%5074%75100%Percentage of studentseligible for FRPL withintea

104、chers schoolsPercentage of teachersusing AI for instructionalplanning at least weekly16131111139*7*35646*5159545329292731291916142025Once a month or less2 to 3 times per month1 to 2 times per week3 times per weekor more10generate instructional materials.The most common uses that teachers mentioned w

105、ere creating student-facing assignments or activities,cited by about 750 teachers,or roughly 40 percent of responding teach-ers.For instance,ELA teachers mentioned using AI tools to develop text-based discussion questions,essay prompts,comprehension questions,or worksheets on practicing grammar.Mean

106、while,math teachers described using AI to create problem sets,practice questions,word problems,and problems with real world applications.One general elementary teacher described how they used AI to create word problems that were engaging for their students:“I have used AI to help create engaging con

107、tent for students using their interests,along with the skill that we are focusing on at the time.For instance,I used AI to generate regroup-ing addition word problems using Minecraft and Fortnite themes.”In addition,about 500 teachersor about one-quarter of responding teachersmentioned using AI to g

108、enerate lesson plans or to brainstorm lesson ideas on a topic.Teachers mentioned using AI to plan lessons if they were experiencing difficulty planning the lesson on their own or wanted to create a lesson that was more engaging or aligned with state standards.For example,one teacher explained,“If I

109、am struggling with planning out a lesson,I use it AI for help.”About 550 teachers reported using AI to generate other types of teaching resources,including exemplar work for students(e.g.,model papers),texts or passages for students to engage with,or slideshow presentations.Teachers use of AI tools

110、to assess students.In addition to using AI to create lesson materials,about one-quarter of the 1,861 teachers who reported on their use of AI for instructional planning said that they used AI to help them assess students.The vast majority of these teachersabout 400said that they used AI tools to cre

111、ate assessments,quizzes,and rubrics.About 80 teachers described using AI to grade student work and provide feedback to students,using such tools as Writeable from HMH(which provides students with AI-generated feedback on their writ-ing),CoGrader(an AI essay grader),MagicSchool AI,Brisk,Class Compani

112、on,and others.Notably,about three-quarters of these teachers were secondary teach-ers or those who reported that their main subject was ELA,which underscores how teachers often used AI to grade or provide feedback on writing.For instance,one high school ELA teacher said,“I have also used MagicSchool

113、 to create specific feedback on student writing,so that I can return the writing quickly and students can use the feedback to revise.”Teachers use of AI tools to differentiate instruc-tion.About 200 teachers,or about 10percent of responding teachers,also described using AI tools to differentiate the

114、ir instruction.Teachers described adjusting the level of texts or reading passages to match students needs.Less commonly,teachers also discussed using AI to differentiate word problems,prompts,assignments,problem sets,lesson plans,and questions for students.About one-third of these teachers describe

115、d using AI to support instruction for specific groups of students,including,most com-monly,struggling learners,followed by multilingual AI Tools Used by Teachers and PrincipalsIn their responses to our open-ended questions,teachers and principals named various AI tools they used in their work.Below,

116、we describe the tools most commonly mentioned by teachers or principals:ChatGPT is a large language model(LLM)that can answer prompts,generate ideas,create resources from scratch,refine materi-als,and act as a thought partner.ChatGPT was by far the most common tool mentioned by teachers and principa

117、ls.Other LLMs,such as Gemini developed by Google or CoPilot developed by Microsoft,function similarly.MagicSchool AI is a comprehensive AI platform containing tools that can assist educators with lesson planning,assess-ments,differentiation,and communication with families.Canva offers a selection of

118、 AI image-generator tools.Khan Academy is an adaptive learning platform.Amira is an AI-powered reading tool that assists with reading practice,assesses oral fluency,and screens for dyslexia.11learners(MLLs),students with IEPs,and more-advanced learners.When differentiating instruction for struggling

119、 learners and MLLs,teachers described adjusting the level of text for struggling readers or creating student assignments(e.g.,worksheets,prob-lem sets,reading passages,questions)to review past material for more practice.A few teachers added that they use AI to help brainstorm ways to teach a lesson

120、when students struggle.When differentiating instruc-tion for students with IEPs,teachers mentioned using AI to help them develop IEP goals,shape lessons or assessments around IEP goals,and brainstorm ways to accommodate students with IEPs.The few teachers who mentioned differentiating instruction fo

121、r their advanced learners used AI to expand on a topic or lesson or create more-challenging materials and more-complex questions.About one-half of teachers who use AI to support differentiation were ELA teachers,which aligns with our finding that teachers using AI for dif-ferentiation were often mod

122、ifying the reading levels of texts.Relatively few math teachersonly about tenreported using AI for differentiation.Diffit,an AI tool that adapts content to differ-ent reading levels,and ChatGPT were the tools most commonly mentioned by teachers using AI for dif-ferentiation.Notably,even teachers who

123、 were not ELA teachers used AI to support differentiation for readers at different levels.One high school science teacher commented,“Our biology PLC professional learning community has used the tool Diffit to develop a few lessons and activities for students at various levels of reading and backgrou

124、nd knowledge.”Teachers use of AI tools for administrative tasks and direct student support.Although our question to teachers asked how they used AI tools or products to“plan their instruction,”some teachers reported other uses for AI beyond instructional planning(see Figure 4).About 100 teachers rep

125、orted using AI to support administrative tasks,such as drafting parent emails,report card comments,student recommenda-tion lessons,and substitute plans.Roughly 50 teachers described using AI with their students.These teachers said that students used adaptive learning systems(such as Khan Academy or

126、i-Ready)or tutoring systems(such as Amira or Khanmigo).Teachers also described teaching students how to create their own practice problems using AI tools or helping students under-stand how to use AI to support their brainstorming,research,or editing.A few teachers also described teaching their stud

