Educators Technology

Perhaps the problem with AI professional development is not that teachers lack confidence.

It is that we keep trying to build confidence through demonstrations rather than experience.

This study followed 50 in-service primary teachers through an eight-month professional development program on integrating AI into English-speaking instruction.

During the program, teachers planned lessons, practised through microteaching, implemented those lessons with students, reflected on the difficulties they encountered, and received feedback from a supportive professional community.

After the program, teachers reported significantly stronger confidence in three areas: classroom management, student engagement, and instructional strategies.

The crucial ingredient was what Bandura calls enactive mastery, confidence developed by doing the work, encountering real problems, finding solutions, and seeing positive results.

This is an important reminder: a one-hour workshop may introduce teachers to an AI tool, but it rarely prepares them to integrate it meaningfully.

If we want confident AI-using educators, professional development must include sustained practice, classroom experimentation, reflection, feedback, and community support.

AI confidence is not downloaded. It is built through supported experience.

Reference

Yang, Y.-F., Tseng, C. C., & Lai, S.-C. (2024). Enhancing teachers’ self-efficacy beliefs in AI-based technology integration into English speaking teaching through a professional development program. Teaching and Teacher Education, 144, 104582

We may not have an AI integration problem. We may have a teacher-support problem.

A systematic review of 95 studies found that 65% examined how AI is used in teaching, while only 35% focused on teacher professional development.

The message seems clear: we are rushing to place AI in classrooms without investing equally in the people expected to use it responsibly.

Teachers need more than demonstrations of the latest tools. They need sustained opportunities to examine pedagogy, bias, privacy, accountability, and the effects of AI on student learning.

If institutions expect teachers to transform education with AI, shouldn’t they first provide the time, training, and support required to do it well?

Reference

Tan, X., Cheng, G., & Ling, M. H. (2025). Artificial intelligence in teaching and teacher professional development: A systematic review. Computers and Education: Artificial Intelligence, 8, Article 100355.

The machine may have helped precisely because it did not care how the teachers sounded.

That is what makes this study so interesting.

Thirty Taiwanese bilingual education teachers, whose English proficiency ranged from elementary to intermediate, used Google Assistant for speaking practice over approximately 17 weeks. They were asked to interact with it at least twice a week for ten minutes.

Their average speaking score increased from 47.73 to 66.29. The researchers reported improvements in fluency, content, vocabulary, and grammar.

Pronunciation, however, did not improve significantly.

But the numbers are only part of the story.

Eighty percent of the teachers associated their improved fluency with more frequent practice. Seventy percent pointed to reduced anxiety. They could speak, make mistakes, repeat themselves, and reformulate their sentences without worrying about embarrassment or “losing face.”

There is a provocative lesson here.

We often assume that the educational value of AI comes from how intelligent it is. But perhaps, in this case, some of its value came from something much simpler:

It was available, patient, private, and non-judgmental.

Reference

Tai, T.-Y. (2025). Exploring the effects of intelligent personal assistants on bilingual education teachers’ L2 speaking proficiency. Interactive Learning Environments, 33(1), 440–451.

A big thank you to Paul McAfee, PhD, MBA for sharing this gem!

I gave it a quick scan and it seems a really interesting read and it is free!

The book is packed with practical assessment ideas organized around three approaches:

1. Co-Creation: teaching students to use AI as a thought partner.

2. AI Friction: designing assessments that require authentic participation, communication, and productive struggle

3. Learning Without AI: protecting opportunities for students to build foundational knowledge and demonstrate their own thinking

The central message is to make the learning process visible, to ask students to show their work.

Looking forward to read it cover to cover!

Link in the first comment

APA reference

Yee, K., Uttich, L., Giltner, E., & Bojanowski, A. (2026). Show your work: Assessment in the age of AI. FCTL Press.

If AI can turn weeks of qualitative analysis into minutes, what exactly has been accelerated: insight, or our distance from the data?

Paulus and Marone examine how ATLAS.ti, NVivo, and MAXQDA market generative AI. Their warning is timely: speed is not rigorand chatting with documents is not the same as understanding people.

AI may assist qualitative research, but should it ever occupy the researcher’s seat?

Definitely an interesting paper I highly recommend!

Reference

Paulus, T. M., & Marone, V. (2025). “In minutes instead of weeks”: Discursive constructions of generative AI and qualitative data analysis. Qualitative Inquiry, 31(5), 395–402. https://doi.org/10.1177/10778004241250065

ChatGPT did not work equally well across classrooms.

After a year-long study involving 169 university students, Guo and colleagues found that students responded more positively when ChatGPT supported a concrete coding activity than when it was used for an abstract physics discussion.

Students generally found ChatGPT easy to use and helpful for confidence, motivation, and engagement. But they did not believe it could adequately address individual needs or replace their instructors.

