Summer preparedness: AI for the 2024-25 school year

The AI landscape in independent schools has been thriving over the last year, with hundreds of schools across the nation wholeheartedly embracing AI tools for their students and teachers — a notion which seemed unthinkable just two years ago, before the launch of ChatGPT.

At Flint (https://www.flintk12.com/), we like to think of ourselves as among the world experts on the integration of AI in independent schools. Throughout the 2023-24 school year, we’ve partnered with dozens of independent and international schools to build AI tools that best meet the unique needs of every school.

As we look ahead to the upcoming 2024-25 school year, it’s hard to pinpoint exactly where things will be a year from now. However, through reflection, strategic planning, and co-learning with the independent school community, we’ve put together a guide for how school administrators can form an approach for the Fall and beyond.

Getting up to speed on the AI landscape

To craft an effective approach to AI for your school, it’s first imperative to have a robust understanding of what AI is. This is a daunting task because AI is constantly evolving, so staying up to date requires keeping abreast of new developments periodically.

Some recent developments that we recommend reviewing and/or sharing with your AI committee include the launch of OpenAI’s GPT-4o (source), the launch of Anthropic’s Claude 3.5 Sonnet (source), Apple’s announcement to put ChatGPT in every iPhone for free (source), and this overview that our team wrote back in December which covers how AI in education evolved over the course of 2023 (source).

Most importantly, there are limitations with AI that you may be familiar with that no longer apply. Unlike the first wave of AI applications, which were essentially a chat interface with a large language model (LLM), many AI applications today use a mixture of LLMs as well as fact-checking tools, translation tools, and calculators under the hood.

This mixture can help to ensure much higher accuracy than what you might get with just an LLM, and means that the frequency of issues like hallucinations and inaccurate answers can be reduced by orders of magnitude by AI application developers who invest time in developing and improving these “under the hood” tools.

The AI in place within Flint, for example, uses a mixture of GPT-4o (the most robust AI model currently available), Google Translate for text-to-speech, OpenAI’s Whisper for speech-to-text, a retrieval-augmented generation system to pull from primary sources, and a proprietary calculator similar to Wolfram Alpha — all to create a much more accurate and safe AI experience for students and teachers.

Addressing concerns of academic dishonesty

When many teachers think of their students using AI, cheating is the first thing that comes to mind. This concern is entirely valid, and all the more pressing given the research showing that AI detectors do not work and have biased false positive results, which make them unsuitable for use in enforcing academic integrity (source).

However, despite the concerns of students using AI to cheat, student use of AI is rapidly growing. The latest studies show that the majority of K-12 students are using AI to help with their work. More worryingly, a cursory search of the App Store for homework help apps reveals an infinite number of apps specifically marketed toward helping students cheat with the help of AI.

With the student use of AI all but inevitable and impossible for schools to ban, it’s essential for schools to form a robust academic integrity policy that sets clear standards on acceptable uses of AI augmentation for student work.

We recommend not placing bans on specific tools (namely because there are a seemingly endless number of similar tools), but rather on empowering teachers with clear guidelines that they can apply to their assignments on a case-by-case basis. By standardizing these “buckets” of acceptable AI use, students also benefit from having a clear mental model that they can then apply to their decision-making when considering using AI tools to augment their work.

An approach that we’ve seen being implemented successfully is a “stoplight model”, whereby schools have given students guidelines for AI use on assignments — via a green, yellow, or red light — to provide clarity. A piece by Dr. Samuel Mormando, the Director of Technology, Innovation, and Online Learning for the Garnet Valley School District, covers this concept more in-depth (source).

Identifying what matters for your school

As important as it is to be aware of the major AI developments, it’s equally important to ensure that your school’s approach to AI isn’t determined by the technology itself. Instead, we recommend that schools focus on the use cases of AI that they feel will be most promising for their school — whether it be student use for academic support, teacher use for time savings, administrator use for workflow automation, or any combination of the above.

This approach, however, is seemingly contradictory. How do you take a use case driven approach to AI implementation if the use cases themselves are constantly shifting and expanding as AI development rapidly progresses?

An approach that we recommend is to first understand the goals of companies like OpenAI (creators of ChatGPT) and Anthropic (creators of Claude). These companies are pushing the forefront of AI model development, and are responsible for the lion’s share of technological progress that has led to the AI revolution in the first place. The plans of these companies are worth paying attention to, and they are fascinating.