127、ents to assess and critique the writ-ing created by AI to demonstrate to students the short-comings of AI-generated writing.Finally,a relatively small number of teachers(i.e.,about 20)reported using AI for each of the following other purposes:checking for student plagiarism or whether students used

128、AI for their assignments,translating text to other languages,or researching topics relevant to their instruction(e.g.,structured literacy or how to solve math problems).Teachers Had Mixed Perceptions About the Helpfulness of AI Tools for Instructional PlanningAlthough we did not ask teachers about t

129、heir percep-tions of the helpfulness or usefulness of AI for instruc-tional planning,some teachers expressed opinions in their responses to an open-ended question about how AI helped them in their work or to overcome chal-lenges.About 100 teachers of the 1,861 responding“I use ChatGPT to gather info

130、rmation for tasks that would take longer than if I did them myself(e.g.,compiling definitions for words in a table,summarizing a long article or adjusting its reading level,creating knowledge check questions with specific criteria I set).”High School Science Teacher12teachers specifically described

131、ways AI was useful or helpful to them.Some of these teachers mentioned using AI tools to personalize student assignments,such as adding students names to word problems or cultural references in exemplars,or to include real-world concepts in lessons.Teachers also reported that AI helped them brainsto

132、rm lesson ideas or interactive activities and streamline tasks.For example,teachers described how AI saved them time by accelerating the process of time-consuming tasks,such as writing out lesson plans,leveling texts,generating writing exam-ples,creating worksheets,and compiling assessment questions

133、,although some added that they still have to modify or double-check the results produced by AI.In the words of one teacher,“AI is not a replacement for what I do,but it can help make some elements of what I do go smoother.”A small number of teachers also described how using the data collected from a

134、 variety of AI platforms(e.g.,adaptive learning systems)can help inform instruction or lesson planning to target student needs or standards.Forty teachers also highlighted potential draw-backs or challenges with using AI for instructional planning.The majority of these teachers described how they ha

135、d to assess the quality of the AI output.For example,they noted that sometimes AI made errors,FIGURE 4Teachers Most Commonly Reported Uses of AI Tools for Instructional Planning-Generate lesson plans,teaching resources,and classroom activitiesCreate assessments,provide student feedback,and support g

136、radingDifferentiate instructionfor a range of learnersSupport administrative tasksHelp studentsuse AI themselves“I used an AI to outline a lesson plan for ethics in using AI.Ive also used AI to help suggest group activities for lessons on literature.After the AI provides input,I edit and adjust the

137、material to better ft the needs of my students.”High School ELA Teacher“I do use ChatGPT on occasion to create test and quiz questions(example instructions to AI:create a quiz with 10 multiple choice and 2 short essay questions about stocks,bonds,and mutual funds for a remedial 12th grade math class

138、 or create a quiz about chi-square distributions similar to this one,and paste in an existing document).I do have to solve all quiz problems myself to correct errors or make the wording and symbolism consistent with our texts,but it is still faster than creating a test or quiz entirely on my own.”Hi

139、gh School Math Teacher“Before ChatGPT I struggled to find middle school texts to match the phonics skills they students were learning.They were very babyish,and the students were embarrassed to read them.With ChatGPT,I can create an appropriate passage that is high interest and contains all the skil

140、ls they students are learning.It has been an absolute lifesaver and takes less than 5 minutes to create!Middle School ELA Teacher“I havent used it to actually plan but I have used it to write recommendation letters for students who failed to turn in resumes or background information.”High School ELA

141、 Teacher“All students have apps installed on their Chromebooks(Grammarly Basic,Snap&Read,G-Suite).Students with identified struggles(IEP and 504)have additional apps installed on theirs(Co:Writer,JAWS).When it comes to writing,students have been taught how to turn on/off the tools to accommodate the

142、ir needs.When introduc-ing a new concept or before they turn in an assignment,I remind them that they have the opportunity to double-check their work with a program before submission.”Elementary ELA Teacher13provided misinformation,or did a worse job on a task than the teacher would have done,which

143、required them to double-check the AI output and make revi-sions before using it.A few teachers also discussed how they showed these mistakes to their students as a way to teach students about the limitations of AI.For instance,one teacher noted,“I used AI to create an example essay and to model the

144、limitations of AI for students to dissuade them from using it as a replace-ment for actually completing work on their own.”How Much and for What Purposes Were K12 Public School Principals Using AI Tools for Their Work in the 20232024 School Year?Nearly 60 Percent of Principals Reported Using AI Prod

145、ucts or Tools in Their Jobs for a Variety of Purposes,Including Drafting Communications and Supporting Teachers A majority of principals(58 percent)reported using AI products or tools in their jobs.In an open-ended question,we asked these principals to describe the ways they used AI in their jobs.We

146、 analyzed the interpretable responses of 1,942 principals.We identified five major categories of AI tool use by principals.In order of prevalence,these categories were to(1)draft communications;(2)support with other school administrator tasks;(3)support teacher hiring,evaluation,or professional lear

147、ning;(4)create instructional resources for teachers or help teachers create them;and(5)do research.We describe these uses in more detail in the remainder of this section.Principals use of AI tools to draft commu-nications.About one-half of those 1,942 principals reported using AI to draft communicat

148、ions with parents,staff,and other stakeholders.This was the most common use of AI cited by principals.They described using AI to create or revise emails,letters,newsletters,speeches,or school announcements and social media posts,flyers,memos,or presentations.Beyond using AI to draft communications,p

149、rincipals also reported using AI to enhance the quality of their writing(e.g.,helping them with proofreading,edit-ing,and revising).Principals use of AI tools for other admin-istrative purposes.About 400 principals,or about one-fifth of responding principals,described using AI to help them complete

