Welcome back! In this series, we take a closer look at how education is shifting as a result of generative AI tools like ChatGPT. Our previous articles touched on how instructors might co-design curricula alongside AI and how researchers consider supercharging inquiry with AI partners. Today, we explore the role that new AI might play in the creative process undergirding both of these areas.

On the surface, it’s all about tools. A new wave of AI apps is enabling creative minds to co-build 360 images, hybridized poetry, stylized videos and even 3D objects on their own. Writers and artists are teaming up with AI engines like ChatGPT to jump-start their design process and evaluate their paintings. In some cases, AI tools are being integrated throughout the entire creative cycle.

But what does this mean for students and educators? Where does the creative mind and learning self fit in an increasingly AI-driven world? More specifically, what do humans bring to the party, creativity-wise? Three aspects of the creative process that might be worth a closer look include: how we question, how we elaborate, and how we nudge.

How we question

Inside the creative process, questions can serve to pivot and to propose, to frame and reframe. Traditionally, this has been a human-to-human affair. But when a thought partner is not available, AI programs can also help to pose questions that inspire more human questions, facilitating loops of iterative inquiry. When the creative process becomes too knotty, AI has also shown the capacity to narrow down ideas and surface hidden insights in complex data.

With this in mind, what types of questions might learners be most responsible for? Ambiguous questions require one to wrangle with uncertainty, which can be problematic for computers. Questions that make intuitive leaps are difficult to explain, much less to program. Lastly, questions that reframe a problem require a feel for an audience that is often tacit and inexplicable.

AI can also bolster creative inquiry by investigating new ways of answering old questions. For example, AI can search literature and non-fiction for abstract ideas like “Why does love change over time?” or “How does knowledge elevate the soul?” While Google search is traditionally based on keywords, new search capabilities can link idea to idea, unlocking new connections and patterns of discovery.

In case learners don’t have access to the right person to ask a question, customized chatbots will increasingly allow users to interview and consult a virtual facsimile of their favorite artist or expert. These so-called “digital twins” are AI systems trained on one person’s data for an extended amount of time. As a result, a basic chatbot grows to mimic one particular person. This opens up the possibility to bounce questions like “How do you get unstuck when you’re in a creative rut?” to renowned creatives like music producer Rick Rubin.

To be clear, this is not exactly the same as talking to Rick Rubin. It’s important to distinguish between people and simulations of people. But it can be more than good enough as a tool to help us spark new thoughts.

How we elaborate

As ideas grow, they inevitably branch out and expand. Are there elements of this part of the creative process that are best suited for human input versus AI input?

We might start with a basic approach to using ChatGPT. A new user of ChatGPT could first think of it as an answer machine: I ask a question, ChatGPT gives me an answer. This might feel like ChatGPT as a research assistant or a boosted search engine.

However, beyond answering questions, generative AI tools can also serve as an experiment engine for elaboration. For example, a user could ask ChatGPT to build on an existing song:

To test out another path, a user might ask ChatGPT to rewrite it with a different tone:

For an alternative approach, you might ask ChatGPT to rewrite it in the style of a haiku:

What we’re doing here is not trying to get ChatGPT to give us the “right answer,” but rather to use ChatGPT to explore a divergent set of possibilities to expand our thinking. A designer might call this exploring sacrificial concepts that push our thinking.

Beyond chatbots, flexible and modular AI-boosted idea boards are also leveling up the iteration process. By clicking and dragging, the user can program AI models to generate images or widgets for idea experimentation. Concepts can be automatically compressed or expanded — and even recombined — with no coding at all. This flexible approach allows the learner to program a digital canvas to serve their particular needs and preferences, and does this in a way that feels like familiar whiteboarding tools like Google Jamboard or MURAL.

Additionally, while learners are good at producing ideas, elaborating on them over time can be challenging. AI tools are beginning to support this part of the creative process by setting up pathways to iterate concepts. For example, to organize ideas for iteration, some programs promise to operationalize your messy stack of notes or to extract and regroup ideas from across your computer. These apps largely integrate AI into existing workflows in order to reduce friction and rapidly find a fit with learning technology.

How we nudge

The right nudge can make or break creative work. Knowing where and when to nudge can be challenging, especially to an artist or learner deep within their creative journey. So, what role might creative learners be best suited for here? What about educators? What about AI?

Consider how we discover new connections. There may be a more human quality in finding unusual connections, especially ones that require intuition or cultural background. For example, could an AI have guessed that peanut butter and pickle sandwiches can be quite good? Or what about intuiting that an old Celtic song could find new life as a punk rock melody?

While AI is less capable at inuiting the right connection, it can nonetheless spur new directions. AI apps can now cook up wild combinations of ideas in user-friendly interfaces — and if you don’t like the concoction, there is no cost to try again until you find the right fit. Designers are packaging this creative partnership so that learners can outline a new direction alongside an AI, then evaluate the output on their own and decide if this propels their creative vision.

For more seasoned creatives, AI can also act like a visual design intern — just give it a word or phrase, and a color palette magically appears. For those students in digital media classes already focussing on industry tools, AI partners are now even integrated into Adobe products to nudge art in new directions.

That’s all for now! Stay tuned for next time as we take a look behind-the-scenes with education-minded folks considering how to develop their own AI tools.

This article was co-written by Josh Weiss, director of digital learning solutions at Stanford Graduate School of Education, and Glenn Fajardo, instructor at Stanford d.school.

Welcome back! This article is part 2 of the series A New Class of AI Tools, which explores the current wave of generative AI tools impacting teaching and learning. Previously, we delved into co-designing curricula and instruction alongside ChatGPT.

Lately, AI enthusiasts are focusing their attention on another key element of education: research. Often, these coders are researchers themselves who developed AI partners in order to overcome bottlenecks in their own process, such as collecting, interpreting, and compiling dense research material. This new batch of AI-aided research tools takes advantage of the co-thinking capacity of AI to bolster productivity, amplify insights, and lower the barrier of entry for curious novices.

