Engaging Teachers to Co-Design Integrated AI Curriculum for K-12 Classrooms
EEngaging Teachers to Co-Design Integrated AI Curriculumfor K-12 Classrooms
JESSICA VAN BRUMMELEN ∗ , Massachusetts Institute of Technology, USA
PHOEBE LIN ∗ , Harvard University, USA
Fig. 1. Representations of the four integrated curricula [6, 9, 20].
Left : The “exemplar” physics and AIcurriculum.
Mid-left : Social studies and AI curriculum.
Mid-right : ESL and AI curriculum.
Right : Literacy andAI curriculum for students with learning disabilities.
Artificial Intelligence (AI) education is an increasingly popular topic area for K-12 teachers. However, littleresearch has investigated how AI education can be designed to be more accessible to all learners. We organizedco-design workshops with 15 K-12 teachers to identify opportunities to integrate AI education into corecurriculum to leverage learners’ interests. During the co-design workshops, teachers and researchers co-created lesson plans where AI concepts were embedded into various core subjects. We found that K-12 teachersneed additional scaffolding in the curriculum to facilitate ethics and data discussions, and value supports forlearner engagement, collaboration, and reflection. We identify opportunities for researchers and teachersto collaborate to make AI education more accessible, and present an exemplar lesson plan that shows entrypoints for teaching AI in non-computing subjects. We also reflect on co-designing with K-12 teachers in aremote setting.CCS Concepts: •
Human-centered computing → Participatory design ; User centered design .Additional Key Words and Phrases: Artificial intelligence, K-12 education, co-design workshop
ACM Reference Format:
Jessica Van Brummelen and Phoebe Lin. 2020. Engaging Teachers to Co-Design Integrated AI Curriculum forK-12 Classrooms. 1, 1 (September 2020), 12 pages.
Artificial intelligence (AI) education is becoming an increasingly popular subject in the eyes ofeducators due to the rapid integration of AI technologies in user-facing services and products[16, 34, 41]. Researchers have called for formal K-12 education to prioritize AI literacy and teachchildren to interact with AI using a critical lens [42]. The AI4K12 research community has alsopublished guidelines for what AI concepts K-12 curriculum should cover, known as the
Big AI ∗ Both authors contributed equally to this research.Authors’ addresses: Jessica Van Brummelen, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge,MA, USA, [email protected]; Phoebe Lin, Harvard University, 13 Appian Way, Cambridge, MA, USA, [email protected]. a r X i v : . [ phy s i c s . e d - ph ] S e p Jessica Van Brummelen and Phoebe Lin
Ideas , and calls for AI researchers to help teachers and students understand AI [35]. As childreninteract more with AI technologies, it is critical that they are able to recognize AI, understand howAI algorithms work, use those algorithms to solve problems meaningful to them, and evaluate theimpact of AI technologies on society [5].Teachers of all subjects should feel empowered to teach AI curriculum, yet teachers often feelthey lack sufficient understanding to teach AI or the capacity to include more curriculum on top oftheir existing curriculum [39]. Despite the proliferation of tools and AI curriculum in response to therecent calls to action, few are widely implemented due to challenges in the classroom that preventthese curricula from being accessible [27]. In order to introduce new practices, researchers anddevelopers should consider the contexts of teachers and invest in additional supports to facilitatethe accessibility of AI resources for teachers.Similarly, AI as a discipline can span many other topics, such as government, journalism, andart [10, 18, 32], therefore AI should not be confined to just computing subjects such as computerscience or data science. Tools and curriculum today often teach AI as an extension of computerscience curricula or as standalone curricula that is difficult to adjust to other contexts [8, 25, 33].Adapting those tools and curriculum then becomes especially difficult for teachers who teach coresubjects, including English, math, social studies, and science, and may not have any AI experience.The lack of integrated AI curricula in core subjects has become one of the barriers to exposing AIto students with little access to computing disciplines.In this paper, we partner with K-12 teachers to design AI curriculum that is integrated with coresubjects. We aim to empower all teachers to incorporate AI into their classrooms and leveragelearners’ interests for other subjects through a two-day co-design workshop with 15 teachers fromdifferent schools. We set out to understand what is necessary and valuable to K-12 teachers toeffectively implement integrated AI curricula, and co-create lesson plans that address those needsand values. Specifically, our research questions are:
RQ1:
How might we address K-12 teachers’ values and considerations when designing AIcurriculum? (Teaching needs)
RQ2:
How might AI be integrated into core subject curriculum? (Integrated curriculum design)To answer these research questions, we organized a multi-session workshop that spanned twodays with fifteen teachers who teach various subjects. The first day of the workshop involvedpresentations and group discussions to level set everyone’s basic understanding of AI. Betweenthe first and second day of the workshop, participants were asked to complete a brainstormingassignment where they identified curriculum of their own to use as a potential base for an integratedAI curriculum. During the second day of the workshop, we split participants into three smallgroups to work together and design a lesson plan that integrates AI into a non-computing subjectcurriculum. The co-design process revealed when teachers design curriculum, they consider fourpractical needs: evaluation, engagement, logistics, and collaboration. Furthermore, our analysis ofthe co-designed lesson plans showed opportunities for connections between AI and a core subject,with three points of integration: data, reflection, and scaffolding for ethics.The contributions of this work are (1) identifying the values and needs of K-12 teachers teachingAI in the classroom and opportunities to address them, (2) showing an exemplar integrated AIcurriculum as an output of the co-design session [6], and (3) reflecting on co-design sessionsinvolving K-12 teachers in a remote setting to solicit design considerations of AI curriculum.