150、a variety of other school administrator tasks.The majority of these principals said that they used AI to write recommendation letters or reference letters for teachers or students.Principals also described using AI to help them develop school policies,such as school handbooks,mission state-ments,str

151、ategic plans,or improvement plans and to plan school programs and events,such as field trips,school celebrations,intervention programs,or themed weeks.AI also supported principals with such tasks as creating master bell schedules,drafting behavior inter-vention plans and IEPs,developing data collect

152、ion instruments(e.g.,parent surveys),and analyzing and summarizing the findings drawn from data.Principals use of AI tools to support teach-ers.Nearly 300 principals,or about 15 percent of responding principals,described using AI to support teacher hiring,evaluation,or professional learning.“AI tool

153、s are currently terrible at suggesting a math lesson plan or creating resources when given specific requirements.any teaching resource I ask AI to create has been terriblenot meeting my constraints and/or below grade level and the rigor my students need.”Middle School Math Teacher14About 80 of these

154、 principals reported using AI to craft job descriptions,interview questions,hiring rubrics,or justifications for a position.About 150 principals also described using AI to support teacher evaluation and to provide teachers with feedback.They used AI to create evaluation templates,sum-marize observat

155、ion notes,and draft observation reports and improvement plans.For instance,one principal described their use of AI:“I enter the descriptor and information about the teacher,and it gives me a starting point to edit narratives for each section.AI reduces evaluation paperwork time by half.”Nearly 70 pr

156、incipals also used AI to help them craft activities for staff meetings and professional learning opportunities.They described using AI to generate ideas for meetings,craft professional learn-ing objectives,and develop professional learning resources(e.g.,examples of best practices,resources for coac

157、hes,or slide decks).About 150 principals described using AI to create instructional resources for teachers or help-ing teachers use AI for this purpose.Similar to our findings about teachers use of AI for instructional planning,principals described using AI to generate classroom activities and asses

158、sments and to differ-entiate instruction.These principals also described the ways in which they helped teachers use AI for planning.For instance,principals described guiding teachers use of AI to support with lesson planning,unpacking standards,creating assessments,building IEP goals,and differentia

159、ting student resources.Principals use of AI tools to research topics.Finally,about 50 principals used AI to research and better understand various topics related to their jobs,such as best practices,education policies,and cur-riculum ideas(see Figure 5).About 15 to 20 princi-pals also used AI to sup

160、port with such administrative tasks as transcription,note-taking,and translation.FIGURE 5Principals Uses for AI Tools in Their JobsDraft communicationsand enhance their writingSupport schooladministrator tasksSupport teacherhiring,evaluation,andprofessional learningSupport instruction ordemonstrate

161、to teachershow to use AI for instructionResearch topicsrelevant to their jobs“I use ChatGPT for weekly newsletters and emails to staff.It helps me dump my thoughts out and puts them in a more concise and coherent form.”High School Principal“I use ChatGPT often to get ideas.I usually type in my reque

162、st,review the answer,add more details due to informa-tion that I felt it was lacking,get a modified response,then modify it myself to get the responses tailored to my needs.For example,I needed to create a parent survey after an event,so I looked to ChatGPT to come up with a survey.In writing a scho

163、ol grant,I consulted ChatGPT with things to consider for specific areas.”Elementary PrincipalWe have used ChatGPT to come up with ideas for professional develop-ment.For example,we were working on restorative practice strategies for dealing with phone use in class.We asked ChatGPT for sentence stems

164、 to help teachers have restorative conversations to address behavior.“High School Principal“I have been using Quizziz to help teachers in PLCsprofessional learning communities to use the personalized learning path tools for assessments and lessons.It is helping us greatly to ease the ability of teac

165、hers to differentiate for students and to utilize more personalized tools for assessing students after remediating standards.”Elementary Principal“I have used AI to help generate some ideas around supporting students with cell phone use and social media,as well as social-emotional health.”Middle Sch

166、ool Principal15Principals in the Highest-Poverty Schools Were Less Likely to Use AI Tools in Their Jobs and Less Likely to Seek Guidance on Using AI Compared with Those in the Lowest-Poverty SchoolsAmong all principals in the nationally representa-tive sample,principals in the highest-poverty school

167、s were somewhat less likely than principals in the lowest-poverty schools to report using AI tools in their jobs(62 percent compared with 54 percent)(see Figure 6).As we also observed,teachers in high-poverty schools were less likely to use AI tools.Principals lower use of AI tools in higher-poverty

168、 schools could discourage teachers in those schools from using AI tools.It could also mean that princi-pals in higher-poverty schools are less equipped to help their teachers use AI tools.At the same time,other factors,such as less access to broadband outside school,capacity to use technology,and ot

169、her pressing concerns,may keep both principals and teachers in higher-poverty environments from accessing AI tools in the first place.Regardless of whether they used AI tools,we also asked principals on whom they relied for guidance on the use of AI,including other principals,professional associatio

170、ns,teachers,district staff,news media,or social media;principals could also report that they did not seek guidance from any source.The most common source of guidance that principals reported was other principals in their districts or professional networks(39 percent),followed by their professional a

171、ssociations(22 percent),their supervisor or other district staff(22 percent),teachers in their schools or districts(24 percent),and social media(20 percent).Reported guidance on the use of AI from the news media was comparatively rare(8 percent).Beyond the sources of guidance we asked about,principa

172、ls wrote in other sources,including conferences,train-ings,educational courses,and,less frequently,their social circles(i.e.,their family,friends,mentors,col-leagues,and staff).Among principals who reported that they did not seek guidance on the use of AI,only 32percent said that they used AI.Howeve

173、r,among principals who sought guidance from any source,the FIGURE 6Percentage of Principals Using AI Tools for Their Jobs,According to School Poverty Status(Based on Percentage of Students Receiving FRPL)020406080100Percentage of principalsNOTE:This figure depicts response data from the following su