Anti-confusion machine

Understanding dense material can be daunting and frustrating, especially for non-experts. One tool, Explainpaper, aims to solve this by narrowing the comprehension gap between reader and research. Within a simple web app, readers can upload a paper, then highlight sections that they find confusing or difficult. Immediately, a clarifying explanation will pop up. In effect, any research paper is accompanied by a rich interactive glossary that decrypts esoteric terminology and language patterns. Plus, if a reader still feels confused, they can pose clarifying questions to a built-in chatbot, which in turn searches and compiles answers from across the text.

Co-founder Aman Jha thinks of Explainpaper as a way to dialogue with a PDF. “Often, you’ll highlight something and the explanation will still be too dense,” he explains. “The follow-up questions are meant to surgically pinpoint whatever little piece you want. ‘What did you mean by that?’ or ‘What about this?’ Eventually, you can break down all the layers of understanding through questions.”

You get a research assistant, and you get a research assistant, and you get a research assistant…

Elsewhere, researchers are teaming up with AI to identify and compile work from journals with products like Elicit. Users can ask a research question and instantly get a list of papers that have tackled the subject matter. The AI can even generate a summary of findings in each paper and organize it into a table. For deeper investigation, Elicit also scrapes the text of each study for key attributes like duration and study type; in parallel, it can identify measurement tools and offer preliminary suggestions on limitations of the study design. As a result, Elicit accelerates the research workflow by bypassing initial steps like preliminary search as well as by surfacing fertile areas for investigation.

So, if everyone now has a research assistant, what does this mean for equity? At the very least, tools like Explainpaper and Elicit lower the barrier of entry to newcomers. By demystifying domain-specific topics and definitions in context, esoteric knowledge will be less of an obstacle to curious minds and new voices. To Aman Jha, tools like Explainpaper have the potential to expand access to complex topics and knowledge that are often locked away in inscrutable texts. “I want more people to go into science or whatever field that they want to go into when they want to go into it,” notes Aman, “and I don’t want them to quit just because it feels like it wasn’t made for them or that they don’t understand research papers.”

“I want more people to go into science or whatever field that they want to go into when they want to go into it”

Ultimately, both experts and newcomers will mix and match these tools as AI is injected into more segments of the research process. And this core suite of research tools is expanding as researchers identify bottlenecks that other AI tools can resolve. In one instance, users are already leveraging ChatGPT to transfer knowledge and terminology across domains and languages, and even acting as “translators” for young learners (see below).

The truth of the matter

How these tools will play out in the wild is still to be seen. Galactica, an AI-powered search engine for research papers, saw pushback after its veracity was called into doubt, and it has since been pulled from public use. The risk of erroneous information is a top concern, and is exacerbated by the confident and authoritative tone that users often see with AI interfaces like ChatGPT.

As a result, transparency has become a top “feature request” for the next generation of tools. At the moment, ChatGPT can fabricate references without providing any ability to “peek under the hood” — but the team at OpenAI is actively working on it. In parallel, Antropic AI is testing a ChatGPT lookalike called Claude which has a built-in mechanism called Constitutional AI to curate responses for traits like reliability. Over at Google, DeepMind is building Sparrow, a dialogue agent that aims to “reduce the risk of unsafe and inappropriate answers” by providing direct evidence for claims. Perhaps most aligned with research formatting is Perplexity.ai, which boasts in-line references to help users trace sources across a multi-sentence response.

Alternatively, some projects are addressing accuracy by fine-tuning AIs for domain-specific information. Models like GPT-3 are being trained on external knowledge bases and particular types of documents, such as academic texts, in order to better hone the context and relationships that show up in responses. The Stanford Center for Research on Foundation Models recently developed PubMedGPT 2.7B, a domain-specific language model trained on a large corpus of biomedical abstracts and papers. While the model is quite accurate for answering questions about biomedicine, it may still fabricate content.

Veracity will need to find firmer footing not only within the research community, but also in industry. Some groups are even filing lawsuits and pushing for greater oversight on how chatbot-generated ideas are to be attributed and traced — which further highlights a need for responsible development and usage, as well as accountability for when things go wrong. Groups like Human-Centered AI at Stanford focus on these questions and connect research insights with industry practices.

In the meantime, professional research communities will continue to wrestle with the implications of AI in-the-mix. Ironically, AI journals are indicating reluctance to incorporate AI partners. In other places, scholarly research listing ChatGPT as a co-author has been accepted. And the pace of production will only accelerate as integrations with knowledge bases like Wolfram unlock AI-powered research on an unprecedented scale.

That’s all for now! Join us for future articles as we delve further into how new AI tools are impacting teaching and learning, storytelling, the creative process, and more.

This article was co-written by Josh Weiss and Miroslav Suzara, and published by The Office of Innovation and Technology at Stanford Graduate School of Education.

Up to now, AI has largely felt like a means to get things done. Google Maps finds the best route, Amazon recommends the right gift, and LinkedIn identifies new work friends. But recent AI advances have opened the door to a new dynamic of human-and-AI co-creation, ushering in a wave of digital tools under the banner of “generative AI.” This burgeoning AI toolset is spreading rapidly throughout education, helping learners to trawl through existential questions one moment and brainstorm counterarguments the next. It is even shaking up cornerstones of the learning process like how we study for tests or assess knowledge.

Though the landscape is taking form, the effect on education is uncertain. Areas like composition, storytelling, research, and creativity are already seeing an influx of tools that are re-orienting how humans develop ideas and solve problems throughout their day. Still, at a higher level, many larger questions remain: Who gets credit for ideas? Should educators encourage the use of AI in classrooms or on exams? What are the future assessment strategies if AI becomes the new word processor? In the era of AI proliferation, what are the future competencies schools should focus on? Will it drive greater inclusion or more disparities in learner outcomes?

Understanding the whole at once can be rather overwhelming. To garner a sense of what these new tools could mean, we’re publishing a series of micro-articles to offer snapshots of some of the corners of education where this new wave of AI is shaking things up.

Proposing and composing

OpenAI’s public release of ChatGPT in early December changed the landscape of co-learning overnight. ChatGPT is a personalized, rapid-response thinking partner that can not only offers ideas, but riff on them, too — and in almost any subject matter imaginable. For example, ChatGPT could be used to provide personalized explanations and feedback to students on homework assignments, practice problems, or exams. Students could also ask directly for support to solve a problem rather than spend time searching for responses on online forums. More extensively, the platform can copilot as a seemingly infinite curiosity machine and potentially supplant Google as a personal intelligence companion.

