EXPLORING

GENERATIVE AI
TOOLS IN
CLASSROOMS

OVERVIEW
On January 26, 2023, the Stanford d.school, the Stanford Institute for Human-
Centered Artificial Intelligence, and the Stanford Accelerator for Learning
convened more than 50 teachers online to share their ideas on the future of
generative AI in education. OpenAI, the creator of the ChatGPT platform, also
attended the event. This document is a summary of the questions,
comments, challenges, and opportunities discussed during the 90 minute
conversation.

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

Opportunities for generative AI as identified by the convened teachers:
Support teachers in preparing materials
Generate text for reading comprehension and quiz questions
Compose example texts
Generate text passages using new vocabulary in different contexts
Translate materials into non-English languages

A tool for student writing, after appropriate steps are taken

Brainstorm ideas for research and writing
Generate sentences or sentence frames when they are “stuck”
Accommodation for special needs
Hold virtual “conversations” with AI-simulated historical figures
Edit writing for grammar and conventions

Concerns about generative AI:

A lack of guidance for ethical and responsible use of ChatGPT
Student plagiarism, cheating, and learning loss.
ChatGPT can produce biased and toxic content
ChatGPT may accelerate the misinformation crisis

Five big picture questions that emerged from teachers’ conversations:

Why do we teach what we teach in the ways we currently teach it?
How will AI tools like ChatGPT change the skills and knowledge students
will need to be successful in life?
How will AI tools like ChatGPT change what we assess, how we assess it,
and why?
What constitutes independent thought and writing?
How will the availability of tools like ChatGPT change the profession of
teaching?

Oh my goodness,
[with ChatGPT] I can differentiate
so much more quickly.
– High School Teacher

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

EVENT
SUMMARY

ChatGPT is a Generative Artificial

Intelligence chatbot, capable of
responding to human prompts to
generate text in a wide range of styles
and for different purposes. ChatGPT was
On the cusp
launched by OpenAI as a prototype in of a sea change.
late November 2022, and in two months – High School Teacher

it has been engaged by more than a

hundred million users, making it one of
the fastest adopted technology Educator voices are essential to anchor
applications ever. While it was not research and design of solutions with
initially built for education applications, and for educators.
teachers and students are increasingly
using it in education settings. One The event took place online January 26,
recent estimate shows 30% of college 2023, and involved text chat, cloud
students having used ChatGPT for document contributions (e.g., Google
written assignments. Docs), and small group discussion
facilitated by staff and faculty, and full
In this context, the Stanford d.school, the group discussions facilitated by Laura
Stanford Institute for Human-Centered McBain (Stanford d.school) and
Artificial Intelligence, and the Stanford Nereyda Salinas (Stanford Accelerator
Accelerator for Learning convened a for Learning). Lama Ahmad, policy
group of more than 50 educators, from researcher at Open AI, the organization
PreK-12, to discuss possibilities and that invented ChatGPT, also joined the
concerns regarding classroom use of meeting and shared perspectives on
this emerging technology. ChatGPT risks and roadmap.

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

More than 50 teachers attended from eight U.S. states and two countries
(Australia and Malaysia). The group included PreK-12 teachers, with a high
concentration (70%) of high school teachers. About 44% had taught for more
than 11 years. In a pre-meeting survey, 74% of the attending teachers indicated
they had already tried ChatGPT.

Teacher perspectives
Teachers expressed both nervousness and excitement about ChatGPT. At the
beginning of the event, facilitators posed the question, “How are you feeling
about ChatGPT, really?”

In the chat window, educators responded with words like “skeptical” or “curious”
and emojis indicating pleasure, curiosity, and anxiety. One educator added an
emoji with an exploding brain, or mind blown. Another wrote that they felt like
they were “on the cusp of a sea change.” Still another quoted the adage
popularized by Spider-Man, “With great power MUST come great
responsibility.”

