When generative AI became widely available, many teachers were thrown into a new and uncomfortable question: How can we know whether a student has actually learned when a chatbot can produce a well-written text in seconds? Matt Miller takes this dilemma seriously in AI for Educators (2023), but the book is far from an alarm bell. It is more a pedagogical toolbox that shows how AI can be used to strengthen teaching, reduce time drains and — most importantly — shift assessment culture toward deeper understanding. From an educational-psychological perspective, the book's core idea is simple: When technology changes what is easy to produce, schools must change what they reward. This article explains the book's main points, supplements them with more recent insights about AI in education, and offers concrete strategies for learning, motivation, assessment and classroom management in an AI-driven school day.
Miller points out something many teachers have already experienced: assignments like take-home essays, reflection notes and "write a text about ..." were designed for a time when producing text was a clear signal of a student's competence. With generative AI, sheer volume of text is no longer a reliable measure of understanding. Educational psychology often distinguishes between performance (what we can produce here and now) and learning (a lasting change in knowledge and skills). When AI can lift performance without the student doing the cognitive work, the risk grows of a dangerous illusion: the product looks good while the learning is weak.
An important concept is cognitive effort. Deep learning often requires that the student retrieves knowledge, fails, adjusts and tries again. If AI is used as a shortcut that removes this effort, it can weaken the formation of conceptual networks, automation of basic skills and the ability to transfer knowledge. Miller therefore does not argue for more policing alone, but for smarter learning design: assessment forms and activities that make thinking visible.
AI for Educators can be read as a shift in the teacher role. Instead of spending energy detecting AI use, Miller encourages building instruction where AI either (1) does not give a free advantage because the assessment requires authentic understanding, or (2) is used openly as a learning tool, in the same way calculators and search engines were integrated into schools. He uses historical parallels: the calculator did not "kill" mathematics, but shifted the focus toward reasoning; the search engine did not "kill" knowledge, but made source criticism and information literacy more central.
A short paraphrase of the book's message can be summarized like this: When technology changes what students can produce, schools must change what they assess. Miller expresses this in the spirit of the following idea: AI is not just a threat to authentic work — it is also an opportunity to make learning more meaningful
. (The quote is brief and used to illuminate the main perspective.)
One of the most actionable chapters in the book is about assessment. Miller proposes several assessment forms that are robust in the face of AI — not because they ban tools, but because they make students' understanding observable. Here are four of the most relevant, with educational-psychological rationale:
Short, unprepared explanations ("explain photosynthesis without notes", "what is the main argument in the text we read?") force retrieval practice — the act of pulling knowledge from memory. This is one of the best-documented learning strategies in cognitive psychology. Oral assessments also reveal misconceptions and superficial understanding that can be hidden in a polished AI-generated text.
When students write, calculate or solve problems while the teacher follows the process, you can assess reasoning, strategy use and metacognition. A simple technique is to require think-aloud explanations or short reflection pauses: "Why did you choose this method?" This reduces the chance that the final product alone becomes the answer key.
Miller highlights assessment forms that require synthesis, perspective-taking and communication. In pedagogy this is about assessing competence through application and transfer: Can the student use the knowledge in a new context? A podcast about historical perspectives or a visual explanatory model makes it harder to pretend to understand, because one must make choices and justify them.
In group discussions, students must respond to counterarguments and adjust their thinking in real time. This resembles authentic competence demands in the workplace: collaboration, argumentation and communication. It also supports social-constructivist learning theories where understanding develops through dialogue.
The book is clear that AI is not just a problem to be "handled", but a tool that can strengthen pedagogy when its use is clearly framed. Miller suggests, among other things:
AI can rewrite texts at different levels, create explanations with multiple examples or provide scaffolded concepts. Educational psychology calls this tailoring support within a student's zone of proximal development. A practical example: the same subject content can exist in three versions — simple, standard and advanced — where all students work toward the same learning goal but with varying linguistic complexity.
Miller describes a "staged feedback model": AI can give the first round of feedback on structure, language and clarity, while the teacher spends time on disciplinary depth, reasoning and misconceptions. For learning, this is important because frequent feedback loops can increase motivation and mastery — especially when students can revise several times before final assessment.
Updated perspective (2024–2025): Experience from schools internationally suggests AI feedback works best when students get clear criteria and must submit a short "change log" (what they changed and why). This promotes metacognition and reduces passive acceptance of AI suggestions.
A concrete technique Miller proposes is using AI live when difficult questions arise: the teacher can demonstrate how to ask good questions (prompting), and then evaluate the answer critically together with the class. This connects digital competence with classic source criticism: What is the claim? Which sources should support it? Which concepts are missing? What might be wrong?
One of the most interesting methods in the book is an extended variant of Think–Pair–Share. Students first think alone, discuss in pairs, consult AI for counterarguments/examples, discuss again, and then share in plenary. From an educational-psychological perspective this is valuable for three reasons:
Practical tip: Have students submit a short note at the end with two columns: "This is what we thought before AI" and "This is what we changed after AI (with justification)". That way you assess learning, not just the final product.
Miller describes AI as a debate partner and as a character in historical conversations. This may sound like edtech entertainment, but it taps into several central mechanisms in educational psychology:
Classroom case example: In social studies, students can debate climate policy with AI representing different stakeholders (municipal economist, youth politician, industry leader). Afterwards, students must identify which values and assumptions guided the arguments and connect this to theory about interest conflicts.
Miller is clear that AI works best as a "first-draft tool." Examples:
Important update: Schools should establish clear routines for privacy. Do not enter sensitive student data into open AI services without approval and data processing agreements. Many municipalities and regions tightened practices in 2024–2025 for this reason.
Toward the end of the book Miller looks ahead: students inherit a future where AI will be part of most professions, but the content of jobs will change rapidly. Therefore, schools should train adaptability rather than try to predict the "right" future jobs. Two competency areas become particularly important:
Communicating precisely with AI — giving context, narrowing scope, iterating — is similar to information literacy from the search-engine era, but demands greater precision. A concrete teaching activity is to have students improve a prompt in three rounds and log what improved in the response and why.
Generative AI can produce convincing but incorrect information. Students should therefore be trained to verify: check against the textbook/primary sources, ask for sources, and assess whether answers are consistent. Practical exercise: Give students an AI response that contains three deliberate errors, and have them find, correct and document them with sources. There you assess critical thinking, not just fact recall.
AI for Educators by Matt Miller is most useful when read as pedagogy, not technology. The book describes a real problem: traditional assessment forms have become less reliable when AI can produce "perfect" student texts. But Miller does not stop at concern. He shows how teachers can redesign assessment and instruction so students must demonstrate understanding through speech, process, dialogue and creative products — and how AI can be used for leveling, idea development and faster feedback without erasing the teacher's professional judgement.
From an educational-psychological perspective the message is particularly relevant: Deep learning requires active thinking, feedback and reflection, and AI should be used to amplify these mechanisms — not replace them. When routine work is automated, the most human aspects of the teacher role become even more important: relationships, safety, motivation, assessment for learning and support for struggling students. AI can change the tools, but the teacher still designs the learning journey.