ChatGPT can also co-produce educational content such as lessons, study guides, and quizzes. For example, a teacher could provide ChatGPT with learning objectives or outcomes, and ChatGPT is capable of suggesting a rudimentary learning plan. This could save time and effort for teachers, and could also provide students with more personalized and tailored learning materials aligned with their needs (see below).

While impressive, there are notable gaps that educators and learners should take into consideration. The quality of ChatGPT’s expertise can vary by domain, and the quality of instruction is not always consistent. To solve this, the field of prompt engineering — or using the right combination of words — is emerging as a skill for learners in order to generate the most useful response from the chatbot. Communities are also pooling their experience to identify optimal lines of questioning for learning. For example, coders on Twitter have explored question patterns through which ChatGPT could potentially augment the debugging process for programmers, discover and narrate solutions as an alternative to Google, and even teach concepts through tutor-like dialogue (see below).

Not least among possibilities — and one that has been getting a lot of buzz — is essay-writing. To combat writer’s block, students can copilot with an AI to generate new ideas or invigorate their writing with more varied language structures. In one case, Mina Lee from Stanford University developed CoAuthor, a human-AI collaborative writing dataset based on GPT-3 (the basis of ChatGPT) to assist creative and argumentative writing. The dataset hooks into a collaborative interface where the writer is presented with several AI-generated suggestions that the writer can choose and modify. Such tools hold promise to augment human creativity by supercharging the ideation process as well as helping to pivot ideas in novel directions. However, this new capability raises important questions around plagiarism and assessment — which, fittingly, AI-driven anti-plagiarism startups are looking into. This all underscores the importance of critical thinking, as students will be evaluating and editing ideas as much as producing them.

That’s all for now! Stay tuned for future micro-articles on storytelling, research tools, creative thinking, and more.

This article was co-written by Josh Weiss and Miroslav Suzara, and published by The Office of Innovation and Technology at Stanford Graduate School of Education.

The first computer mouse was nothing more than a block of wood with a single red button. Doug Englebart developed the revolutionary device at SRI International in 1968 to make navigating a computer screen more intuitive — a big leap from the inscrutable adding machines of the day. In the process, Englebart’s laser-focus on intuitive design birthed a new field, human-computer interaction (HCI). In its highest form, HCI produces a frictionless relationship between people and machines, the kind of symbiotic flow state that sci-fi writers and tech visionaries have hinted at for over half a century. Naturally, many of these early technologies were applied to learning and collaboration, and education technologies common in classrooms today are in many ways products of Englebart’s vision.

Dr. Hari Subramonyan builds on this tradition. In his role as Assistant Professor at the GSE and Computer Science (by courtesy) and a Faculty Fellow at Stanford’s Institute for Human-Centered AI (HAI), Dr. Subramonyam considers how intuitive digital design makes learning happen. His HCI work focuses on augmenting critical human tasks with AI by incorporating principles from cognitive psychology. More specifically, he aims to balance automation with other cognitive goals such as learning, creativity, and sense-making. “The focus of intelligent systems,” posits Dr. Subramonyam, “is on ‘making things easy’ through automation. However, there is such a thing as too easy or too automated.”

Inviting exploration

Dr. Subramonyam conveys these HCI principles throughout his course, EDUC 432: Designing Explorable Explanations for Learning. The objective is to teach students how to design digital learning experiences that feel more attuned to how humans think and feel, such as using text as an environment to think in. Throughout the course, students apply concepts from visualization theory and instructional design. Guest experts also share their insights into design, data visualization, and representations of research. By the end of the quarter, students create their own explorable explanations.

With HCI design principles in mind, students’ prototypes address an array of challenges in education. “Into the Chaparral: A Fire Experience,” an author-driven tool developed by GSE student Vicky Z. Chan, introduces the learner to the California chaparral ecosystem and how its plants are adapted to wildfires. “I really appreciated learning about instructional design and information visualization, discussing examples of explorables, and getting to create my own explorable,” says Vicky. “I can see these concepts informing the digital science communication work I want to do!”

In another project, “Holding onto our Best Defenders of Student Learning,” GSE student Helen Higgins prompts the learner to deeply engage with issues affecting teacher retention as well as the impact on students. “As someone without a technical background, this class helped me break down core aspects of interactive learning,” explains Helen. “Since this class, I’ve returned to some of these tools, like metaphors, focal points, and hypothesis-building when designing in-person and digital learning experiences.”

Augmenting learning alongside AI

In addition to the course, Dr. Subramonyam’s research has explored a wide range of learner-centered interfaces, including with AI. “Today, there are many things that AI can do,” he notes. “However, shaping the right learning experience requires a multidisciplinary perspective to set expectations about what AI can do — to predict when it might fail, to align human values with artificial intelligence, and, most importantly, maintain human agency in learning.”

Towards this effort, Dr. Subramonyam has developed prototypes to intelligently support learning activities such as annotation and visual sense-making. With texSketch, Dr. Subramonyam and his collaborators created a prototype that makes it easier to produce diagrams while engaging in active reading strategies through the use of AI and natural language processing. Another application, VideoSticker, supports visual note-taking on video content through a process that enables object detection and links with the transcript.

As part of his research, Dr. Subramonyam has introduced three strategies for combining AI with human effort. The first strategy, automation after human-effort, involves a learner receiving automation support after they have demonstrated learning. This is seen in texSketch, in particular, where important relationships from a reading are automatically visualized only when it senses that the user has learned key concepts.

A second strategy, automation as a complementary to human-effort, also involves automation support as a strategy for human-AI interaction. However, complementarity is emphasized in a way that ensures the human is still in charge of the process. This is evidenced in TakeToons, a tool that helps to design animations by automatically aligning spoken dialogue to script and enabling flexible edits through speech-based commands.

The final strategy, automation as last-mile optimization, considers automation support to help humans make sense of large and complex data sources. AffinityLens, a computer vision-based tool, embodies this well; the application reduces the effort of diagramming by providing overlaid data insights on top of sticky notes and extending human-made clusters with additional recommendations.

