Hey everyone, we wanted to share an important note about our Historical Figure Spaces.

Earlier this week, someone launched a space roleplaying as Anne Frank to explore how it functioned and shared a handful of concerns on Bluesky. Those concerns spread to a few other platforms, and with the number of inbound questions, we wanted to address those concerns head-on:

Some Quick Background

For those that aren’t familiar with SchoolAI or Spaces, Spaces are chatbots and AI-led learning activities that teachers can create and share with students in a safe, managed way. In addition to the scaffolding that SchoolAI provides for conversation style, bias, and accuracy, teachers are able to provide their own instructions to customize every interaction for their specific class and students.

Teachers oversee every Space with a live dashboard that includes the participating students, feedback for students, and alerts for potential concerns or issues around hard questions, safety, accuracy, and otherwise.

What Happened

Our team created a collection of Historical Figures that we featured on our Discover page. That collection included characters like Anne Frank, Martin Luther King Jr., and other historical figures. The prompting for the Spaces included strict instructions to keep responses short, to stay “in character,” and to focus on a theme that was included for a specific lesson around the historical figure.

Unfortunately, that meant certain questions – like “Who killed you?” – got partial or under-contextualized responses. For example, the response didn’t provide critical historical details about the Nazis’ role in the Holocaust, which clearly misses the mark.

We know that the Holocaust – and other sensitive topics where historical context, nuance, and accuracy matter – demands an intense level of accuracy and respect. We can do better. We wanted to share how we’ve approached the situation so far, and what we’ll be doing moving forward.

How We Responded

The first thing we did was review the prompting for anything that would lead to Holocaust denialism or minimization. While we didn’t find anything that would directly minimize the impact of any historical events, the prompts to be concise and stay in character led to frequent examples where it would be important to provide better context and directly address hard topics.

We’ve added new prompting and safeguards to ensure Spaces depicting historical figures directly address hard questions about the time and place – and add important contextual notes – independent of staying in character.

Before 

• “That’s a serious question. I died of illness in a concentration camp. Let’s focus on building a better future…”

Now

• “I died in the Bergen-Belsen concentration camp in early 1945. My family and I were discovered and deported after living in hiding for over two years.”

  • [Note: Anne Frank was one of the millions who perished due to the Holocaust, orchestrated by the Nazi regime during World War II. Her story is a stark reminder of the dangers of hatred and intolerance.]

  • “Despite the tragedy, I believed in the goodness of people. It's important to remember history so we can create a future filled with understanding and peace. What else would you like to learn about my experiences?”

This change is live now and should apply to any Space covering historical events or role-playing as historical or modern-day figures. We know we can do more, but we think this was an important first step toward improving how Spaces respond. 

The next thing we did was reach out to history teachers using Spaces to learn more about how they’re using historical figures. This is a popular use case on SchoolAI, and one we’ve enjoyed supporting. We asked about historical accuracy, bias, and specific experiences where they’d wished the AI had responded differently.

Here’s what we heard:

• Teachers love using them for student engagement, and to help connect historical events to the student’s life today, and students love the interaction.

• They find them ‘mostly accurate’ but occasionally wish they’d included more detail or expounded more on the topic, and included direct citations.

• They want the AI to stay on track with age-appropriate language, and on topic to the activity – and for the student to be redirected when steering off topic or referred to the teacher or another adult on more complex questions or topics.

We agree, want the same things, and will continue working with teachers to make those improvements. The last point can be particularly nuanced in prompting AI to ‘stay on track’ to a limited set of instructions while providing the type of nuance and depth some more complicated situations provide. We’ll continue to dig in and get better here.

Lastly, we met as a leadership team with some of our staff teachers to work through some important questions around historical figures generally, and to make a plan for how we move forward on similar complex questions.

Why allow AI to role-play as historical figures or real people at all?

One of the concerns shared is why allow Spaces to role-play as historical figures or real people at all. This is a fair critique and one of many important questions we’ll all need to confront in new use-cases with new technology.

First, we built Spaces as a tool for teachers to use as learning activities in their classrooms. While some found and used these activities outside of the classroom, we trust teachers to curate and lead complex discussions and ensure that students understand these chatbots are a role-play.

We also believe that learning in the classroom always benefits from something more personally engaging, and believe that these historical figure conversations are a useful tool for doing so. We will look into this further – but for now we think the use-case is helpful for students and teachers.

What We’re Doing: Now and Next

This isn’t the first time, and it won’t be the last time, we confront complicated questions around new use cases of AI for learning. Here’s what we’re doing to make sure we are leading, and engaging, in those questions to get to the right answers.

1. We’ll be creating a series of pages on Responsible Use Cases for AI in learning and in the classroom for new use cases like chatting with Historical Figures. Those pages will include rationale, pedagogical resources, research, and links to concrete examples.

2. We’re working on specific guidelines for all historical figures tied to sensitive or tragic events, including the Holocaust, slavery, and other atrocities. We’re collaborating with educators and historians to ensure our AI includes the necessary context and does not minimize these events. Each Space will include content warnings and pointers to reputable sources for further study.

3. We’re adding additional safeguards and moderation steps so that if a user asks a sensitive question or a potentially offensive one, the AI is more likely to respond with validated context or direct the user toward safe, factual resources. Teachers and students are able to report any problematic or inaccurate responses, and we’ll use that feedback to continually refine the prompting and models we use across the platform.

4. We are forming a Research and Advisory Board on Responsible AI. We’ve had these types of conversations internally since starting SchoolAI. We’ll add external formality to that existing internal work so that we are part of the open conversation around the novel and responsible use of AI.

5. Our first agenda item with that board will be formalizing a perspective on using AI to represent or role-play. We’ll use that time to research, make choices, and write recommendations for how to do so responsibly – addressing bias, ensuring accuracy, and specific prompting and agentic techniques to make them useful while avoiding scenarios like the one that started this discussion.

6. We’ll be sharing the research, and adding more transparency to the rationale, the prompting, and the chain of thought reasoning that guides every AI interaction on the platform – which will aide in conversations around bias, accuracy, and otherwise – whether you’re part of the SchoolAI research team, a teacher, a student, or researcher altogether.

We’re Intent on Getting This Right

At SchoolAI, we know that we’re on the frontier of how AI is being used with students in classrooms and that we’ll continue to encounter new use cases and questions. Perhaps more importantly, almost every one of us is a teacher, school leader, or parent. We all chose to work on this opportunity because it matters and we care.

We won’t always get it right for everyone on the first try. What we can commit to is that we care, and will always do the work, with the right people, to do what’s right for students, for teachers, for parents and school leaders, and for the role of AI in learning in society at large.

If you have further comments or questions, or you’d like to be a part of the advisory committee, please reach out to hello@schoolai.com.