To the authors’ knowledge, there are no papers describing a co-design process with teachers tointegrate AI concepts into core curriculum, and few papers that purposefully integrate AI concepts ngaging Teachers to Co-Design Integrated AI Curriculum for K-12 Classrooms 3 into core curriculum. Other related research includes the development of AI education tools andcurricula, as well as co-design of other course materials with teachers.
Many K-12 AI tools and curricula exist as standalone products or extend computer science curricu-lum. Two widely-used AI teaching tools are
Teachable Machine [9] and
Machine Learning for Kids [20], which empower learners to develop classification models without needing to program. Otherstandalone AI teaching tools include
Any-Cubes , which are toys to teach machine learning (ML)concepts [29];
Calypso for Cozmo , which is AI curriculum for a toy robot [37]; and extensions for
MIT App Inventor , which enable students to develop AI-powered mobile apps [23]. Each of thesetools could be integrated and taught in core classes; however, are presented as standalone AI tools.In terms of K-12 AI education research involving instructors, most involve researchers rather thanK-12 teachers as the instructors, and likely miss valuable expertise and feedback from professionalswho have worked in the classroom. Nevertheless, some works involving K-12 teachers include anAI summer program for high school girls [38], an AI engineering course for high school students[31], and a STEM workshop for middle school students [28]. Each of these studies saw value inengaging with K-12 teachers.Other works have also integrated core curriculum content into AI tools and curricula; however,most of these involve researchers as instructors and are often not in regular classroom settings. Forexample, one physical education curriculum involves students developing sports gesture classi-fication models with researchers as facilitators [43]. Another science-based curriculum involvesstudents teaching a conversational agent about animals, and observing it classify the animals intoecosystems with researchers as facilitators [21]. Although these works are state-of-the-art in K-12AI education, it is unknown whether they are suitable for K-12 classrooms, since they have notbeen tested in regular classrooms and teachers were not involved in the design process.In our literature review, we found one example of AI curriculum that was both integrated intoa core course and designed or taught by K-12 teachers alongside researchers. This curriculuminvolved AI and science concepts, and was taught in Australian K-6 classrooms [17]. Although thisexample is insightful, much further research is needed to integrate teacher expertise and addresswidespread, integrated AI curriculum in K-12 classrooms [39].
Although K-12 AI education has not yet benefited from tools and curriculum co-designed with K-12teachers, other areas of education have. For example, in one study, researchers collaborated withteachers to develop new science curriculum materials. Researchers recognized the value in teachers’K-12 expertise and in promoting their agency throughout the design process [30]. Another sciencecurriculum co-design study argued that the process of working with teachers had substantial effectson adoption of the tools and curricula, in addition to bringing social value and innovative ideas[13]. In order to catalyze such benefits, one paper presents key considerations to co-designingwith teachers. These include addressing a “concrete, tangible innovation challenge”, investigating“current practice and classroom contexts”, and involving a “central accountability for the quality ofthe products of the co-design”, among others [27]. In our study, we utilize these considerations andpresent a co-design for AI-integrated core curricula development.
We conducted a two-day co-design workshop with fifteen instructors, ranging from K-12 teachersto educational directors. Participants completed pre-work before each day’s activities, as well aspre- and post-workshop surveys. The co-design activity was split into three smaller group sessions
Jessica Van Brummelen and Phoebe Lin to enable us to better identify differences in value and process of different teachers. This study wasapproved by the universityâĂŹs Institutional Review Board (IRB).
Fifteen teachers participated in the study, whom we recruited from a mailing list and our personalnetwork. The only inclusion criteria was that they teach or previously taught in a K-12 classroomand were able to commit to the time and pre-work for the two-day workshop. Seven participantsidentified as female, four participants identified as male, and the remaining did not say. Their ageranged from 25 to 50 (M = 40.6, SD = 11.6). In our selection process, we prioritized participantswho primarily taught non-computing subjects, such as English language arts (ELA), and thenparticipants who taught computer science, with the idea that small group sessions could havediverse perspectives. Their work background is detailed in Tab. 1. All participants provided informedconsent to participate in compliance with our institutionâĂŹs IRB. As the workshop was conductedin groups, we collected participants’ availability and selected times where the greatest numberof participants could join. Each day of the workshop lasted 2.5 hours. Every workshop sessioninvolved two researchers.
Table 1. Participants were selected to represent diverse profiles and/or subject areas.