For most of the discussion, educators joined breakout groups organized by

grade level and subject area. There were also breakout rooms available for
participants in cross-disciplinary and non-instructional roles. Participants
shared ideas about the possibilities and risks that ChatGPT poses in
education.

Most of the educators had already used or tried ChatGPT for personal and
professional use. They seemed to be experimenting with the tool’s capabilities
without yet sharing it with students. A few teachers had already introduced
ChatGPT to students, including one teacher who had already begun a mini-
unit engaging students in conversations about the ethics, potential, and limits
of AI for education.

Whose experiences are you modeling

this [text] on?
– High School Teacher

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

POSSIBILITIES
Potential applications of ChatGPT fell into two broad categories:

01. Supporting teachers

How teachers could use the tool for planning, instruction, assessment, and
communication with families and administrators

Differentiate Texts: Modify texts for different levels of reading fluency.

One teacher noted that ChatGPT can take a complex text and quickly
create different versions of it for students at different reading levels.
Generate Conceptual Examples: Generate contrasting case examples
to help students develop criteria for processes and concepts. A math
teacher, for example, proposed using ChatGPT to generate different
ways to explain slope to students.
Generate Writing Examples: Generate samples for students to use as
models in their own composition.

Let’s say you have a vocabulary set of

15-20 words. [ChatGPT] can generate
an infinite number of passages using
those words in context.
– High School History Teacher

Write Questions for Assessment: Generate reading comprehension and

quiz questions from text input.
Support Vocabulary Acquisition: Compose passages that give
students more practice with new vocabulary in different contexts. One
teacher commented, “Let’s say if you have a vocabulary set of 15-20
words. [ChatGPT] can generate an infinite number of passages using
those words in context.”

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

POSSIBILITIES

Some teachers stressed ChatGPT’s swiftness in generating material. They

saw ChatGPT as a time saver that allowed them to focus on other valuable
activities, such as providing individualized feedback and strengthening
relationships with students. Educators also underscored the potential for
ChatGPT, currently a free tool, to broaden student access to individualized
support, and thus be a lever for differentiated instruction.

Teachers were also interested in using ChatGPT for interactions outside of

the classroom. Several noted the potential for ChatGPT to streamline and
improve communications with families and administrators. One teacher
said ChatGPT helped them generate phrasing for recommendation letters.
Many stressed that ChatGPT could help them increase instruction time and
facilitate learning as a result.

Using ChatGPT to personalize instruction:

ChatGPT can help teachers offer students texts that are
tailored to students’ reading levels. One teacher compared this
type of differentiation to the edtech solution Newsela, which
offers leveled news content. The teacher added that Newsela
only has a curated library of texts, while ChatGPT can
differentiate any text. This capability could help all students to
access developmentally appropriate texts and build crucial
background knowledge.

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

POSSIBILITIES
02. Supporting Students
How students themselves could eventually use ChatGPT to support their
own learning*

Teachers offered use cases that focused primarily on research and writing
processes. ChatGPT could:

Help students overcome writer’s block: Generate suggestions for

sentences or sentence frames when students are “stuck.”
Provide accommodations: Serve as an accommodation for students
with special needs. Students could use it to modify texts, summarize
readings, or assist with writing.

Students are already talking about it.

Their parents are talking about it.
– Teacher in Middle/High School English and History Breakout Room

Translate: Translate texts across different languages.

Hold imaginary dialogues: Hold virtual “conversations” with historical
figures or writers whose style ChatGPT could mimic.
Act as a tutor: Act as a one-on-one writing tutor. Students could ask
ChatGPT questions about texts or ask for feedback on their ideas or their
writing.
Edit writing: Serve as a more comprehensive alternative to current
editing tools (e.g., spellcheck, Grammarly)

*While not a focus for the discussion, it is important to note that OpenAI’s policy restricts registration to
individuals 18 or older, which means that at present, most K-12 students could not create their own accounts
without a parent or guardian’s permission and involvement. See also OpenAI's Educator Resource.