The most important finding may be this: students’ responses improved when teachers redesigned the activity to include greater collaboration, exploration, and discussion.

In other words, the value did not come from adding ChatGPT. It came from improving the pedagogy around it.

Perhaps there is no such thing as an effective AI tool in education. There are only more or less effective ways of designing learning with it.

Reference:

Guo, F., Li, T., & Cunningham, C. J. L. (2025). One year in the classroom with ChatGPT: Empirical insights and transformative impacts. *Frontiers in Education, 10*, Article 1574477.

Perhaps generative AI has not destroyed assessment. Perhaps it has exposed how narrowly we have been assessing learning.

In this paper, Catherine Hartmann describes replacing traditional take-home essays with oral examinations in an upper-level humanities course. The goal was to make students’ thinking visible through explanation, argument, evidence, and real-time responses to questions.

The redesign required more than adding an oral exam at the end. Hartmann restructured the course to prepare students through discussion, social annotation, technology-free classes, and regular participation.

Students reported positive experiences, and the instructor’s time commitment was comparable to grading papers.

Oral exams will not suit every course. But they raise an important question:

If AI can produce the final written product, should assessment focus more closely on the thinking students can explain, defend, and adapt in the moment?

Reference:

Hartmann, C. (2025). Oral exams for a generative AI world: Managing concerns and logistics for undergraduate humanities instruction. College Teaching. Advance online publication.

What if students use AI unethically because academic writing feels irrelevant, impersonal, and disconnected from who they are?

This study of 24 multicultural undergraduates found a troubling disconnect: students understood that plagiarism was unethical, yet some still justified using GenAI because of high academic expectations, language difficulties, time pressures, and the perceived irrelevance of academic writing to their future careers.

This suggests that academic integrity cannot be addressed through rules and punishment alone.

Perhaps we should also ask, “What is it about our writing tasks that makes students want to disappear from them?”

Read Hysaj et al.'s paper for more details!

Reference:

Hysaj, A., Dean, B. A., & Freeman, M. (2025). Exploring the purposes and uses of generative artificial intelligence tools in academic writing for multicultural students. Higher Education Research & Development, 44(7), 1686–1700.

The most important AI adoption story in schools may be happening where students cannot see it.

A new midyear report examined how teachers across 12 Washington State school districts used a teacher-facing generative AI platform during the first four months of the 2025–2026 school year.

By the end of December:

2,891 teachers had used the platform
They exchanged more than 412,000 messages with the AI
Lesson-planning features accounted for 60% to 99% of usage in every district

AI was used in these classrooms primarily to create lesson plans, instructional materials, differentiation, assessments, images, and classroom activities.

While much of the public debate remains fixated on students using AI to cheat, teachers are quietly incorporating it into the infrastructure of everyday instruction.

The report also found that teachers serving larger proportions of English language learners and students receiving special education services used AI more frequently for differentiation, accessibility, student profiles, and lesson-delivery resources.

There was even a small positive association between teacher AI use and students’ midyear reading scores. But the effect was modest, and the authors correctly stress that the finding is correlational.

Teachers who voluntarily used the platform may differ from those who did not. No comparable association was found in mathematics.

So this is not proof that AI improves learning.

But it is evidence that AI is already reshaping how teaching gets designed.

And that raises a more urgent question than whether teachers are “using AI”:

What happens to instructional quality when lesson planning becomes increasingly mediated by systems created outside the school?

If AI is becoming part of the invisible infrastructure of teaching, then AI literacy cannot be limited to prompt writing, and school policy cannot be limited to student misconduct.

Link in the first comment!

Reference:

Esbenshade, L., Liu, A., Xiao, M., Tian, Z. V., Sun, M., Zhang, Z., Han, T., Lapicus, Y., & He, K. (2026). Generative AI in K–12 classrooms: A midyear implementation report (arXiv:2605.16277). arXiv.

“Transparency” may be the most overused word in AI governance.

But what does transparency accomplish if regulators still cannot see how the industry operates, inspect what is happening inside the systems, or intervene when something goes wrong?

In this important paper, Ferrari et al. argue that meaningful generative AI governance requires three interconnected conditions:

Industrial observability: Can we trace the companies, infrastructure, data flows, and power relations behind AI systems?

Public inspectability: Can qualified public authorities examine the models and their technical foundations?

Technical modifiability: Can these systems actually be changed when harms, risks, or legal violations are identified?

The authors’ point is crucial: observing AI without being able to inspect it is inadequate. Inspecting it without the power to modify it is equally inadequate.

Governance requires all three.

This reframes AI in a way I find especially useful. AI systems are not mysterious forces beyond human control. They are designed, financed, deployed, and maintained by identifiable institutions. They are regulatory objects that can be negotiated and governed.

So here is the uncomfortable question:

If governments cannot observe, inspect, and modify an AI system, are they governing it, or simply trusting the companies that built it?