ID Grade taught Subject taught Location
P1 6th grade English Language Arts (ELA) North Carolina, USAP2 5th grade Science Connecticut, USAP3 6th-8th grade Computer Science Tunisia, North AfricaP4 9th-12th grade Computer Science Cuneo, ItalyP5 9th-12th grade Chemistry and Math British Columbia, CanadaP6 6th-8th grade STEM Florida, USAP7 6th-8th grade STEM Florida, USAP8 - STEM Pennsylvania, USAP9 6th-8th grade Computer Science California, USAP10 9th-12th grade Career Exploration Rhode Island, USAP11 9th-12th grade Computer Science Massachusetts, USAP12 9th-12th grade Library Science Rome, ItalyP13 6th grade History California, USAP14 6th-9th grade Computer Science TurkeyP15 6th-12th grade English as a Second Language (ESL) Pennsylvania, USA
The entire co-design workshop spanned two days, Session 1 on the first day and Session 2 onthe second day. Session 1 consisted of discussions and a "What is AI" presentation to level set allparticipants (see Tab. 2), and Session 2 consisted of the co-design activity and an ethics presentation(see Tab. 3). Here, we describe our rationale and the activities in detail.
Before the first session, we asked participants to complete a pre-workshop question-naire asking about participants’ familiarity with AI, whether they have taught AI in the classroombefore, and if so, what their experience was. This was to understand their backgrounds and enableus to tailor the content of Session 1 appropriately. Participants were also given detailed instructionson how to install and use Zoom [3], Slack [4], and Miro [2]—the tools used throughout the entireworkshop. We started Session 1 with breaking participants into small groups on Zoom to discusswhy participants thought AI is or is not important to teach their students. Having them describe ngaging Teachers to Co-Design Integrated AI Curriculum for K-12 Classrooms 5 what and why AI was important allowed us to understand their preconceptions about AI and theirpriorities as teachers. During the “Let’s learn AI” presentation, participants learned the
Big AIIdeas [36], categories of AI, and how to recognize what is and is not AI. During the “Let’s learnAI tools” presentation, we demoed four distinct AI learning tools and provided participants withresources and links to explore further. We then used Miro for a card sorting activity [40], wherewe asked participants to generate categories for Google’s A to Z of AI cards [1], where categorieswere limited to subjects taught in the classroom. The card sorting activity showed participants’enthusiasm for integrating AI topics into every classroom subject, including English language arts(ELA), writing and reading, social studies, math, science, economics, and social-emotional learning.
Table 2. Schedule for Session 1
Time Activity15 min Introduction20 min Why AI? (Discussion)50 min Let’s learn AI! (Presentation)15 min Break25 min Card sorting activity25 min Let’s learn AI tools! (Presentation)
Participants were asked to complete “pre-work” before Session 2. Participantshad two days to complete their pre-work between Session 1 and Session 2. The pre-work askedparticipants to explore the rest of the AI learning tools, select one of the tools to go along with acurriculum they currently use or have used in their classrooms, and identify areas where they seepotential to teach AI using the selected tool. Participants uploaded their submissions into a sharedGoogle Drive folder. Participants had access to the workshop Google Drive folder, which containedall of the presentations and resources from Session 1, at all times, and could also post questions inthe workshop Slack group, which was monitored closely by the researchers. From the pre-worksubmissions, we selected one idea to develop into an exemplar curriculum (see [6]).For Session 2, participants were split into three groups of 4-5. Each group was asked to analyzethe exemplar curriculum and discuss what they noticed. The co-design activity part 1 then beganwith each group responding to a prompt asking them to devise integrated AI curricula for specificsubjects. We created the prompts from the pre-work submissions and organized the groups suchthat each would have a domain expert. For example, the group responding to the prompt askingparticipants to create a curriculum for students who are learning English as a Second Language(ESL) had an ESL teacher, who would be familiar with ESL students’ needs. Each group was alsopaired with a researcher, who provided technical input and answered participants’ questions aboutAI or learning tools. During the “Ethics & Diversity” presentation, we presented definitions ofAI ethics, diversity statistics within the field, and resources for teaching and learning AI ethics.Participants then continued working in their groups on their integrated curriculum in co-designactivity part 2.Every group was successful in producing a first draft of an implementable AI curriculum thatintegrated with a core subject. The drafts can be found in the appendices [6]. Lastly, participantsdiscussed why they thought AI was or was not important to teach for a second time, which actedas a reflection and a way to see if their mindset or preconceptions changed after the workshop.Participants were asked to complete a post-workshop questionnaire that asked how familiar theywere with AI, how comfortable they felt teaching AI in their class, as well as feedback on theworkshop itself and their demographics (i.e. age, gender, ethnicity).