Building out and learning across

As more learning moments occur on digital interfaces, HCI will take an increasingly significant role in the lives of educators and students. Traditionally, HCI integrates expertise in cognitive psychology, software engineering, and design. But weaving students and educators into design and development, as in EDUC 432, will be important going forward for inclusive design that centers as much on how the learner learns as what the technology can do. “The co-design processes, design representations, principles, and tools that I am developing in my research enable teachers, students, engineers, ethicists — and, of course, designers — to collaborate across expertise boundaries,” adds Dr. Subramonyam.

Moreover, building education technologies now means acknowledging how humans think, feel, and move around a screen. The key is connecting minds, technologies, and classrooms — and spreading those intentional practices to everyone building and designing technology. HCI offers a toolkit for these connections and a way to calibrate the digital boundaries of today’s learners. “Ultimately,” says Dr. Subramonyam, “by centering people in technology design, HCI can drive the creation of ethical, inclusive, and usable AI experiences.

This article was co-written by Josh Weiss and Miroslav Suzara, and published by The Office of Innovation and Technology at Stanford Graduate School of Education. For more information, contact josh.weiss@stanford.edu.

For Dr. Francis A. Pearman, authenticity is core to any learning experience. “At the heart of my teaching is a desire to foster deep learning,” he states, “the type of learning that transforms a student’s engagement with the world.” In turn, his own engagement with the academic world revolves around authentic issues: as Assistant Professor of Education at Stanford Graduate School of Education, his research and policy interests focus on how poverty and inequality shape the life chances of children and how policy levers can improve outcomes for low-income children. He works with authentic data — quantitative puzzles inside of complex governance systems — and knows what it takes to hash out policy that makes sense.

The goal of Dr. Pearman’s course, Education Policy in the United States, is to cultivate these hard-won, real world policy-making skills in up-and-coming leaders and researchers. The course draws a diverse group of students, as it is cross-listed with the Graduate School of Business and welcomes undergraduates, Master’s students, and PhD students to join. With the post-lockdown policy landscape wide open, Dr. Pearman and his teaching assistant, Carrie Townley-Flores, saw an opportunity for authentic learning. “We wanted to show students with limited experience as practitioners and policy-makers the benefits of good education policymaking, but also how messy policy design and implementation can be,” reflects Carrie. “This was our chance.”

Setting a course

Dr. Pearman and Carrie set out to design their Spring 2022 course around a classic technique for authentic learning: case studies. Case studies traditionally focus on a challenge that an institution, organization, business, or policy group have wrangled with. The case study document that students typically read entails data, interviews, and outcomes in a tidy narrative. However, Dr. Pearman and Carrie envisioned something that felt closer to what leaders would actually encounter in a public school district — scattered data, unfiltered interviews, and messy outcomes. Their vision centered on experiential learning that the cohort of budding leaders could work through, challenges and all, alongside peers and the teaching team.

Building a series of interactive case studies in a few short months would require careful planning. Dr. Pearman connected with Josh Weiss, Director of Digital Learning Solutions, to brainstorm. Drawing inspiration from previous simulations, the group settled on developing a series of web-based exploratory maps where students could investigate artifacts from each dilemma. The learning design would center around three guiding principles: 1) How can a case study support and enhance authenticity? 2) How can a case study facilitate learning in new and memorable ways? and 3) How can a case study be structured to support the diverse interests and aspirations of students?

Accordingly, the team selected school choice policies in Detroit as a first interactive map to build out, as it felt especially relevant to the cohort’s diverse interests. Carrie began to curate artifacts to place around the case study map — podcast clips, diagrams, town hall meetings, research papers, etc. At the same time, software engineer Jonathan Lai repurposed code he had developed for previous courses and built additional visualizations and integrations on top. Video and audio players were also embedded within the map itself to afford more immersion and explorability.

Connecting with class

With a prototype ready, the team turned their attention back to classroom experience. Dr. Pearman and Carrie began the course with students reading, writing, and debating around key aspects of education policymaking. Students eventually distilled insights into four key “ingredients” that add up to high-quality educational policymaking. These ingredients carried forward into a guiding framework that students used for the remainder of the quarter to anchor case study engagement and discussion.

With the framework settled, Dr. Pearman and Carrie next designed a brief orientation during class and within Canvas to set up norms and expectations for a self-directed excursion around a school district. They also set up a written assignment in Canvas to transition ideas from the map back into class discussion later in the week. To gauge points of friction, Josh developed a brief survey for students to answer, which Carrie placed within the course itself.

Meanwhile, the team built out four additional case studies. New York City’s gifted education program, Oakland school closures, California’s universal preschool initiative, and teacher recruitment in several districts across the United States fleshed out a broad-ranging set of real-world cases. As Dr. Pearman and Carrie sourced more digital assets to plug into the interactive maps, the Digital Learning Solutions team — including Josh, Jonathan, digital media producer Joe Sherman, and academic technology specialist Mae Bethel — further advised on learning design and found creative ways to arrange and embed a wide range of audio, video, and text-based artifacts into a digital landscape.

One-by-one, the case studies were rolled out during weeks 3–7 of the course. Dr. Pearman and Carrie, along with teaching assistant Jessica Boyle, unpacked each case study with students in class and modeled their own analytical process when approaching messy data. Student feedback indicated that this unpacking was a crucial step; the variety and depth of material within each map could be intimidating, and having a collective post-analysis session with peers and experts helped to stitch together their own conclusions from a complex policy terrain.

In addition, in-class activities allowed students to build off of one another’s experiences with the case studies. For instance, after analyzing the Detroit school choice case map, students were grouped and assigned policy-elite roles from which to debate and develop an improved school choice policy for Detroit. These in-class activities were designed to push their thinking and to make decisions rather than just theorizing about the various policy cases; students had only 45 minutes to synthesize their thoughts, collaborate on a solution, and produce something consequential.

In the making

After analyzing case studies from around the country, it was the students’ turn to generate a case study rooted in their own interests. Within a digital expo hall on Zoom, each student mindfully laid out a case and presentation with scrupulously selected articles and readings. Crucially, students were encouraged to bring their own experiences to their final projects, exposing participants to different perspectives and infusing a wide range of sources into their own debates and cases. The discussion groups responded in kind by taking time to read through the materials and asking thoughtful questions, which fostered extended conversations and analyses amongst the students.