174、rvey question:“Do you use AI products or tools in your job as a principal?”We display the percentage of principals who responded,“Yes.”An asterisk(*)indicates that percentages for that subgroup significantly differ at the p 0.05 level from the reference group before controlling for any principal-or

175、school-level characteristics.The black bars represent 95-percent confidence intervals.The universe of respon-dents was all principals.ref.=reference group.N=3,593.percentage of principals using AI was nearly double:56 percent to 76 percent.Across all sources of guidance that we asked about,with the

176、exception of the news media and social media,principals in the highest-poverty schools were about 10 to 15 percentage points less likely than their counterparts in the lowest-poverty schools to report that they sought guidance on whether and how to use AI in their job.For instance,48 percent of prin

177、cipals in the lowest-poverty schools said that they sought guidance on using AI from other principals,in comparison with 34 percent of principals in the highest-poverty schools.Relatedly,principals working in the highest-poverty schools were more likely to report that they have never sought any guid

178、ance about using AI tools than prin-cipals in the lowest-poverty schools(37 percent com-pared with 23 percent).AllprincipalsPercentage of students eligible forFRPL within a principals school024%(ref.)2549%5074%75100%5862585954*16A Lack of Professional Learning,Concerns About Data Privacy,and Uncerta

179、inty About How to Use AI Were Principals Top Concerns About Using AI Tools We also asked principals about the extent to which any of the seven factors listed in Figure 7 had an influence on whether or how they used AI tools or products.Factors cited by the most principals as having a major or minor

180、influence on their use of AI tools were a lack of professional development for using AI tools or products(72 percent),followed by concerns about data privacy(70 percent)and uncertainty about how AI can be used for their jobs(70 percent).Relatively few principals indicated con-cern that AI would make

181、 them less efficient in their job(22 percent).Principals in the highest-poverty schools were somewhat more likely than principals in the lowest-poverty schools to say that their con-cerns about data privacy had a major influence on whether and how they use AI in their jobs(31 percent compared with 2

182、4 percent).We observed a similar pattern for a lack of district guidance as a factor influencing AI use(27 percent of principals in the highest-poverty schools compared with 19 percent in the lowest-poverty ones).Sixteen percent of principals also noted additional factors beyond the seven we asked a

183、bout as influenc-ing their use of AI.One of the factors that principals wrote in most frequently was concern about ethical issues related to AI,including whether it is acceptable to use AI and whether using AI constitutes plagia-rism.Another common concern from principals was whether the AI output w

184、as accurate or high quality.FIGURE 7Percentage of Principals Reporting the Extent to Which Each Concern Influenced Whether or How They Use AI Tools in Their JobsNOTE:This figure depicts response data from the following survey question:“To what extent have any of the following concerns influenced whe

185、ther or how you use Artificial Intelligence(AI)tools or products for your job as a principal?”We display the percentage of principals who responded,“Minor influence”or“Major influence.”Principals were also able to respond“Not at all.”The universe of respondents was all principals.Percentages within

186、the bars do not always add up to the sums listed due to rounding.N=3,5473,542,depending on the item.0Concerns that AI will make me less efficient in my jobConcerns about my ability to useAI tools and products competentlyLack of district guidance on usingAI tools and productsConcerns about potential

187、biasin AI tools and productsUnsure about how AI tools andproducts could be used for my job10020406080Percentage of principals17424044484142417222023293072707064625922Lack of professional developmentfor using AI tools and productsConcerns about data privacyMinor influenceMajor influence17To What Exte

188、nt Are Schools and Districts Setting Policies for the Use of AI Tools?One in Five Principals Reported That Their School or District Provided Guidance on the Use of AI Tools in the 20232024 School Year,but Guidance Was Less Common in Higher-Poverty SchoolsLast,we asked principals whether their school

189、s or districts were providing guidance on the use of AI among various populations,including students,teachers,and other school staff(see Figure 8).Eigh-teen percent of principals reported that their schools or districts had provided some type of guidance for any of those populations on the use of AI

190、 tools for the 20232024 school year.Thirteen percent of principals in the highest-poverty schools reported that their schools or districts had issued some type of guidance FIGURE 8Percentage of Principals Reporting That Their School or District Provided Guidance on the Use of AI Tools,According to S

191、chool Poverty Status(Based on Percentage of Students Receiving FRPL)Percentage of principalsNOTE:This figure depicts response data from the following survey question:“This school year(202324),has your school or district provided any policy or guidance on how Artificial Intelligence(AI)products and t

192、oolsincluding ChatGPTcan be used?”We display the percentage of principals who responded,“Yes,my school or district has provided policy or guidance on how teachers can use AI products and tools,”“Yes,my school or district has provided policy or guidance on how students can use AI products and tools,”

193、“Yes,my school or district has provided policy or guidance on how other school and district staff besides teachers can use AI products and tools,”and“No,because my school or district is in the process of developing a policy.”We categorized principals as reporting that their school or district provid

194、ed guidance on any use of AI products and tools if they reported that their school or district provided guidance for teachers,students,or other school or district staff members use of AI.An asterisk(*)indicates that percentages for that subgroup significantly differ at the p 0.05 level from the refe

195、rence group before controlling for any principal-or school-level characteristics.The universe of respondents was all principals.N=3,597.School-or district-provided guidanceon any use of AI products and toolsSchool-or district-provided guidance onteachers use ofAI products and toolsSchool-or district

196、-provided guidance on students use of AI products and toolsSchool-or district-provided guidance onother school and district staff membersuse of AI productsand tools010152025305School or districtis in the processof developinga policyAll principals024%2549%5074%75100%Percentage of students eligible fo

197、r FRPL within principals school18141052325*20*14*827*21*16*13*2418*13*84221396215518reported using AI more than their more-experienced counterparts and those in higher-poverty schools.Most teachers who used AI reported doing so for their instructional planning on tasks like creating lesson activitie