Both OpenAI and Anthropic (and dozens of other top AI labs) have the explicit goal of developing artificial general intelligence, or AGI. AGI is defined as “artificial intelligence that matches or surpasses human capabilities across a wide range of cognitive tasks”. In short, if these companies successfully keep iterating on their AI technology (which they have had an incredible track record of doing over the last few years), they are headed towards creating AI in the image of ourselves. AI systems in the very near future could be just as capable, generalizable, and self-learning as humans.

So, as a thought experiment, let’s say we jump ahead a number of years — five, ten, however long it’ll be until AGI is widely available.

In that future, where your school could give every teacher a teaching assistant as good as a human expert, what would your school’s philosophy on effective use of those teaching assistants be? What would you have them do? Material creation? Grading papers? Extra academic support for students?

While we’re not there yet, working backward from an AGI future can help contextualize AI developments today, and help your school take a more use case-driven approach to implementing AI tools.

Our take: academic support is the “killer application” of AI in schools

With every new technological paradigm, there is typically a “killer app” — an application of the technology that is so impactful that it paves the way forward for the use of the technology itself. Microsoft Word and Excel were the killer apps for early Windows PCs, for example. For the iPhone, it was the ability to easily browse the web over 3G. For some technologies, like VR, the lack of a killer app has led to disappointingly slow adoption.

At Flint, we strongly believe that the killer application of AI in schools is academic support for students. While there is significant popularity at the moment for teacher-facing material generation tools, there are a few key reasons why we feel that AI for student support is the ideal path forward.

First, academic support is most closely aligned with the main goal of schools (educating students) and the results are in — AI tutoring works (source). For schools that lack academic support centers, AI can stand in to provide every student with 24/7 extra help in any subject. Every time that teachers feel burdened by the workload created by serving the unique needs of dozens of students, there’s an opportunity for AI to work alongside the teacher to both unburden them and to meaningfully personalize the learning experience for students — a goal that often falls to the wayside because it’s so challenging to do without AI.

Second, we’ve seen firsthand just how impactful AI for personalized learning can be for schools. We’ve worked with some of the world’s top independent and international schools (Westminster*, Cary*, Gilman*, Kinkaid, Crystal Springs, and UNIS, just to name a few) and have seen Flint being used for personalized learning by students across every subject at nearly every grade level (source). These schools have witnessed a transformation not only in the attitudes of their students and faculty towards AI, but also a significant shift towards more personalization at every step of the learning journey of their students (source).

At Flint, we’re building towards a future where AI for academic support — both in and out of the classroom, both student- and teacher-driven — comes to define what AI looks like in schools.

** denotes MISBO school*

Learn together, not alone

Finally, we feel that it’s critical that schools (even independent schools!) navigate AI integration together, not alone. We recommend reaching out to peer schools to learn how they’re approaching AI integration, and to understand what’s worked for them and what hasn’t.

The path forward for every school will be unique, and we believe that’s a good thing. AI shouldn’t lead to more standardization of the practice of teaching and learning. Instead, it should augment your school’s current approach, values, and philosophy. To give an example, we’ve seen Christian schools embrace AI in a way that strengthens theological foundations (source). While this specific example may not apply to your school, it goes to show just how unique every school’s path can be.

In addition to learning together with peer schools, our team highly recommends the work of the following two educators who have been working firsthand with schools across the country on AI:

Eric Hudson, former Chief Program Officer of Global Online Academy and current ATLIS Board Member, has been documenting his work and insights in his blog, Learning on Purpose (source).

Amanda Bickerstaff, the founder and CEO of AI for Education, has been doing wide-ranging PD work with schools nationwide, and has put together a library of free educator-centric resources on AI (source).

We have no affiliation with Eric or Amanda, but we’re a fan of their work! And, of course, if you’re interested in exploring Flint for your school or just in having a chat, feel free to reach out to either of us (the authors of this piece) at sohan@flintk12.com (Sohan, Co-Founder and CEO) or sami@flintk12.com (Sami, Head of School Partnerships and Business Development)!

Sohan Choudhury, Co-founder and CEO, Flint

Sohan is the Co-Founder and CEO of Flint, an AI tutoring platform for independent K-12 schools. Before founding Flint, Sohan co-founded and sold a video chat company. Prior to that, Sohan studied Computer Science at Georgia Tech where he was a Stamps Presidential Scholar. As a former tutor and a child of immigrants, Sohan is passionate about democratizing access to academic support for every student with AI.