The expo nicely encapsulated the authenticity that the teaching team was aiming for. “We thought this new design would help students from a wide range of backgrounds approach education policy with a critical analytic lens that is grounded in both research and the lived experiences of folks in the communities affected by a given policy,” notes Carrie. With the expo wrapping up the quarter, students’ case studies left a lasting impression on the teaching staff. The abundance of case study material and extensibility of digital maps offers potential for iteration. And there is rich possibility to draw on cases that students may have grappled with deeply and personally in their own neighborhoods or through their own stages of life. “We took notable risks with this course in terms of rethinking how education policy could be taught and engaged with,” notes Dr. Pearman. “Considering how students responded, those risks were well worth it.”

Blockchain lingo gives the impression of an adrenaline-fueled world full of eccentric financiers. “Miners” and “whales” pride themselves in “staking” their investments and declaring that they will forever “HODL” (hold on for dear life) to their “altcoin.” But beyond the headlines, a less frenzied subset of blockchain projects is emerging with long-term and socially-minded goals around governance, public services, and education.

So how does it work? Blockchain is an unalterable, fully transparent public record that tracks the transfer and ownership of digital things. Anyone can exchange (ie, buy, sell, transfer, barter, auction) internet-based assets (eg, cryptocurrency, an image, a contract signature). Crucially, the system is designed for trust: transaction data is freely accessible to all, and anything recorded to the blockchain is “chained” to “blocks” of existing data to make it tamper-proof. This system ultimately serves to regulate digital economies by keeping track of the data that is exchanged online.

Building skills, paying bills

Recently, this new mode of ownership and exchange has inspired projects that look to address an age-old challenge in education: motivation. One subgroup of blockchain projects called “learn-to-earn” (also sometimes called “earn-to-learn”) aims to incentivize learners to engage with educational content and, just as importantly, stick with it. Initial experiments by startups incentivized users to watch tutorials and take a quiz in exchange for earning cryptocurrency. CoinMarketCap’s “learn crypto, earn crypto” program has refined their program one step further so that once a learner demonstrates knowledge mastery, rewards are directly released to the learner via a digital wallet.

However, emerging programs in the K12 space focus more on the unique needs of hard-to-reach students. Learning Coin, a project headed by the World Bank, incentivizes students in rural communities to stay in school and improve academic performance. The program evaluates completion and consistency of student work, then releases digital funds accordingly. While some conventional conditional cash transfer programs are vulnerable to corruption and fail to scale due to inefficiencies, blockchain supports the World Bank’s program by ensuring transactions are recorded publicly on the blockchain ledger. As a governance tool, these automated transfers also reduce administrative overhead and record-keeping, which can be challenging for education programs in remote locales.

Another platform, Mygrants, allows learners to access skills training and build new competencies while developing credit through digital cash transfers performed at a low cost by blockchain technologies. The training content is broken up into short, personalized learning “pills” based on personal goals. As students answer questions, they collect points and receive formative feedback to develop critical thinking skills. Learners benchmark their progress against peers with similar goals, and receive badges, points, and a digital payout at the end of the month if they reach their goals.

Towards lifelong learning, the Learning Economy Foundation (LEF) aims to create a decentralized, blockchain-based network where skills and credentials are stored within a digital identity that follows the learner. Recently, LEF partnered with LEGO Foundation to create a gamified learning experience, SuperSkills!, where elementary school students can select adventures and collect gifts as a result of learning core skills. Underneath the hood, the app uses the W3C’s Universal Wallet, a framework developed by MIT and LEF to store credentials within a blockchain-based identity. This identity is not locked down to one app or company, allowing learners to own their data and use it as they wish across their academic and professional lifetimes.

Ramping up is hard to do

As with any emerging technology, equity must be at the core. Early research indicates that blockchain adoption skews towards students with technical backgrounds and entrepreneurial mindsets. However, there is encouraging data around access and utility for under-priveleged communities. “Play-to-earn” projects with well-designed user interfaces such as Axie Infinity have seen significant adoption among low-income groups, and currently supplement household incomes in the Philippines. Burgeoning projects with national governments may broaden opportunities for student credentials in Ethiopia, skill validation in the country of Georgia, and more distributed and inclusive communities via decentralized autonomous organizations (DAOs).

At the same time, these new learning pathways will likely face technical drawbacks. Accessibility with older systems and devices, like those commonly used in developing economies, will be problematic (although browser-based applications may offer a short-term solution). While blockchain’s interconnected and open nature is key to data ownership and exchange, individuals must be vigilant with data security to prevent hacking incidents.

Finally, as learn-to-earn projects and digital wallets mature, learner-centered design will become more crucial. As any teacher or parent knows, extrinsic rewards will only go so far; balancing extrinsic motivation with intrinsic motivation is crucial throughout a learning trajectory. And while extrinsic motivation may get students in the door, teaching strategies like sense-making and project-based curricula have been shown to keep students authentically engaged in a task. A new community of technologists and educators will need to rise to the challenge to design a layered and adaptive system of rewards and strategies — a concept referred to by blockchain enthusiasts as “tokenomics.” To find success with learners, blockchain projects that reach into the classroom will be looking more to educators to co-architect incentives and journeys that meet the student where they are at personally, academically, and financially.

This article was co-written by Josh Weiss and Zoha Salman, and published by The Office of Innovation and Technology at Stanford Graduate School of Education. For more information, contact josh.weiss@stanford.edu

Gaining knowledge often requires us to notice moments deeply. This is especially true during training: a medical student must focus on instantaneous shifts in symptoms, a developing musician must track pitch in microseconds, and a budding teacher must follow, moment-to-moment, how students make sense of new ideas. Typically, we are trained in the art of noticing via one-on-one interactions when a peer, parent, or instructor points out where something important is happening and how to make sense of it. This process, however, is time- and resource-intensive, requiring as much energy from the teacher as the learner.

