Humans in the Loop: Insights from Codesigning AI in Real Classrooms

Since the debut of ChatGPT in November 2022, artificial intelligence in education technology has exploded. New products and features claim to revolutionize learning and change teaching as we know it, adding to an already saturated edtech market. In the 2023-2024 school year, Learn Platform tracked over 57 billion interactions across 9,000 edtech products, while Reach Capital identified over 280 generative AI edtech tools in 2023 alone. 

Analysis of the AI tools in education database compiled by Edtech Insiders highlight the bold, optimistic promises of these tools: 123 feature “personalization” or “personalized” content/experiences, 26 claim to ‘automate’ tasks, and 16 are referred to as “revolutionary” or “revolutionizing.”

At Leanlab, we wondered: Are these tools living up to their promises? How do these tools work in actual classroom environments? Between January and August 2024, Leanlab worked with five companies developing AI-powered edtech tools to understand how well the tools were meeting teacher expectations, and how the tools could be improved.

We also wondered: What do the teachers using these new technologies think of them? Surveys of educators find excitement at the prospect of using AI-powered edtech to generate content and prepare instructional materials like lesson plans and assessments, brainstorm creative new ideas for classes, and differentiating lessons for students. Surveys also find optimism about the future of AI in education, alongsides concerns about privacy, lack of training and support, and students using AI for cheating. What would teachers think of the tools being trialed during codesign research? 

After engaging more than 40 teachers in codesign research studies, we found a considerable gap between their expectations and the current state of AI-powered edtech. 

We also found that participation in codesign research helped bridge this gap, bringing together educators and developers to address challenges, build trust, and create better tools for teachers and learners.

The Great Divide: Teacher Expectations vs. AI Reality

While the potential of AI in education is exciting, there’s often a gulf between what teachers expect AI-powered edtech to do and what it can actually deliver. Teachers approach AI-powered tools with high hopes. They want: 

• Tools that save them time 

• Tools that provide personalized learning experiences for their students, and

• Tools that can take over tedious administrative tasks.

However, the reality of using AI tools in the classroom often falls short.

Leanlab worked with five companies developing AI-powered edtech in 2024 to proactively address this gap. All of the products were supplemental solutions, from teacher copilots to instructional support tools. Leanlab undertook six studies, with two consisting of literature reviews and four engaging teachers and students in the codesign research process. 

Based on the research goals of each company, Leanlab researchers developed logic models, conducted research ready audits, and partnered with educators in codesign studies. From product validation to feasibility studies, researchers engaged teachers and students in the research process: codesigning the research questions, trialing the tool in the classroom, and providing actionable feedback for the companies.

Research participants included dozens of educators, along with their students, in a variety of contexts—elementary, middle, and high schools from coast to coast. 

Across the six studies for the five companies, Leanlab researchers identified consistent themes regarding the intentions of the companies engaging in research, teacher expectations of the AI-technology, and the gap between these intentions and expectations. 

Expectations

In the four studies with teachers, teachers consistently identified three priorities for AI-powered edtech. They wanted tools that:

1. Save them time

2. Support differentiated instruction and enhance student engagement

3. Support them in instructional design, delivery, and feedback

These findings reinforce sentiments in national teacher surveys that found generating content and differentiating lessons for students to be top priorities for respondents. They also largely map to the intentions of the five companies we worked with.

Intentions

Predominant themes from their logic models or theories of change include the intention to:

1. Personalize and differentiate learning

2. Enhance student engagement and agency in learning

3. Improve student engagement and outcomes

4. Support educators in instructional design, delivery, and feedback

While the findings from Leanlab’s six studies show strong alignment between teacher expectations and company intentions, the studies surfaced significant gaps between the expectations and reality of actually using the tools in classrooms.

Reality

Across all four studies that engaged educators and students, there were challenges to expectations specifically caused by: 

1. Technical issues

2. Differentiation shortcomings

3. Trust in AI-generated content

4. Integration challenges

5. Teacher familiarity and confidence with technology

For example, a teacher generating content was met with slow load times that impacted student engagement, saying “I think I expected things to load quicker in this day and age.” Similarly, educators testing another product faced delays, with content generation taking up to an hour—eating into valuable class preparation time.

In another study, a teacher found that differentiation capabilities were not sufficient for them to customize or modify content to meet their students’ individual needs.

These experiences highlight a common theme: while AI tools promise to save time and enhance instruction, technical glitches, differentiation shortcomings, and usability issues often get in the way of these benefits. Teachers start to question whether these tools are worth their time and their experiences reinforce any existing distrust of the technology or the content it produces.

Leanlab’s findings also indicate that teachers are still skeptical about the accuracy, capabilities, and consistency of AI-powered edtech tools. Beyond accuracy concerns of jumbled facts, there are concerns that these tools lack the intuition and pedagogical knowledge to sufficiently differentiate and personalize, effectively scaffolding up or down. This sentiment is reinforced every time a teacher has an experience with inaccurate or inconsistent AI-generated content, and may explain why not many teachers are currently using AI tools for grading.

Humans in the Loop: Bridging the Gap with Codesign 

These findings and experiences highlight the power of codesign and the benefit it brings to both edtech companies and educators. By facilitating the hand-in-hand work of product teams and school design partners during research and development, we ensure that humans are “in the loop” as AI tools evolve to meet real classroom needs. 

Edtech companies can use codesign to validate that initial AI use cases are solving an actual problem faced by teachers or students in the classroom, and they can build out functionality based on their candid feedback. By working with intentionally diverse cohorts of teachers and students, they can design for multiple contexts early on in the development process. 

During Leanlab’s AI codesign studies, researchers made product recommendations based on teacher feedback, allowing developers to quickly address technical issues and integration challenges while incorporating recommended changes in their product roadmap. Take the earlier example of differentiation limitations faced by a teacher. Initially, the tool didn’t support students with lower writing skills—those who needed the most help. Through codesign, teachers were able to voice this concern, leading to a recommendation for integrating additional scaffolding to make the tool accessible to all learners.

Building trust with educators is as essential as addressing technical issues for companies interested in sustainable use of their AI-powered product. Keeping humans in the loop through codesign research allows educators and researchers to inform the data sets and foundational pedagogy guiding these tools. Companies can build trust by:

1. Incorporating learning science principles in their product

2. Sharing a literature-based logic model that clearly articulates a tool’s intended outcomes

3. Tightly training AI models on reliable, high-quality resources

4. Working directly with teachers to codesign tools

The Leanlab team found that using a multi-phase study structure or rapid cycle evaluation (where feedback is shared and product changes are made in cycles throughout the study) also allowed for real-time product improvements, which didn’t just improve the product, but also built trust with teachers. 

Partnering with a third-party research organization like Leanlab to conduct codesign research provides companies with timely and evidence-based recommendations for product improvements, as well as engages diverse participants in a variety of contexts—helping companies develop equitably-designed and accessible AI tools. 

Codesigning Better AI Edtech

The future of AI-powered edtech is undeniably bright, but only if we continue to close the gap between expectations and reality. At Leanlab, we believe that codesign research is the key to making this happen. By bringing teachers and developers together to collaborate on the design and testing of AI edtech, we can ensure that these technologies meet the diverse needs of teachers and learners, and live up to their potential to transform education.

Codesign Product Research

Curious how codesign research can help bridge the gap for your AI-powered product? Learn more about Leanlab’s approach to creating trustworthy and effective tools by diving into Codesign Product Research. Let’s shape the future of AI in education, together.