198、s,generating assessments,or adjusting texts to different reading levels for their students.Teachers who reported using AI were not typi-cally doing so frequently.For example,the majority of teachers who were using AI tools or products in the 20232024 school year reported using them only once a month

199、 or less often for their instructional planning,and only 20 percent of those teachers reported using AI for that purpose at least once a week(4percent of the entire U.S.sample).Thus,AI is likely not affecting many classrooms in substantive ways at this point.A bigger surprise may be the widespread r

200、eported use of AI tools by principals;nearly 60per-cent of principals reported using AI in their jobs for a variety of purposesmost connected to writingincluding drafting and revising such communica-tions as school announcements to parents,teachers,and students;writing letters of recommendation;and

201、providing teacher feedback.That said,only one in five principals reported that their schools or dis-tricts were providing guidance on the use of AI to teachers,staff,or students.Furthermore,just as with teachers,more principals in lower-poverty schools reported using AI and noted the presence of sch

202、ool-or district-provided guidance on its use to teachers relative to those in higher-poverty schools.Clearly,teachers and principals are finding useful purposes for AI.Similar to such lesson reposi-tories as Teachers Pay Teachers,AI is one of the next big technologies that will affect what happens i

203、n classrooms.Whether AI will improve the quality of instruction and student learning in substantive ways is a separate question and one that is not clari-fied by these data.For example,by generating texts at different reading levels,teachers could be helping students to more easily read that text an

204、d possibly gather information they need to comprehend those texts.Yet,at the same time,teachers could be reduc-ing the level of rigor for students who can and should be reading complex texts regularly;furthermore,just feeding a text into an LLM that is not designed to generate text at a desired read

205、ing level may not reli-in comparison with 25 percent of principals in the lowest-poverty schools.Principals reports of school or district guidance for teachers use of AI tools(14 percent)was slightly more common than guidance for students use(10 percent)or other school or district staff members use(

206、5 percent)of AI tools.Principals in the lowest-poverty schools were about twice as likely as princi-pals in the highest-poverty schools to report the pres-ence of any AI guidance,regardless of whether the guidance was for teachers,students,or other staff.Twenty-three percent of principals also repor

207、ted that their schools or districts were in the process of developing guidance.In addition to reporting less guidance in the 20232024 school year,principals in the highest-poverty schools were less likely to report that their schools or districts were working to develop guidance compared with those

208、in the lowest-poverty schools.This finding suggests that the gap in guidance on the use of AIand eventually the use of AIbetween high-and low-poverty schools may become even wider over time.Summary and ImplicationsAlthough AI tools,such as ChatGPT,were first introduced toward the end of 2022,one in

209、four U.S.teachers reported using such tools in the 20232024 school year,according to the spring 2024 survey data.However,these data also indicate that teachers use of AI tools has not risen much since we last measured it in fall 2023.Given the stability of our results between fall 2023 and spring 20

210、24,we do not think that the much higher estimates on teachers use of AI provided by other surveys(e.g.,Impact Research,2023)are warranted.That said,reporting that average25 percent of teachers use AI toolsmasks a large variation in use among different types of teachers:Nearly 40 percent of teachers

211、whose main subject is ELA or science(most of whom are middle or high school teachers)reported using AI tools or products in the 20232024 school year.Only about one-half that many teachers reported using AI tools when their main subject was math or elementary education.In addition,teachers with less

212、experience and in lower-poverty schools 19this report,we offer the following implications and related recommendations.All Districts and Schools Should Develop Intentional Strategies for Supporting Teachers Use of AI in Ways That Could Most Improve the Quality of Instruction and Student LearningGanim

213、ian,Vegas,and Hess(2020)suggests thatto realize the potential of educational technologydecisionmakers should focus on the“compara-tive advantages”of technology over other sorts of interventions.The authors specifically point to four comparative advantages of technology:“scaling up quality instructio

214、n,”“facilitating differentiated instruction,”“expanding opportunities to practice,”and“increasing learning engagement.”AI technolo-gies,in particular,seem well suited to differentiating instruction to meet the needs of learners at different levels,as well as provide more practice sets for stu-dents

215、and personalize instruction to support student engagement.The teachers in our sample reported all of these uses.Importantly,the extent to which AI will be able to scale up quality instruction will depend more on the platforms that can harness standards-aligned,high-quality instructional activities a

216、s the basis for what is generated,allowing teachers to access curated,high-quality materials rather than sifting through the entire internet for teaching and learning content.For all these reasons,states,districts,and schools should lean in to help both school leaders and teachers understand which A

217、I tools have the most evidence of supporting good instruction and student learning.As noted in the recent toolkit for education leaders(Office of Educational Technology,U.S.Department of Education,2024),the effective use of AI requires leaders to investigate the evidence associated with different AI

218、 applications,listen to the needs of their communities,and design intentional strategies to improve the use of AI applications.The differentiation of reading materials is a case in point.Research suggests that students benefit from expo-sure to complex,grade-level texts,even when they are reading be

219、low grade level(Fisher and Frey,2014;ably yield appropriate or accurate results.In another example,ChatGPT could generate a bevy of assess-ment items or problem sets for students more easily than ever before,freeing up teachers to focus on other aspects of their instruction.However,ChatGPT is not an

220、 arbiter for the quality of those assessments or problem sets.Some teachers in our sample noted that they are taking an active role in revising the out-puts of AI when they do use it to ensure that output meets their students needs.However,the extent to which teachers modifications of AI output supp

221、ort student learning is also unclear.One particular challenge reflected in these data are disparities in the guidance for and use of AI in lower-and higher-poverty schools.Both principals and teachers use of AI was higher in lower-poverty schools relatively to higher-poverty ones,and school guidance