Jessica Van Brummelen and Phoebe Lin
Table 3. Schedule for Session 2
Time Activity60 min Co-design activity part 115 min Break40 min Ethics & Diversity (Presentation)20 min Co-design activity part 215 min Why AI? (Discussion)
Our dataset consists of the audio recordings of the entire co-design workshop, participant ques-tionnaires, and the deliverables of each participant, which include their pre-work submissions andtheir group work during the co-design activity. All audio recordings were transcribed to text andthematically coded by two researchers using open coding. We specifically examined their process,priorities, and challenges.Nine out of 15 participants had never taught AI in the classroom. While some participantshad had experience teaching AI, they were interested in learning how to allow non-CS studentsexperience AI and integrate AI into their teaching. Participants came into the workshop ratingtheir own familiarity with AI an average of 4.8 out of 7, and finished the workshop with an averagerating of 5.8 out of 7. Teachers also rated their confidence about integrating AI into their owncurriculum with an average rating of 5.6 out of 7.
Our teacher participants teach students with diverse needs. The co-design activity prompted richdiscussion with three groups completing three curriculum drafts that integrated AI with a topicof their choice. The topics were: (1) “How Does Data Affect Government Policy?” (Social StudiesCurriculum), (2) “Learn Vocabulary with an AI” (Literacy curriculum for students with learningdisabilities), and (3) “Build an AI-powered Pronunciation Application” (ESL curriculum), as shownin the appendices [6]. During the co-designing process, all groups shared certain considerations forthe curriculum, though each group addressed them differently. In the first section of results, weanswer the first research question by outlining what the shared values and considerations wereand showing how each group addressed them. We then answer the second research question byshowing how each curriculum effectively integrated AI.
We identified four categories of values and considerations that our teachers had while creating thecurriculum drafts:
Evaluation , Engagement , Logistics , and
Collaboration . All groups considered student evaluation to be critical to a curriculum. Teacherswanted to see evidence for learning and know their students understand relevant concepts correctly.To do so, teachers first considered their own objectives: “Do we have an end goal in mind, or like,what do we consider a success?” (P9). In the ESL curriculum, P12 referred to the
Big AI Ideas toidentify the what the group called the “AI objective”. P12 and P15 also frequently referred to theexemplar curriculum, suggesting that teachers require frameworks and scaffolding to devise the AIobjective. To evaluate students, P5 and P12 both suggested non-traditional forms of evaluation,such as an “exit interview or on-the-fly assessments where students talk through all of the details,so we get a really good idea from a conversation with them whether they understand what they ngaging Teachers to Co-Design Integrated AI Curriculum for K-12 Classrooms 7 were doing” (P5) and “an engineer’s log where you’ve got their design and you’ve got to do it allofficial” (P12). In these drafts, teachers wanted to evaluate students on their conceptual knowledge,and not on their technical knowledge.
In a K-12 setting, engagement tends to be particularly challenging, which was aconcern for our teachers. P8 and P10 grounded the Social Studies curriculum in law and governmentdiscourse by having students review an article around the Crown Act. Introducing context to theproject gives students an “anchor” (P8) or hook to prompt further inquiry. Other anchors includedasking students the “hard questions” about real-world applications of AI, such as “how do Siriand other personal assistants get to be at that point?” and “who used the machine learning anddesigned the app?” (P7). P5 and P15 both mentioned student-driven learning as a way to leveragestudents’ interests. For example, “I can see a sixth grader coming in and going, I went to the baseballgame and I couldn’t say all these words. And they decide they’re going to do baseball that day”(P12). Lastly, multiple groups brought up competition and gamification as effective methods ofengagement: “the class creates a game that students use to quiz themselves on vocab by trying tobe better than the system” and “module 1 can be a rock paper scissors game so that students getfamiliar with the interface” (P2).
By logistics, we mean factors that enable the curriculum to be smoothly run in theclassroom. Teachers tended to think about how the lesson itself would take shape before addressingwhich core standards the lesson intended to cover. For example, at the beginning of the co-design,P10 explained that what would be most beneficial was “thinking of how to structure the lesson andwhat resources we can use to pull in to have the engagement component”. Most teachers struggledwith identifying which technology resources and learning tools to use, for example, whether to use
Machine Learning for Kids [20] or
Google Quick Draw [19]. Our teacher participants generally lookedto the researcher for guidance, suggesting that tools can be more explicit about when and how theycan be applied in K-12 classrooms. Teachers also paid close attention to grade-level considerations.They felt more comfortable having older students drive their own learning, but recognized evenyounger students are capable of deep reflection: “posing some challenging questions will vary alittle depending on age, but you can get pretty deep with some—even fifth graders. They can getinto this, and I think it’s a good way of opening the door” (P15).