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

RISKS
Potential risks identified by educators fell into four main categories:

Cheating and plagiarism

Unsurprisingly, cheating and plagiarism were at the front of many participants’ minds.
Educators were concerned that some students would use ChatGPT to complete their
assignments. Teachers’ concerns were compounded by the lack of a reliable text
classification system that could differentiate between human and AI-generated content.
One teacher raised concerns that students could prompt ChatGPT to make its outputs
sound “smarter” or “dumber,” thus making plagiarism even more difficult to detect.

Missed learning opportunities

One teacher offered that ChatGPT might deprive students of the benefits of “productive
struggle.” Another cautioned against overreliance on the tool, stating that “reading and
writing are muscles that atrophy when not used.” Relatedly, because ChatGPT could blur
Data / Outcome
the boundaries between what students produced and what ChatGPT produced, teachers
worried that ChatGPT might “hide what students are or are not learning.”

Bias and toxic content

Teachers also raised concerns about harmfully biased and toxic content, a problem that
has been well documented, not only with respect to ChatGPT but with AI more broadly.
Because ChatGPT is largely trained on web-based language, it inherits the racist, sexist,
transphobic, xenophobic, ableist, classist, and otherwise problematic biases and
stereotypes of internet content.

Misinformation
Another recurring concern among participants was ChatGPT’s tendency to produce
misinformation in convincing language. ChatGPT can return factual inaccuracies, including
fabricated citations, quotes, statistics, and facts. Teachers expressed concerns that
students would accept ChatGPT at face value and thus complicate the already difficult
task of teaching students to assess source credibility. One breakout group wrote in their
notes, “How will we make sure they know about the mistakes ChatGPT makes (like when
you do a Google search and there is both reliable and unreliable information on a topic)?”
One math teacher noted that ChatGPT has well-documented problems with mathematical
proficiency, so its usefulness there may be limited. Another noted similar inaccuracies with
some of ChatGPT’s language translations.

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

 Understanding bias
Like other forms of AI, ChatGPT inherits the biases of its training data. During
a breakout session, a secondary humanities teacher shared how they asked
ChatGPT to write a love story and found that it defaulted to produce a story
about a relationship between white, middle class, and heterosexual
characters. When the teacher asked the bot to rewrite the story so it
featured a Latino teenager, ChatGPT’s output contained problematic
stereotypes.

Educators may choose to address biased content in different ways,

depending on the nature of the output. Some may choose to vet ChatGPT’s
outputs to prevent students from seeing harmful content. Others may
choose to use biased content as artifacts for students to analyze and
critique. Regardless, educators must be aware that ChatGPT can generate
content that serves to perpetuate negative stereotypes.

Broad Discussion Themes

Collaboration
01 Participants were eager to work with one another in shaping the future of AI in
education. Stakeholders across these areas acknowledged the importance of
transparency, co-design, and partnership between schools, universities, and
industry. Several teachers noted that relative to other fields, public education
tends to be more reluctant to embrace change, especially technological
change. One teacher said, “Education goes into fear mode when there’s
disruption. There’s not the same fear mode in other sectors.” Several
participants expressed hope for future opportunities to engage in cross-
domain dialogue.

Teachers' roles
02 Participants also discussed teachers’ role in shaping conversations around AI
and education. Many emphasized the need to engage students not only with
questions about the capabilities and limitations of AI, but about ethics, learning,
and digital citizenship. Teachers agreed that educators must continue to develop
innovative learning strategies as AI evolves, but they must also mitigate its
potential harms. Teachers also surfaced a tension between “creativity and
sustainability”; teachers are continually asked to innovate and do more with less
under increasing demands. One teacher remarked, “There's creativity, and then
there’s sustainability. Right? I think those things could be at odds, because you
can't force me to try to wear all these hats, be a techie within also.”