222、 for use of AI was also more frequently pro-vided in lower-poverty schools during the 20232024 school year.Yet,when teachers used AI,teachers in higher-poverty contexts reported using it more frequently.These data suggest that,although teach-ers in higher-poverty schools may not have as much opportu

223、nity or access to use AI,they are more likely to find consistent uses for it when they do use it(e.g.,differentiating instruction).Yet,at the same time,if teachers in higher-poverty contexts are receiving little guidance on the use of AI,they may not be fully apprised of the drawbacks of such tools

224、when they do use them,such as the questionable quality of AI output.In their reflections on how education technol-ogy can improve learning for all,Ganimian,Vegas,and Hess wrote in 2020 that technology has“changed how we communicate,access information,work,and even play.”Yet,“its impact on schools,te

225、aching,and learning has been much more limited”(Ganim-ian,Vegas,and Hess,2020).The authors posit that one reason for this is that technology is often not designed to address critical learning needs or inte-grate clearly into instructional processes or school infrastructure.Relatedly,in 2024,Kleiman

226、and Gal-lagher wrote that generative AI can only support learning if policymakers,technology developers,and educators focus on critical educational challenges and how AI can address them.Taking that research into account,alongside our own investigation in 20tional materials that have been vetted for

227、 their qual-ity,developers could best support more and better uses of those instructional materials.Furthermore,given our findings thus far that educators in high-poverty schools are not taking advantage of AI as much as those in lower-poverty contexts,investors,developers,and decisionmakers should

228、reflect on how to make the most-promising AI developments freely available,alongside profes-sional learning opportunities that target teachers in more disadvantaged and higher-poverty contexts.Those professional learning opportunities should,in particular,help teachers assess the quality of materi-a

229、ls generated by AI and how to use their own pro-fessional judgment to revise AI outputs as needed.Beyond just targeting teachers in these contexts,further research and needs assessments could con-sider other obstacles to teachers use of AI in addition to training and availability.For example,teacher

230、s in higher-poverty contexts could lack the digital infrastructuresuch as adequate internet access or hardwareto support experimentation with AI.Researchers Should Work Hand in Hand with AI Developers to Study Use Cases and Develop a Body of Evidence on Effective AI Applications for School Leadershi

231、p,Teaching,and LearningStates and school systems have an extremely chal-lenging job:to guide the use of AI in schools in a sea of resources with very little evidence on what uses of AI are most effective for improving the quality of instruction and student learning.Policymakers at the federal and st

232、ate levels must provide support for research to guide school systems in this fast-moving environment,in which the majority of principals are using AI for a variety of leadership tasks and one-quarter of U.S.teachers are using AI.Studies should particularly focus on two distinct and very separate use

233、s for AI:(1)innovative uses of AI that are specifi-cally designed to improve teaching and learning and(2)uses of AI that improve efficiency among princi-pals and teachers.Importantly,studies should con-sider whether uses of AI that improve efficiency of a principals or teachers work conflict with or

234、 decrease Morgan,Wilcox,and Eldridge,2000).Although lev-eled texts that differentiate according to students reading levels may also support students reading,a balance between the use of leveled texts and grade-level texts is likely optimal(Liben and Liben,2017).Thus,if teachers are using AI liberall

235、y to adjust the reading level of texts,districts should provide teach-ers with guidance on how to use AI for this purpose in a way that is most helpful to students,about which AI tools may be most appropriate for creating leveled texts,and on how to assess the quality of leveled texts that AI might

236、provide.4 In addition,according to our data,teachers in higher-poverty schools are receiving less guidance on the use of AI,which could create gaps in which teach-ers are able to best harness AI for use in their class-rooms.States could provide more-equitable access to guidance by creating and provi

237、ding such guidance for free across districts,and districts could identify educators with more AI expertise to help provide support across lower-and higher-poverty schools.AI Developers and Decisionmakers Should Consider What Useful Applications of AI Have the Greatest Potential to Improve Teaching a

238、nd Learning and How to Make Those Applications Available in High-Poverty ContextsTeachers are using AI in a variety of ways,potentially without a deep knowledge of which uses of AI are going to drive student learning the most.AI devel-opers,along with state and district policymakers,can drive more b

239、eneficial uses of AI that capitalize on the comparative advantages that AI might offer.The comparative advantages of educational technol-ogy cited by Ganimian,Vegas,and Hess(2020)the potential to scale up good instruction,differentiate,provide more practice opportunities,and engage studentsare a goo

240、d starting point for AI developers and decisionmakers to assess which uses of AI could be most helpful.Teaching Lab,for example,is explor-ing how AI could improve the usability of current standards-aligned curriculum materials(Teaching Lab,undated).By anchoring the use of AI in instruc-21In an open-

241、ended question,we asked teachers to describe how they used AI for planning,but we did not ask them to describe how they introduced AI tools or products to students.Thus,we are unable to determine definitively what teachers were thinking of when they said they introduced AI to students.Third,our anal

242、ysis of the teacher survey data is limited to only those who teach ELA,mathematics,and science,including general elementary teachers of multiple subjects.We are not able to report on the use of AI among teachers of other subjects.Fourth,when analyzing teachers responses to the open-ended question,so

243、me teachers provided input on the usefulness of AI tools and the challenges they experienced with using AI.However,our open-ended question to teachers did not explicitly ask about these topics.Thus,we urge caution in interpreting these findings,because the perspectives expressed by our surveyed teac

244、hers could be more or less prevalent in the general population of teachers than our data cur-rently suggest.Additionally,there could be more rea-sons that teachers find AI useful or challenging to use than are captured by our data.Last,as noted previously,these data do not pro-vide any information a

245、bout whether AI is being used in ways that support student learning.Much more research is needed to understand optimal uses of AI for student learning and how to harness that use across districts and schools.the quality of teaching and learning.To do this work,researchers could partner closely with