All groups discussed the value of collaboration. In the ESL curriculum, teach-ers had their students collect data in groups and input the data into multiple models using MachineLearning for Kids. In the literacy curriculum, teachers had every student contribute 10 imagesto a class dataset to input into Google Teachable Machine. The presence of group work not onlyhelps overcome the need to create many training examples for a machine learning model, but alsoprovides students with opportunities to discuss design and ethics decisions with their peers andteacher. This also aligns with Long and Magerko’s design consideration for
Social Interaction [22].Teachers also consider how collaboration can be implemented most effectively when designingcurricula. For example, P8 described how “it’s important to think about the group size becauseyou want to make sure that students have a voice in the work. And when you start doing largegroup things those kids that process information internally never get to be heard.” She went on todescribe how, in her experience, “duos [of students] work really, really well” and how it is generallybetter to “go with smaller groups [of students in the classroom], but if you’re using technology [...]you’re bound by what you have.” Thus, it is important to consider how AI tools can best facilitategroup work to ensure all students have a chance to contribute and learn.
Jessica Van Brummelen and Phoebe Lin
During the co-design, teachers made connections between the core subject material (e.g., socialstudies) and AI in three main ways: (1) relating an AI tool or concept to the core subject , (2)relating content from the core subject to AI , and (3) noticing overlapping concepts in AIand the core subject . For example, P14 related the AI tool, Arbitrary Style Transfer [24], to thecore subject of history when he said, “If we give an image as input and try to modify [it] accordingto the old art [using] Style [Transfer][...] This can give us an idea about the history when we lookat the picture, [...] but if you change the picture, the students may understand how people thoughtin the past”. Other teachers related real-life applications of AI to core subjects, like how
YouTube suggestion algorithms can be “tunnel visioned” in what they suggest, similar to how people can be“tunnel visioned” when considering politics or how recidivism risk analysis algorithms [7] can berelated to social studies concepts (P10).Teachers also often made connections by starting with a core subject concept and relating itto AI. For instance, one teacher connected physics data from one of their student’s 3D printingprojects to an AI flight prediction algorithm (P12). The same teacher also started with an Englishunit and asked, “What tools do we know that we [can] connect to language?”, ultimately connectingEnglish to a Shakespeare natural language processing algorithm. Another teacher began with theELA concept of “argumentation” and connected it to the reflection and “data analysis” processes inAI (P8).In terms of overlapping concepts between AI and core subjects, teachers often found connectionsusing the
Big AI Ideas [36]. For instance, the Big AI Ideas of
Societal Implications and
Representationand Reasoning are also core concepts in social studies. The AI concept of iterative development inML was also directly connected to the social studies concept of iterative opinion making through“go[ing] back and forth” (P8) and adjusting beliefs.Using these methods of connection, participants co-designed integrated curricula containing AIconcepts and supports for teaching core subject requirements. The curricula contained three mainpoints of integration: (1) data , (2) reflection , and (3) ethics . Educational activities often produce data, and AI systems often require data. Thisprovides an obvious access point for AI systems to be integrated into ready-made educationalactivities. In our co-design workshops, participants used this fact to generate integrated curriculum.For instance, in the exemplar curriculum (see Fig. 1 and [6]) (which was based on a teacher’s ideaduring the workshops) students would produce data as they construct airplanes for a physicsactivity. The paper airplane dimensions and time-of-flight data would then be used to train a MLmodel to predict the effectiveness of other potential paper airplanes, combining AI systems andphysics concepts into a single curriculum.For the ESL integrated curriculum, students produced data as they were practicing word pro-nunciation, which was then utilized in a pronunciation teaching app. For example, students wouldcreate data by recording saying a word correctly (as guided by a teacher) and incorrectly, whichwould then train a classification model for an app developed in
MIT App Inventor [23]. This appwould then be used to further help students learn correct pronunciation. Future AI-integratedcurricula might consider utilizing the data inherent in core curricula activities, such as speechpronunciation data, to teach AI. This may be through teaching data-related AI competencies, like
Data Literacy , Learning from Data , and
Critically Interpreting Data [22], or through using data totrain ML models, which can teach other AI competencies.For the literacy curriculum for students with learning disabilities, participants also used data fromcore curriculum—vocabulary words—to integrate AI concepts. From the vocabulary words, studentswould find relevant images, generating further data, and use this to train a classification model. ngaging Teachers to Co-Design Integrated AI Curriculum for K-12 Classrooms 9
This addressed the aforementioned data-related AI competencies, as well as other competencies,including the
ML Steps and
Human Role in AI , in addition to relevant English literacy concepts.
Another point of AI integration was student reflection on core curriculum contentand AI methods. Many common core standards as well as AI competencies can be addressed throughstudent reflection. For example, the common core standard, 1-ESS1-1: “Use observations of the sun,moon, and stars to describe patterns that can be predicted.” [26], and the AI concept, “Learningfrom Data”, could be addressed by reflecting on patterns in a constellation classification model’sinput and output. In the exemplar curricula, students were asked to reflect on what did and didnot work and why, and on the real-world implications of a biased dataset in airplane development.This reflection addressed both a standard from the common core, 3-5-ETS1-3: “Plans and carriesout fair tests in which variables are controlled and failure points are considered to identify aspectsof a model or prototype that can be improved” as well as a number of AI literacy competencies,including
AI Strengths & Weaknesses , Critically Interpreting Data , and
Ethics [22].Teachers also used this method to integrate AI concepts into the social studies curriculum. Forexample, students were asked to reflect on the amount of data in each image category, socialnorms and peer opinion, people’s ability to access resources, and consensus agreement in thiscurriculum. These reflection questions address a number of the AI competencies, including
DataLiteracy , Critically Interpreting Data , and
Ethics [22], as well as core social studies and Englishlanguage arts standards, including NSS-EC.5-8.1:
Scarcity , NSS-C.5-8.3:
Principles of Democracy ,NL-ENG.K-12.4:
Communication Skills , and NL-ENG.K-12.7:
Evaluating Data [14].
