Key questions

Five hard questions, honest answers.

The questions a skeptical but interested school board member should ask about rethinking school in the age of AI, and where the evidence stands today.

  1. Question 1

    How do we know an AI-enabled model produces better learning? What would change our mind?

    We don’t know yet, so we shouldn’t pretend to. Early research is promising when AI is designed to teach, and worrying when it simply hands over answers (see the research below).

    So: pilot first, compare against a control group, and judge by what students can do without AI. If a year of guided AI use doesn’t improve device-free results, scale it back.

  2. Question 2

    What must students master without AI?

    Reading, writing, number sense and core math, the foundational knowledge of each subject, and independent problem-solving. You can’t judge AI’s answers without knowing things yourself.

    These are taught explicitly, practiced deliberately and checked in “Red,” device-free work every day. Memorization still has a place where it builds that foundation.

  3. Question 3

    What should teachers stop, start and continue? What training makes that realistic?

    Stop: lecturing by default and assigning homework AI can simply complete.

    Start: designing Red, Yellow and Green tasks, adding oral checks, and using AI for planning and differentiation.

    Continue: explaining hard ideas, catching misconceptions, holding high expectations and building relationships.

    Training has to be hands-on, ongoing and done in teams, with time for it built into the schedule, not a one-day workshop.

  4. Question 4

    Does it work for students who are behind, have disabilities or lack technology at home?

    It has to, or it fails. Students who are behind need more human help, not less: tutoring during the school day has some of the strongest evidence in education. AI practice should free teachers to give it.

    Vetted accessibility tools (read-aloud, translation, speech-to-text) can open doors for students with disabilities. School-issued devices and on-campus time cover access gaps. And every change should be checked for whether it widens gaps.

  5. Question 5

    What three measurable outcomes would convince a board it’s working?

    1. Growth on device-free assessments in reading and math, against a normed benchmark.
    2. Living-wage qualified graduates: the share who finish with a credential local employers pay a living wage for.
    3. Graduates on track one year later: the share in college, training or a living-wage job.

What the research says, and doesn’t

Bold ideas still need evidence.

Five findings worth knowing, with their limits. Knowing things and applying them isn’t either/or. Students need both.

Explicit teaching works best for beginners. For students new to a topic, clear explanation with practice and feedback beats leaving them to discover it alone. Kirschner, Sweller & Clark, Educational Psychologist, 2006
Tutoring is one of the strongest tools we have. Across 96 randomized studies, tutoring raised learning by about 0.37 standard deviations, with bigger effects during the school day than after it. Nickow, Oreopoulos & Quan, NBER, 2020
AI without guardrails can hurt learning. High school students with open AI access did better in practice but 17% worse on the exam. A tutor designed to guide, not answer, avoided the loss. One school, one subject. Bastani et al., PNAS, 2025
A well-designed AI tutor can work. In a Harvard physics course, students learned more than twice as much in less time with a custom AI tutor than in an active-learning class. Small, short and college-level, so promising, not proof. Kestin et al., Scientific Reports, 2025
Competency-based learning: promising, but mixed. Evidence favors models built on student choice and projects over flexible pacing alone, and schools must watch for widening achievement gaps. RAND, 2014
The takeaway Strong teaching, plenty of guided practice, more human support for students who need it, and AI designed to make students think. Then measure what they can do on their own.

Have a harder question? Disagree with an answer?

Thoughtful pushback makes these ideas better.

Get in touch