Dr. Victor Lee, Associate Professor in the Graduate School of Education, is keenly aware of this challenge — namely, that teaching others how to develop good habits in noticing is hard to do and hard to scale. To address this, Dr. Lee has been developing new ways for students to help each other notice and understand important moments through a first-of-its-kind course in the Stanford Teacher Education Program (STEP). The aim of the course is to empower up-and-coming teachers to incorporate important ideas about data, and from data science, into any subject matter. This involves training new teachers how to notice moments of student input relevant to quantitative data, and then utilize these moments to support and amplify students’ ideas. “In education and cognitive research, we talk a lot about how effective instruction really builds on what students already know related to the topic being taught,” notes Dr. Lee. “One important step in developing excellent teachers is getting them to see when students are mentioning good ideas and making decisions about how to build on those until a learning goal is met.”

Planning with intent

Inspired by the Transforming Learning Accelerator, a campus-wide initiative anchored at the GSE, Dr. Lee reconsidered the utility of video footage, originally collected for research, for instructional purposes. Dr. Lee approached Josh Weiss, Director of Digital Learning Solutions at GSE IT, to consider how the skill of noticing might be taught via this video footage. The two ideated around unique properties of digital video, including its capacity to crowdsource and network patterns of noticing across a class of students. Dr. Lee and Josh sketched out an initial paper prototype, applied for a grant, and went to work assembling a roadmap for the project.

Project goals were laid out. First, the technology would optimally leverage the unique many-to-many sharing properties of digital media. Second, the platform would be learner-centered and promote organic learning moments in the classroom. Third, project development needed to move quickly, so rapid prototyping and iterative development would be key.

Building and iterating

As development began, Josh pulled in Jonathan Lai, software engineer at GSE IT, to start researching web technologies that supported media annotation. One platform, Frame.io, stood out for its stability and rich data-sharing features. Commonly used by production studios, Frame.io allows multiple contributors to comment, upvote, and even doodle on a video frame-by-frame. Just as importantly, the platform also exports data in a way that can be remixed and analyzed on-the-fly.

Although Frame.io supported video-based group work well, the tool did not crowdsource noticings in the manner that Dr. Lee had envisioned. While noticings were collected efficiently, the off-the-shelf data visualizations in Frame.io were ineffectual; the data presentation was not conducive to the discussions Dr. Lee had hoped to foster among students around what and how they were collectively noticing.

To address this shortcoming, Jonathan and Josh developed a custom data visualization. The initial goal was to visualize all noticings at once while also conveying the depth and frequency of noticings in particular parts of the video. For version 1.0, a vertically stacked chart of all comments across a horizontal timeline gave a sense of the pacing of comments. All comments were exported and imported manually across the two platforms during class, and students were able to see an aggregate representation of their comments within only a few minutes of starting the activity.

However, the overlapping geometry of comments with long duration hindered the legibility of the graph. As a result, version 2.0 of the data visualization emphasized compactness; the chart became more compressed vertically and aggregated data at intervals to structure discrete sections of the video. Version 2.0 also integrated data via an API back-end that allowed professors to refresh the data visualizations in real-time.

Still learning

With the essential technology developed, the team performed user testing to map out the full learning experience. Alpha and beta testing was initially performed with colleagues to evaluate rudimentary learning pathways and delivery models, then expanded out to students and professors to fit their needs. Conceal-and-reveal methodologies were tested out, along with prediction, analysis, and reflection techniques. With some groups, specific prompts and deliberate pacing were employed; with others, more organic discussions evolved from open-ended prompts and looser time constraints.

Based on these deployments, Dr. Lee submitted a report on this effort for publication in the Proceedings of the 2022 Annual Meeting of the International Society of the Learning Sciences. “In some ways,” reports Dr. Lee, “crowdsourcing noticing is a natural extension to what we already are capable of doing. But we haven’t seen it done in this way, and sharing what GSE IT has helped develop with the broader academic community is a great way to inspire and spread this innovation.”

As Dr. Lee looks forward, he sees additional opportunities with noticing, data, and digital media. Paring down features to key functionalities for the classroom could streamline the experience for learners. Richer data visualizations could foster deeper and more revelatory discussion among participants about how they take note of classroom phenomena. Garnering more annotation data over time opens the possibility for comparing patterns across cohorts of learners. “A lot of the technological innovations we appreciate now — such as recommender systems — rely on large crowds sharing what they find to be interesting and important,” Dr. Lee notes. “There is a big opportunity here to leverage this crowdsourcing for education, and it is really exciting to explore what sort of crowdsourced noticing could be possible.”

Andre Nudelman is currently chairman of the Digital Media Academy, founded at Stanford in 1999 to provide technology education to K-12 teachers and develop technology immersion courses for students. Today it offers STEM online summer camps and courses in such topics as artificial intelligence, game design, and film-making to students in more than 125 countries. In addition to owning the company, Nudelman also supports thousands of full scholarships for talented students through the Nudelman Family Trust.

What are your goals as Entrepreneur in Residence (EiR)?

My goal at the Graduate School of Education is to contribute to its positioning as the center and the forefront of innovation in education and also to help improve education globally by sharing my experience with students and faculty members, and also learning from them.

Which topics interest you most?

The topic of the moment that interests me the most is to observe what’s going on in the schools because of the pandemic, where teachers have to overcome their fear of technology and start using this technology to deliver their curricula online to their students. Observing what happened, what is happening, and what’s going to happen after the pandemic will provide us educators with very valuable data that we can use to optimize the educational system for the future.

What challenges are you looking to address?

The challenge I’m looking to address is to use my previous experience in delivering curricula online and in digital format for K-12 schools to enhance the quality and effectiveness of this digital initiative. Together with another professor, I’m developing a course that will address the effects of technology in K-12 schools.

What unique viewpoints do you bring to the GSE as an entrepreneur?

I bring to the GSE a different perspective to the way the problems should be addressed, because it is a very practical way. This are based on my experience doing businesses. In the last two years I had 6 companies in education, and before that I had 32 companies in a series of different areas. That brings a different point of view that enriches discussion of the major topics.

What advice do you have for budding education entrepreneurs?

Some advice that I can offer to education entrepreneurs are: first, don’t be afraid of failure. Failure brings very valuable lessons that will enhance your probability of success in life. Second thing, be very diligent, work hard, treat people very well, and you have much better chances in business.