The final point of integration we present is through ethics, which is one of the AIliteracy competencies [22]. Ethics can also be found in many common core standards [14, 26]. Forexample, environmental ethics can be found in life science standards (K-ESS3-3: “Communicatesolutions that will reduce the impact of humans on the land, water, air, and/or other living things inthe local environment.”), and engineering standards (MS-ETS1-1: “Define the criteria and constraintsof a design problem [...] taking into account [...] potential impacts on people and the naturalenvironment”) [26]. Furthermore, social justice principles, which are highly related to AI ethics,are commonly advocated for within standards-based K-12 education [11, 12]. By teaching ethicalprinciples with respect to AI, teachers can also address standards related to the common core.Each curricula designed in the workshops had an ethics component. In the exemplar, studentswould engage in a brainstorming session about how AI bias affected the accuracy of ML modelsand relevant implications in the real world. Similarly, the ESL curriculum addressed ethics throughdiscussing AI bias, socioeconomic norms for “correct” pronunciations, and the implications of anAI system judging people’s pronunciations in the real-world. The social studies curriculum wasdeveloped around the ethics of the “CROWN Act” [15], what it means for students to design AIalgorithms to classify outfits and hairstyles as “professional” or “unprofessional”, and how thismight affect different people groups. The literacy curriculum for students with learning disabilitiesaddressed ethics through discussion about the accuracy of the image classification system andreasons for any bias observed. Each of these curricula touched on environmental, social justice orother ethical issues, addressing both AI and common core ethics standards.From the workshops, we found that teachers were highly interested in teaching ethics (e.g., thesocial studies curricula was entirely focused on ethics); however, they also seemed apprehensiveabout actually implementing ethics activities in the classroom. For example, P5 described howthere is a “barrier that comes up for teachers” when “ kids often bring ethics up with questionsand sometimes teachers will avoid it because theyâĂŹre afraid to say something wrong [...] eventhough those discussions would be so rich.” Nevertheless, P5 also mentioned how if it was in a“planned lesson”, it would be “less scary because you know what youâĂŹre going to say”. Designing scaffolding for AI ethics lessons would not only enable core curricula integration, but would alsoempower teachers to more confidently teach students about ethics.
Due to Covid-19, we organized and ran this co-design workshop completely remotely. Amongour activities, the perceived helpfulness from most to least helpful was: Presentations (11 votes),Co-design activity (9 votes), Why AI? and Ethics discussions (both 7 votes), and the Card sortingactivity (4 votes). Participants also indicated meeting like-minded educators from around the worldand having access to the list of tools and links to be particularly rewarding takeaways. Overall, wenoticed a slight increase in familiarity with AI after the workshop and a high level of confidence forintegrating AI into their classrooms, though we did not establish that baseline. When asked if theworkshop changed their opinion about teaching AI, teachers cited “introducing AI is the gateway toso much learning...now I am seeing and starting to understand the vast world of opportunities thatexist for coding beyond being video game designers” (anonymous), as well as seeing the necessityof teaching AI and understanding that AI can be accessible to not “just the computery people”(anonymous).At the beginning of the workshop, we established norms as an entire group to make facilitatingeasier. For example, setting expectations for “warm” calling to ensure equal representation of voicesin the room meant participants expected to be called on to share their thoughts. Other normsincluded being present, having discussions in breakout rooms, and keeping cameras on. Since mostteachers were unfamiliar with teaching AI, we grouped them into smaller groups of 4-5 so that eachgroup could have a researcher co-designing with them. However, this meant some teachers workedon curriculum that was unrelated to their discipline. We believed this trade off was necessarygiven the complexity of the task, and was mitigated by the benefits of collaboration. This setupmay have worked better if teachers from the same school joined, and groups could be formed byschool. Several teachers also requested more time to play with the AI learning tools and digest thepresentations. This could have been addressed by scheduling more time between Session 1 and 2,so teachers would have more time to complete the pre-work for Session 2. One suggestion from aparticipant was to introduce the AI tools using a jigsaw game where every teacher explores anassigned AI tool and presents it back to the group.