246、AI develop-ers to provide formative feedback on the use of AI to improve products and summative feedback to educa-tors and the public to ensure a full understanding of the benefits and drawbacks of particular AI tools.Research LimitationsThis report provides an in-depth look at the use of AI in scho

247、ols and classrooms and the presence of AI policies across schools and districts.However,there are several caveats that readers should consider when interpreting the results presented in this report.First,our analysis of educators responses to close-ended survey items relies on their self-reports on

248、their use of AI tools.These self-reports should be interpreted with caution,because they rely on teachers and principals understanding of what AI entails,especially considering that AI is a term whose meaning has shifted over time and for which experts cannot agree on a universal definition across d

249、isci-plines(Kaplan,2016).Although we defined the term artificial intelligence in our teacher survey,teachers own ideas of what constitutes AI may or may not align with our provided definition.We did not pro-vide a definition for AI in our principal survey,mean-ing principals could similarly have a v

250、ariety of differ-ent interpretations of what constitutes AI.Second,we asked teachers whether they used AI tools or products to plan their instruction or whether they introduced AI tools or products to students.22References Bernard,H.Russell,and Gery Ryan,Analyzing Qualitative Data:Systematic Approac

251、hes,Sage Publications,2010.Brisk Teaching,“Customize Reading Material in Seconds:Adjust Levels and Languages with AI,”webpage,undated.As of December 2,2024:https:/ Council of Chief State School Officers.Council of Chief State School Officers,“CCSSOs High-Quality Instructional Materials and Professio

252、nal Development(IMPD)Network,”webpage,undated.As of December 2,2024:https:/learning.ccsso.org/high-quality-instructional-materialsDiliberti,Melissa Kay,Heather L.Schwartz,Sy Doan,Anna Shapiro,Lydia R.Rainey,and Robin J.Lake,Using Artificial Intelligence Tools in K12 Classrooms,RAND Corporation,RR-A9

253、56-21,2024.As of December 2,2024:https:/www.rand.org/pubs/research_reports/RRA956-21.htmlDoan,Sy,Joshua Eagan,David Grant,and Julia H.Kaufman,American Instructional Resources Surveys:2024 Technical Documentation and Survey Results,RAND Corporation,RR-A134-24,2024.As of January 17,2025:https:/www.ran

254、d.org/pubs/research_reports/RRA134-24.htmlEAB,The Voice of the Superintendent:Eight Themes from Responses to EABs Annual Survey of District Leaders,2024.Fisher,D.,and N.Frey,“Selecting Texts and Tasks for Content Area Reading and Learning,”Reading Teacher,Vol.68,No.7,2015.Ganimian,Alejandro J.,Emili

255、ana Vegas,and Frederick M.Hess,Realizing the Promise:How Can Education Technology Improve Learning for All?Brookings Institution,2020.As of February 5,2025:www.brookings.edu/articles/realizing-the-promise-how-can-education-technology-improve-learning-for-all/Impact Research,“Americans See Need to Be

256、tter Prepare Students for National Security Careers of the Future,”memorandum,July 18,2023.Kahoot!,“AI-Powered Features in Kahoot!”webpage,undated.As of December 2,2024:https:/ Intelligence:What Everyone Needs to Know,Oxford University Press,2016.Kaufman,Julia H.,Jill S.Cannon,Shelly Culbertson,Magg

257、ie Q.Hannan,Laura S.Hamilton,and Sophie Meyers,Raising the Bar:Louisianas Strategies for Improving Student Outcomes,RAND Corporation,RR-2303-BRAF,2018.As of December 2,2024:https:/www.rand.org/pubs/research_reports/RR2303.htmlKaufman,Julia H.,V.Darleen Opfer,Michelle Bongard,and Joseph D.Pane,Change

258、s in What Teachers Know and Do in the Common Core Era:American Teacher Panel Findings from 2015 to 2017,RAND Corporation,RR-2658-HCT,2018.As of December 2,2024:https:/www.rand.org/pubs/research_reports/RR2658.htmlKleiman,Glenn M.,and H.Alix Gallagher,“State Education Policy and the New Artificial In

259、telligence,”State Education Standard,Vol.24,No.3,2024.Notes1 Impact Research did not report how it selected its teacher sample.This makes it difficult to compare results from that survey with the RAND survey,which relies on a probability-based,randomly selected sample of teachers across the United S

260、tates.2 Readers should keep in mind that not all elements of Khan Academy are powered by AI.For instance,Khan Academy has a library of instructional videos and lessons that,alone,might not be considered AI,but it also leverages an adaptive learning approach that personalizes learning for students.In

261、 addition,Khan Academy includes a feature called Khanmigointroduced in March 2023that can generate a variety of teacher-facing and student-facing resources for teachers.3 About 80 percent of those teaching grades K5 indicated that“elementary education”was their main subject;the remaining 20 percent

262、of K5 teachers reported that a specific subject,such as ELA,math,or science,as their“main subject.”We assume that the remaining 20 percent are departmentalized elementary teachers who focus on specific subjects rather than multiple subjects.Conversely,more than 80 percent of the math and ELA teacher

263、s and more than 90 percent of the science teachers in our sample are secondary teachers.4 One study demonstrated that ChatGPT 3.5 and 4.0 do not reliably create readable education materials for patients(Marder et al.,2024).At the same time,MagicSchool and Brisk Teaching,among others,have recently cr

264、eated AI-powered text-leveling tools.More study may be needed to understand how these tools work to provide accurate texts at the desired grade level(Brisk Teaching,undated).23Liben,David,and Meredith Liben,“Both and”Literacy Instruction:A Proposed Paradigm Shift for the Common Core ELA Classroom,St