“Never felt more useless in my life,” tweeted DotACapitalist, a professional gamer, “but we’re having fun at least.” An artificially intelligent (AI) computer system had just thrashed the world’s top players in Dota 2, a video game that demands ingenuity and teamwork. Most striking, perhaps, was the AI system’s observable behavior, particularly pace and quality of decisions. “[It] just doesn’t need the processing time that humans require,” reported The Verge, “which made its play appear unnatural — but only in the speed and crispness of the decision-making, not in the content of those decisions.”

While AI may feel unnatural, its utility is undeniable. In the case of Dota 2, the computer system gained the experience of 120 years’ worth of playing in a single 24-hour period, all the while tweaking its strategies against more capable versions of itself. This capacity for self-improvement with an established data set and rules, known as machine learning, has existed for decades. However, recent improvements in training methods, in particular a branch called deep learning, have accelerated AI systems in areas like gameplay; chess, Go, and Dota 2 masters have all fallen at the hands of deep learning in recent years.

For many of these masters, defeat spurred a new perspective on how humans and AI systems interact. Following his defeat, Go master Lee Sudol marvelled at the computer’s computational breadth and focus: “I cannot accept a technical superiority, but in that [concentration/psychological] area, I believe it [AlphaGo] will be difficult to beat for humans.” Indeed, the unique combination of human-plus-AI teams are frequently more effective than either AI- or human-only teams, as evidenced in hybrid team formats like Advanced Chess.

The new role of AI operator

Chess grandmaster Garry Kasparov, whose defeat in 1997 by IBM’s Deep Blue spurred human-computer hybrids in chess, goes one step further. He sees human-computer interaction (HCI) and machine learning as soon-to-be defining traits of everyday tasks. Computational breadth and focus is only useful to a point, he says; without humans to anticipate, integrate, and communicate unforeseen factors, decisions fail. “Are we doomed to a future where things maybe work more effectively, but we’re frustrated all the time because there’s no human-to-human interaction to ease [comprehension]?” The answer: “There will be other humans who will be supervising machines.”

Human-computer teams that use machine learning are increasingly popping up in education. On a fundamental level, search engines have amplified learning through discoverability and optimization. For each student that has let Google autocomplete or refine via “Did you mean…?”, a better and more precise resource has been delivered. But now this partnership has received a boost through better interface design coupled with the iterative power of machine learning.

A notable example is Wekinator, a kid-friendly machine learning interface. Its creator, Dr. Rebecca Fiebrink, Senior Lecturer in Computing at Goldsmiths, University of London, envisioned a system that could enable non-experts like artists and musicians to train a computer in much the way you would train a puppy. When one behavior (say, a hand wave) is performed, you can train the machine learning system to perform a response (say, a musical note). Over time, you provide more examples and contexts. Just as a trained dog would accumulate norms and commands that guide its behavior, the machine learning system comes to exhibit a unique set of expressions.

Crucially, this all occurs without any coding. Dr. Fiebrink designed the interface so that “you’re making a program by providing examples rather than writing code,” opening the door for non-technical domain experts like composers to leverage rapid computation and iteration. “In a sense, people are spending time doing things they’re good at or interested in,” she notes. “In the best case, machine learning takes over for some of the things we can’t do as well.”

Looks like an app, feels like an assistant

Time-consuming tasks once performed by teachers are increasingly managed by intelligent systems. Take, for example, a peer review activity. Traditionally, instructors would tally up peer responses per student, evaluate the quality of feedback, and identify those students not pulling their own weight. Peerceptiv, an edtech company that employs machine learning techniques, now provides all of these services in a single, code-free interface. Moreover, the machine learning ties into rubrics and data visualizations to streamline the feedback flow.

Many edtech companies are taking the same approach, marrying machine learning with code-free interfaces. Cerego, a learning platform, maps out skill trajectories and recommends templates and materials. As the program observes the user’s habits over time, it can adjust to the pace and instructional needs of the learner. Separately, Pearson, a publisher, teamed up with IBM Watson, a machine learning engine, to devise a digital tutor. Each click or idea expressed by the learner is tracked and then integrated into a learning profile, producing customized resources and feedback akin to one-on-one instruction.

Learning alongside AI increasingly occurs on mobile devices. Quizlet Learn, a smartphone app, looks through a database of study patterns to determine an optimal study schedule for someone in your situation. Liulishuo, popular in China, will polish your English pronunciation by noting patterns phoneme-by-phoneme. And AdmitHub, a virtual admissions counselor, can walk prospective students through deadlines and paperwork in 100 different languages. Each of these AI-enabled experiences is designed mobile-first to match the needs of modern learners.

Out in the world

With interface design becoming more intuitive, intersections of non-technical fields with machine learning will continue to blossom. Musician Brian Eno’s app, Wavepaths, invokes principles of neuroscience to generate therapeutic soundscapes that morph over time to fit each patient’s needs — all with basic visual cues. Mozilla, a digital research organization, has opened up $225,000 in grants to “media makers” that can creatively communicate machine learning’s effect on society.

All this points to a tighter integration of interaction-minded machine learning with day-to-day activities, whether teaching in a classroom or playing with a toy. To address these potentially disruptive forces, courses are popping up at MIT and Stanford to guide ethical questions around intelligence augmentation and machine learning’s role in society. On the industry side, Palo Alto-based Institute for the Future has disseminated Ethics OS, a guide to building foresight into the product development cycle.

As machine learning finds its place in the pedagogical process, educators will become even more instrumental to student achievement. As Kasparov notes, machine learning systems need operators, namely “[s]omeone who can work out the most effective combination, bringing together human and machine skills.” These operators must have deep expertise in the applicable domain, in this case pedagogy.

In the end, adaptivity will be essential for all parties involved. New applications and spaces will emerge with the promise of efficacious learning. But perspective will remain paramount. For designers of machine learning systems, imagination is fundamental. “The first step,” notes Dr. Fiebrink, “is getting your head around ‘what do I do with it?’ and ‘what is the space of possibilities?’” Once you understand the system, the learning can really begin.

This article was written by Josh Weiss and published by The Office of Innovation and Technology at Stanford Graduate School of Education. For more information, contact josh.weiss@stanford.edu

Iterating and upgrading is crucial to good instruction. Professor Walter “Woody” Powell knows this well. His course, Organizational Behavior and Analysis, has steadily transformed over the years with a persistent aim towards relevance and applying theory to practice. “This course gets people to think about organizations as things to be designed and improved,” he notes. “It got tweaked a bit each time in response to students’ interests and current events.”