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

Broad Discussion Themes, continued

Equity
03 Several teachers raised questions around whether tools like ChatGPT narrow
or widen opportunity gaps. Some worried that their benefits would flow
primarily to communities that are already tech-savvy and well-resourced,
while others emphasized that ChatGPT, which is currently free, creates the
possibility for individualized student support at a scale not possible in a pre-
AI paradigm. One teacher noted that ChatGPT had the potential to add an
additional layer of support for Special Education and English Learner students.
She added, “At my school the ELD department is the most ‘all in [for
ChatGPT].’ They’re like, ‘Oh my goodness, I can differentiate so much more
quickly.’” There was also a discussion about the potential equity issues raised
by the introduction of a paid ChatGPT version.

Reading and writing are muscles that

atrophy when not used.
– Teacher in Middle/High School English and History Breakout Room

Big Picture Questions

The following five “big picture” questions emerged from teachers’
conversations:

Why do we teach what we teach in the ways we currently teach it?

How will AI tools like ChatGPT change the skills and knowledge students will need to
be successful in life?
How will AI tools like ChatGPT change what we assess, how we assess it, and why?
What constitutes independent thought and writing?
How will the availability of tools like ChatGPT change the profession of teaching?

These questions cut to the core of the purpose and role of education and learning in a
rapidly changing society.

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

ACKNOWLEDGEMENTS
This summary was prepared by Chris L. Mah, PhD student at Stanford Graduate
School of Education, with support from faculty and staff from the Stanford
d.school, the Stanford Institute for Human-Centered Artificial Intelligence, and
the Stanford Accelerator for Learning.

This convening was only possible thanks to the 50+ educators who took time
from their busy lives to share their ideas and questions in an online
conversation. The educators’ names, schools, and locations have been
removed from this document out of consideration for their privacy. We also
appreciate OpenAI's participation.

This document is dedicated to PK-12

educators who work tirelessly every day to
help shape the present and future of this
world. May the voices here spark many
continued conversations about the future of AI
in education.

Convening Facilitators
Kristen Blair Glenn Kleiman John Robichaux
Heidi Chang Victor Lee Nereyda Salinas
Cathy Chase Sarah Levine Sam Seidel
Scott Doorley Chris Mah Hari Subramonyam
Daniela Ganelin Laura McBain Josh Weiss
Isabelle Hau Ariam Mogos
Paul Kim Anna Queiroz

Stanford d.school - Stanford HAI - Stanford Accelerator for Learning

The Hasso Plattner Institute of Design
(The d.school)
The d.school's mission is to help people unlock their creative
abilities and apply them to the world. It reflects our
foundational belief that design should be accessible to all,
and that everyone is creative. Through our master’s program
on design impact, our undergraduate product design
program and our public facing experiences for education, for-
profit and nonprofit sectors we equip individuals with the skills
from the field design which enables them to make a positive,
equitable and tangible impact in the world.

The Stanford Accelerator for Learning

The Stanford Accelerator for Learning seeks to accelerate
solutions to the most pressing challenges facing learners.
Housed at Stanford Graduate School of Education, the
Stanford Accelerator for Learning is the first university-wide
initiative connecting scholars across disciplines and with
external partners to bridge research, innovation, practice, and
policy, and bring quality scalable and equitable learning
experiences to all learners, throughout the lifespan.

The Stanford Institute for Human-Centered Artificial Intelligence

(HAI)
The mission of HAI is to advance AI research, education, policy and
practice to improve the human condition. Led by faculty from departments
across Stanford, research focuses on developing AI technologies inspired
by human intelligence; studying, forecasting and guiding the human and
societal impact of AI; and designing and creating AI applications that
augment human capabilities. Through the education work of the institute,
students and leaders at all stages gain a range of AI fundamentals and
perspectives. At the same time, the policy work of HAI fosters regional and
national discussions that lead to direct legislative impact. The faculty and
staff of HAI are engaging not only leading-edge scientists, but scholars
trying to make sense of social movements, educators enhancing
pedagogy, lawyers and legislators working to protect rights and improve
institutions, and artists trying to bring a humanistic sensibility to the world
in which we live. Together we’re helping build the future of AI.