The above findings contribute to the under-explored need to collaborate with teachers whendesigning AI curriculum, as well as the potential for AI to be integrated into K-12 core curriculum.Combining teaching expertise with research expertise through co-design allowed for thinkingbeyond the context of a research study and into actual classrooms. The adoption of learning toolsand AI curriculum is influenced by complex factors outside the locus of control of the people creatingthe tools (i.e. designers and researchers) and the people using the tools (i.e. teachers). However,without teacher buy-in, adoption in the classroom would be impossible, and understanding theircontexts is necessary and often understated. Through this co-design, teachers experienced thepotential for AI to be embedded in subjects like social studies and English, which could allownon-CS and non-technical learners to experience AI in new classrooms. This method also leveragesstudents’ existing interests in non-technical subjects as a pathway into AI. This work serves as apush for further explorations to expose a wider range of students to AI.While the above findings could provide useful insights to AI education researchers and designers,we acknowledge the limitations of our study. Because our explorations focused on integratingAI with non-technical subjects with a small group of teachers, applying and extending generalimplications beyond this context should be done with caution. ngaging Teachers to Co-Design Integrated AI Curriculum for K-12 Classrooms 11
In this paper, we explored the needs of teachers in K-12 classrooms and how AI education can beintegrated with existing core curriculum. We engaged K-12 teachers and researchers in a two-dayco-design workshop, where we co-created lesson plans that embedded AI concepts into curriculafor social studies, ESL, and literacy for students with learning disabilities. We found that teachersvalue curriculum that address evaluation and engagement of students, which could be built intothe learning tool or curriculum. Teachers also successfully connected AI with their subject byhaving students examine subject-related datasets, as well as reflect on real-world implications andAI ethics. Our work highlights an opportunity to increase accessibility of K-12 AI education byembedding AI into core subjects (e.g., English, social studies), and reaching students outside of CSand technology classrooms.
ACKNOWLEDGMENTS
We thank the teachers who were a part of this study; Randi Williams, who provided co-designguidance; and Hal Abelson, who made this work possible.
REFERENCES [1] [n.d.]. A to Z of AI with Google. https://atozofai.withgoogle.com/intl/en-US/about/. Accessed: 2020-08-13.[2] 2011. Miro. https://miro.com/. Accessed: 2020-08-13.[3] 2011. Zoom. https://zoom.us/. Accessed: 2020-08-13.[4] 2013. Slack. https://slack.com/. Accessed: 2020-08-13.[5] Safinah Ali, Blakeley H Payne, Randi Williams, Hae Won Park, and Cynthia Breazeal. [n.d.]. Constructionism, Ethics,and Creativity: Developing Primary and Middle School Artificial Intelligence Education.[6] Anonymous Authors. 2020. Appendix. http://anonymized.for.peer.review. Accessed: 2020-09-16.[7] Richard Berk. 2017. An impact assessment of machine learning risk forecasts on parole board decisions and recidivism.
Journal of Experimental Criminology
13, 2 (2017), 193–216.[8] Harald Burgsteiner, Martin Kandlhofer, and Gerald Steinbauer. 2016. IRobot: Teaching the basics of artificial intelligencein high schools. In
Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence . 4126–4127.[9] Michelle Carney, Barron Webster, Irene Alvarado, Kyle Phillips, Noura Howell, Jordan Griffith, Jonas Jongejan, AmitPitaru, and Alexander Chen. 2020. Teachable Machine: Approachable Web-Based Tool for Exploring Machine LearningClassification. In
Extended Abstracts of the 2020 CHI Conference on Human Factors in Computing Systems . 1–8.[10] H. Chen. 2009. AI, E-government, and Politics 2.0.
IEEE Intelligent Systems
24, 5 (2009), 64–86.[11] Andrea L Dixon, Catherine Tucker, and Mary Ann Clark. 2010. Integrating social justice advocacy with nationalstandards of practice: Implications for school counselor education.
Counselor Education and Supervision
50, 2 (2010),103–115.[12] Alison G Dover. 2013. Getting âĂIJup to codeâĂİ: Preparing for and confronting challenges when teaching for socialjustice in standards-based classrooms.
Action in Teacher Education
35, 2 (2013), 89–102.[13] Eva Durall, Merja Bauters, Iida Hietala, Teemu Leinonen, and Evangelos Kapros. 2020. Co-creation and co-design intechnology-enhanced learning: Innovating science learning outside the classroom.
Interaction Design and Architecture(s)
The High School Journal
AI Magazine
38, 2 (2017), 3–4.[17] Clint Heinze, Janet Haase, and Helen Higgins. 2010. An action research report from a multiyear approach to teachingartificial intelligence at the K-6 level. In .[18] Aaron Hertzmann. 2020. Visual Indeterminacy in GAN Art.
