Nowadays, when you hear about students and generative AI, probabilities are you may get a taste of the debate about adopting tools like ChatGPT. Are they a help? (Yes! Great for research! Fast!) Are they harmful? (Whistle! Disinformation! Cheating!). However, some startups perceive the arrival of generative AI in the school environment as something positive and certain. And they create products to meet what they imagine shall be a certain market opportunity.
Now one of them has raised money to make this dream come true.
Magic School AI, which creates generative AI tools for educational environments, closed a $15 million Series A round led by Bain Capital Ventures. Denver-based MagicSchool began with tools for teachers, and founder and CEO Adeel Khan said in an interview that it now has about 4,000 teachers and schools using its products to plan lessons, write tests and create other educational materials.
It recently began building tools for students as well, delivered by their schools. MagicSchool will use the funds to proceed building on each of those paths, in addition to work on acquiring more customers, hiring talent, and more.
This latest round also includes support from some very high-profile investors. They include Adobe Ventures (whose parent company Adobe is very heavy on AI on its platform) and Common Sense Media (a specialist in age-based technology reviews that has entered generative AI with an AI guidelines partnership with OpenAI and chatbot rankings). The people in this round include Replit founder Amjad Masad, Clever cofounders Tyler Bosmeny and Rafael Garcia, and OutSchool cofounder Amir Nathoo. (Some of them were also seed investors in the company: it previously raised about $2.4 million.)
Khan didn’t reveal MagicSchool’s valuation for this round, but investors imagine that supporting app bets like this one is a natural next step for AI startups after investing tons of of hundreds of thousands of dollars in infrastructure corporations like OpenAI, Anthropic and Mistral.
“This is the moment of AI in education, a huge opportunity to create assistants for both teachers and students,” Christina Melas-Kyriazi, a partner at Bain Capital Ventures, said in an interview. “They have an opportunity here to help teachers with lesson planning and other work that takes them away from students.”
From Teacher to Preacher AI
MagicSchool, despite its name, didn’t appear out of thin air.
Khan started off as an educator, working initially for Teach for America when he first left academia. (And his interest in public service and the role of education may have began even earlier: At Virginia Tech, he was student government president during Virginia Tech shooting (Unfortunately, I have witnessed firsthand the devastation of gun violence.)
As a teacher, he showed early signs of developing an interest in entrepreneurship and leadership when he moved to Denver with the intention of starting his own school.
After working in various administrative positions at local schools, he eventually founded his own charter highschool called DSST: Conservatory Green High School, where the first group of graduates achieved a 100% acceptance rate to four-year colleges.
Taking a profession break from this frenzy of activity, Khan got here up with the idea for MagicSchool.
“It was around November 2022 when ChatGPT was dominating the headlines and generative AI was hitting the airwaves for most of the country,” he recalled. “When I was thinking about what I would do next, I started tinkering with it and immediately realized how much useful this new technology was for teachers.”
He developed early versions of the use of generative AI to create tools for teachers, visiting the schools where he taught and giving his former colleagues a tour of the possibilities. But it didn’t work.
“They found the interface clunky and just not catchy,” he said. Khan’s demonstrations for them inspired the desired “wow,” but left to their very own devices, teachers would use it once and never again.
“They told me, ‘I spent so much time trying to cajole him and get him to do what I wanted that I ended up not saving time, but costing him time.'”
His solution was to make more detailed customizations.
“Behind the scenes, we were just implementing really sophisticated prompts and making sure the results were what the teacher wanted,” he said.
Examples of what teachers create in MagicSchool include lesson plans, quizzes and tests, course materials, and reworkings of existing materials for lower and higher grade levels. MagicSchool is still tinkering with all of this. Khan said it works thoroughly with OpenAI APIs, but also with Anthropic and others. Behind the scenes, he said, the company is doing AB testing to determine what works best in a given scenario.
However, convincing teachers (who didn’t pay to use the product) and then schools (who did) to join MagicSchool wasn’t that easy.
“When we started working on the product, I didn’t get to meet with any schools or districts, including the one I worked at; there was so much fear in it all,” he said. “A negative headline about the use of artificial intelligence in schools… about how artificial intelligence will take over the world and robots” was enough to end any conversation.
That regularly began to change as society and industry more broadly adopted AI and more advanced models were introduced. Saving time was the most evident reason to use it, he said, but in addition they found it good for brainstorming and even offering supplements to what they may learn on their very own.
“I think teachers didn’t really know or expect what AI could do for them and their audiences,” he said.
What’s more, he makes a second argument for why bringing more AI into the classroom makes sense: It shall be a part of how all the pieces gets done, so it’s the school’s job to make sure students are ready for it.
Artificial intelligence is intelligent, but it is not “humanly intelligent”
That said, there are limits to the use of AI in any scenario, including the classroom.
“AI has a completely different type of intelligence than human intelligence. Humans have developed an emergent intelligence that is somehow the product of millions of years of pruning through natural selection. It is very holistic. It is very flexible, cognitively,” said Mutlu Cukurova, professor of education and AI at University College London, where a research laboratory has been operating for years examining various permutations of AI and learning. (One very realistic conclusion from recent article:There needs to be a hybrid approach, involving each AI and humans.)
“AI is designed intelligence, not emergent intelligence. That is, it is designed for a very specific purpose or set of purposes. AIs are brilliant at that specific purpose and show significant signs of intelligence, but this is a different kind of intelligence.”
This could also be particularly necessary for students and how they’ll learn in an AI world, or for teachers who may not have enough experience to know when an AI-powered version of learning materials resembling a quiz is not sufficiently All right.
Cukurova said that automating some tasks might be a helpful use case, but “the problem comes when teachers… don’t have enough experience to learn how to do these types of things on their own.”
Khan said MagicSchool intends to concentrate on this, especially when it comes to students. He said schools control what facilities they supply to students on the platform, and it is clear when they have used MagicSchool for an project.
This all sounds great in theory, but ultimately vulnerabilities may only develop into apparent in stress tests.
For example, will a cash-strapped school district rely on more AI input during teacher instruction? Or how will schools have the option to recognize when students are using AI tools outside of the classroom in ways that their teachers have not approved?
Cukurova says it will require a different form of AI education. “This is an important piece of the puzzle: How do we educate and train people to use AI effectively and ethically?”