265、udent Achievement Partners,last updated December 15,2017.As of February 5,2025:https:/achievethecore.org/content/upload/Both%20And%20Literacy%20Instruction%20K-5%20%20A%20Proposed%20Paradigm%20Shift%20for%20CCSS%20ELA%20and%20Literacy.pdfMarder,Ryan S.,George Abdelmalek,Sean M.Richards,NicolasJ.Nade

266、au,Daniel J.Garcia,Peter J.Attia,Gavin Rallis,and Anthony J.Scillia,“ChatGPT-3.5 and-4.0 Do Not Reliably Create Readable Patient Education Materials for Common Orthopaedic Upper-and Lower-Extremity Conditions,”Arthroscopy,Sports Medicine,and Rehabilitation,October 9,2024.Morgan,Alisa,Bradley R.Wilco

267、x,and J.Lloyd Eldredge,“Effect of Difficulty Levels on Second-Grade Delayed Readers Using Dyad Reading,”Journal of Educational Research,Vol.94,No.2,2000.National Center for Education Statistics,“202223 Common Core of Data(CCD)Universe Files,”dataset,U.S.Department of Education,undated.Office of Educ

268、ational Technology,U.S.Department of Education,Empowering Education Leaders:A Toolkit for Safe,Ethical,and Equitable AI Integration,October 2024.As of February 5,2025:https:/tech.ed.gov/files/2024/10/ED-OET-EdLeaders-AI-Toolkit-10.24.24.pdfOpfer,V.Darleen,Julia H.Kaufman,and Lindsey E.Thompson,Imple

269、mentation of K12 State Standards for Mathematics and English Language Arts and Literacy:Findings from the American Teacher Panel,RAND Corporation,RR-1529-1-HCT,2016.As of December 2,2024:https:/www.rand.org/pubs/research_reports/RR1529-1.htmlRay,Kecia,“AIs Big Deal:AI in the Classroom Continues to E

270、volve,”Tech&Learning,June 26,2023.As of December 2,2024:https:/ Pandey,David Grant,Julia H.Kaufman,Elizabeth D.Steiner,and Dorothy Seaman,American Mathematics Educator Study:2024 Technical Documentation and Survey Results,RAND Corporation,RR-A2836-4,2024.As of January 9,2025:https:/www.rand.org/pubs

271、/research_reports/RRA2836-4.htmlTeaching Lab,homepage,undated.As of January 24,2025:https:/www.teachinglab.ai/RR-A134-25RAND is a research organization that develops solutions to public policy challenges to help make communities throughout the world safer and more secure,healthier and more prosperou

272、s.RAND is nonprofit,nonpartisan,and committed to the publicinterest.Research IntegrityOur mission to help improve policy and decisionmaking through research and analysis is enabled through our core values of quality and objectivity and our unwavering commitment to the highest level of integrity and

273、ethical behavior.To help ensure our research and analysis are rigorous,objective,and nonpartisan,we subject our research publications to a robust and exacting quality-assurance process;avoid both the appearance and reality of financial and other conflicts of interest through staff training,project s

274、creening,and a policy of mandatory disclosure;and pursue transparency in our research engagements through our commitment to the open publication of our research findings and recommendations,disclosure of the source of funding of published research,and policies to ensure intellectual independence.For

275、 more information,visit www.rand.org/about/research-integrity.RANDs publications do not necessarily reflect the opinions of its research clients and sponsors.is a registered trademark.Print and Electronic DistributionRightsThis work is licensed under a Creative Commons Attribution 4.0 International

276、License.All users of the publication are permitted to copy and redistribute the material in any medium or format and transform and build upon the material,including for any purpose(including commercial)without further permission or fees being required.For more information on this publication,visit w

277、ww.rand.org/t/RRA134-25.2025 RAND CorporationAbout This ReportDrawing on American Instructional Resources Surveys data from spring 2024,the authors examine the use of AI tools and products among teach-ers and principals and the presence of school policies on the use of AI.The American Educator Panel

278、s are nationally representative samples of teachers,school leaders,and district leaders across the country.The panels are a proud member of the American Association for Public Opin-ion Researchs Transparency Initiative.If you are interested in using AEP data for your own surveys or analysis or in re

279、ading other publications related to the AEP,please email aeprand.org or visit www.rand.org/aep.RAND Education and LaborThis study was undertaken by RAND Education and Labor,a division of RAND that conducts research on early childhood through postsecond-ary education programs,workforce development,an

280、d programs and policies affecting workers,entrepreneurship,and financial literacy and decisionmaking.This report is based on research funded by the Gates Foundation,the Charles and Lynn Schusterman Family Foundation,and the Walton Family Foundation.The findings and conclusions presented are those of

281、 the authors and do not necessarily reflect positions or poli-cies of the foundations that supported this research.More information about RAND can be found at www.rand.org.Questions about this report should be directed to jkaufmanrand.org,and questions about RAND Education and Labor should be direct

282、ed to educationandlaborrand.org.Acknowledgments We are extremely grateful to the U.S.public school teachers and leaders who agreed to participate in the panels.Their time and willingness to share their experiences are invaluable for this effort and for helping us understand more about how to better

283、support their hard work in schools.We thank Daniel Ibarrola and Brian Kim for serving as survey man-agers,Casey Hunter for serving as the survey data manager,and Julie Newell and Tim Colvin for programming the survey.Thanks to Joshua Eagan for producing the sampling and weighting for these analyses.

284、We also greatly appreciate the administrative support provided by Tina Petrossian.We thank our quality assurance manager,Ben Master,for providing helpful feedback that improved this report.We also thank our reviewers,Christopher Doss and Kristen Pilner Blair,for helpful feed-back that improved this report.Last,we thank Maria Vega for her edito-rial expertise and Monette Velasco for overseeing the publication process for this report.www.rand.org