Leonardo
53, 4 (2020), 424–428.[19] Jonas Jongejan, Henry Rowley, Takashi Kawashima, Jongmin Kim, and Nick Fox-Gieg. 2017. Google Quick, Draw!https://quickdraw.withgoogle.com/. Accessed: 2020-09-16.[20] Dale Lane. 2020. Machine Learning for Kids. https://machinelearningforkids.co.uk/. Accessed: 2020-09-05.[21] Phoebe Lin, Jessica Van Brummelen, Galit Lukin, Randi Williams, and Cynthia Breazeal. 2020. Zhorai: Designinga Conversational Agent for Children to Explore Machine Learning Concepts.. In
AAAI . AAAI, New York, NY, USA,
Proceedings ofthe 2020 CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA) (CHI âĂŹ20) . Association forComputing Machinery, New York, NY, USA, 1âĂŞ16. https://doi.org/10.1145/3313831.3376727[23] MIT App Inventor. 2020. Artificial Intelligence with MIT App Inventor. https://appinventor.mit.edu/explore/ai-with-mit-app-inventor. Accessed: 2020-09-10.[24] Reiichiro Nakano. 2019. Arbitrary Style Transfer in the Browser. https://reiinakano.com/arbitrary-image-stylization-tfjs/. Accessed: 2020-09-16.[25] Taro Narahara and Yoshihiro Kobayashi. 2018. Personalizing homemade bots with plug & play AI for STEAM education.In
SIGGRAPH Asia 2018 Technical Briefs . 1–4.[26] NSTA. 2014. Access the Next Generation Science Standards by Topic. https://ngss.nsta.org/AccessStandardsByTopic.aspx. Accessed: 2020-09-14.[27] Jeremy Roschelle, William Penuel, and Nicole Shechtman. 2006. Co-design of innovations with teachers: Definition anddynamics. In
Proceedings of the 7th international conference on Learning sciences . International Society of the LearningSciences, 606–612.[28] Bawornsak Sakulkueakulsuk, Siyada Witoon, Potiwat Ngarmkajornwiwat, Pornpen Pataranutaporn, Werasak Surare-ungchai, Pat Pataranutaporn, and Pakpoom Subsoontorn. 2018. Kids making AI: Integrating Machine Learning,Gamification, and Social Context in STEM Education. In . IEEE, 1005–1010.[29] Alexander Scheidt and Tim Pulver. 2019. Any-Cubes: A Children’s Toy for Learning AI: Enhanced Play with DeepLearning and MQTT. In
Proceedings of Mensch und Computer 2019 . 893–895.[30] Samuel Severance, William R Penuel, Tamara Sumner, and Heather Leary. 2016. Organizing for teacher agency incurricular co-design.
Journal of the Learning Sciences
25, 4 (2016), 531–564.[31] Ahuva Sperling and Dorit Lickerman. 2012. Integrating AI and machine learning in software engineering coursefor high school students. In
Proceedings of the 17th ACM annual conference on Innovation and technology in computerscience education . 244–249.[32] Jonathan Stray. 2019. Making artificial intelligence work for investigative journalism.
Digital Journalism
7, 8 (2019),1076–1097.[33] Danny Tang, Yuria Utsumi, and Natalie Lao. 2019. PIC: A Personal Image Classification Webtool for High SchoolStudents. In
Proceedings of the 2019 IJCAI EduAI Workshop . IJCAI.[34] David Touretzky, Christina Gardner-McCune, Cynthia Breazeal, Fred Martin, and Deborah Seehorn. 2019. A year inK-12 AI education.
AI Magazine
40, 4 (2019), 88–90.[35] David Touretzky, Christina Gardner-McCune, Fred Martin, and Deborah Seehorn. 2019. Envisioning AI for K-12: Whatshould every child know about AI?. In
Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 33. 9795–9799.[36] David Touretzky, Christina Gardner-McCune, Fred Martin, and Deborah Seehorn. 2019. Envisioning AI for K-12: Whatshould every child know about AI?. In
Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 33. AAAI,Honolulu, Hawaii, USA, 9795–9799.[37] David S Touretzky and Christina Gardner-McCune. 2018. Calypso for Cozmo: Robotic AI for everyone. In
Proceedingsof the 49th ACM Technical Symposium on Computer Science Education . ACM, Baltimore, MD, USA, 1110–1110.[38] Marie E Vachovsky, Grace Wu, Sorathan Chaturapruek, Olga Russakovsky, Richard Sommer, and Li Fei-Fei. 2016.Toward more gender diversity in CS through an artificial intelligence summer program for high school girls. In
Proceedings of the 47th ACM Technical Symposium on Computing Science Education . 303–308.[39] A. Vazhayil, R. Shetty, R. R. Bhavani, and N. Akshay. 2019. Focusing on Teacher Education to Introduce AI in Schools:Perspectives and Illustrative Findings. In .71–77.[40] Jed R. Wood and Larry E. Wood. 2008. Card Sorting: Current Practices and Beyond. 4, 1 (Nov. 2008), 1âĂŞ6.[41] Xiaozhe Yang. 2019. Accelerated Move for AI Education in China.
ECNU Review of Education
2, 3 (2019), 347–352.[42] Michelle Zimmerman. 2018. Teaching AI: Exploring New Frontiers for Learning.
International Society for Technologyin Education (12 2018).[43] Abigail Zimmermann-Niefield, R Benjamin Shapiro, and Shaun Kane. 2019. Sports and machine learning: How youngpeople can use data from their own bodies to learn about machine learning.