The scope of the course mirrors Prof. Powell’s expertise. He is the Jacks Family Professor of Education, and, by courtesy, Professor of Sociology, Organizational Behavior, Management Science and Engineering, and Communication at Stanford University. The Organizational Behavior and Analysis course presents learners with real-life cases and practical knowledge that draw from a broad spectrum of frameworks in organization theory, economic sociology, and the study of civil society organizations.

In early 2019, Prof. Powell came across an additional opportunity for iteration: developing a scaled, fully online version of the Organizational Behavior and Analysis course for education leaders. He was approached by the EdLEADers team, led by Executive Director of EdCareers Nereyda Salinas, EdLEADers Program Manager Heidi Chang, and GSE IT Director of Digital Solutions Shawn Kim. EdLEADers, a professional certificate program and collaboration with Stanford Graduate School of Business, was established in 2018 to address the unique opportunities and challenges of current and aspiring superintendents nationwide via a 16-month, 100% virtual, cohort-based program. Having produced courses with GSE faculty like Janet Carlson and David Brazer, Prof. Powell’s leadership-minded course made for an ideal complement to the program.

“This course gets people to think about organizations as things to be designed and improved”

Prof. Powell saw value in the collaboration, as well. First, it would give a diverse set of education leaders an opportunity to connect. The virtual format would also afford a broader dissemination of expertise, as EdLEADers cohorts include participants from districts across the US. Further, the online format provides flexibility, which is crucial for an executive audience with busy schedules.

Planning for iteration

The group first established a vision for the project. Principally, the course would take shape around authenticity and rigor. Prof. Powell saw it as a unique challenge: “Is there a way to take important and somewhat complex ideas, and make them accessible and real to people in the trenches? That’s a fun process.” Authentic and meaningful material would be tailored for full-time education leaders like early-career superintendents, charter management officers, and educational consultants.

As a first step, the EdLEADers team familiarized itself with the course material. Team members sat in on classes and took note of ideal content to convert to an online context, all the while considering the technical feasibility. The team noted how Prof. Powell made an effort to incorporate relevant current events, and how these materials complemented “time-tested” materials such as theory-based readings.

The team then considered Prof. Powell’s instructional style. His personable lecture format and theory-plus-application approach, signatures of the Organizational Behavior and Analysis course, would be incorporated and preserved as much as possible. Equally crucial to interaction was the variety of cases from inside as well as outside education that were discussed in class. With cases, notes Prof. Powell, “people can have divergent points of view. That’s one of the most important things.” To get the full flavor of the course, EdLEADers team members sat in on the full 3-hour class sessions each week.

From classroom lecture to high-quality video

The team then set out to capture the unique qualities of the live graduate level course in a way that could be translated to a virtual experience. The first element was lecture. Digital Media Producer Joe Sherman recorded class sessions, while Instructional Designer Leslie Cook took notes on key points of communication and interaction. Leslie and Joe then condensed the lectures into bite-sized videos that could easily be viewed online.

From the transcriptions, Julie Braly, a freelance instructional designer and GSE alumna, distilled Prof. Powell’s live presentation material into a series of compact, seven-minute scripts. The process was initially met with incredulity. “We just had a three-hour class, and you turned it into seven minutes,” commented Prof. Powell, laughing. But then he discovered there were often three scripts per class, and that creating a high-quality seven-minute segment required significant collective effort.

With polished scripts in hand, the team headed to the studio. Joe arranged sessions at Stanford Video, a professional video recording facility on campus, and guided the team in producing a series of high-quality videos replete with motion graphics and expert editing. “From a faculty member point of view, translating a course into a series of videos was easy, interesting, and energizing,” reflects Prof. Powell. “The actual filming part was a blast.”

Capturing the on-campus experience

In addition to lectures, the team wanted to preserve the nature of classroom discussions and assignments. A central feature of the course was the weekly memo assignments in which participants analyze cases by applying theory learned in class. The argumentation in these memos was crucial to connecting concepts throughout the course. Leslie and Julie regularly met with Prof. Powell to determine the fit for these adapted assignments, taking pains to maintain the spirit of the activity in the new online format.

Interaction was the last piece of the puzzle. A novel online platform, NovoEd, was chosen based on its collaboration model. NovoEd features an interface and module design that is collaboration-friendly and conducive to remixing ideas with peers. Additionally, in an effort to re-create the live course’s in-class breakout groups and peer-to-peer interactions, the team leveraged NovoEd’s functionality for weekly team discussion and share-out assignments; perspectives were analyzed within teams, and then shared out across the cohort via the course’s discussion forums.

Take-aways and reflection

Looking back at the process, Prof. Powell is proud of what he and the EdLEADers team created — what both he and his online students endearingly term “Survival 101” for education leaders. The case method endures, with real-world scenarios to anchor discussions, allowing learners to practice theory application, share points-of-view, and strengthen decision-making skills. Discussions around complex issues thrive online, and are finding new purchase with a diverse set of busy, full-time professionals.

At the same time, Prof. Powell notes the obstacles along the way. “Iterating on the scripts was a bit of work” and required flexibility on the part of everyone, he admits. “For many scripts, I said ‘this is too dense, let’s break it up.’ And so one session evolved into several.” Flexibility and communication were paramount, particularly as content was honed and media was developed.

The most indispensable element of the course building process was ultimately the shared sense of mission. As a partner in the process, Prof. Powell kept an open mind to the possibilities of digital media, and gave leeway for bolstering instructional materials if it meant a better student experience. In turn, the team is appreciative of Prof. Powell’s learner-centered approach, how far the Organizational Behavior and Analysis course has come, and how they could be a part of this iteration to move the course online.

If you would like to explore creating a blended or fully online experience for degree-bearing courses, contact shawnkim@stanford.edu; for professional learning, contact heidic@stanford.edu.

This article was written by Josh Weiss and published by The Office of Innovation and Technology at Stanford Graduate School of Education. For more information, contact josh.weiss@stanford.edu