One Thing to Teach in the Age of AI | Bryan Cassady | TEDxUniversityofSalford

Main Ideas / Concepts

• “The greatest exam question”: Exam difficulty can come less from what it asks, and more from what it doesn’t ask—for example, requiring students to produce good questions (not just provide answers).

• A training gap (banking model critique): Traditional education often follows Paulo Freire’s “banking model”—teachers deposit knowledge into students’ heads. Students are valued for holding more information.

• AI changes where knowledge lives: With AI, knowledge doesn’t need to be stored internally; it’s available on demand. The bottleneck shifts from “having answers” to asking the right questions.

• Danger of AI without good questioning:

  • AI can confidently generate responses to the wrong question.
  • The biggest risk is not hallucinated answers, but confident, correct-sounding answers to incorrect prompts.

• Questions as the multiplier: Better questions lead to better outcomes; bad or unexamined questions amplify errors.

• Learning should be about enough knowledge + practice time:

  • Teach people just enough to ask good questions.
  • Then provide time for application and thinking (“soak time”).
  • Learning loop: learn enough → ask good questions → apply quickly.

• Method: “How might we?”:

  • A prompt format emphasized as easy to teach and broadly usable: “How might we?”
  • Add tension/challenge to turn vague prompts into actionable problem-framing questions.

• Assessment focus changes: Rather than only “what to teach” or “what to cram,” focus on what to test—especially whether learners can generate and use effective questions.

• Einstein-style prioritization of questions:

  • Spend most effort on the question, because solutions are easier by comparison.
  • The contrast is with modern behavior: people often don’t spend even ~5 minutes thinking deeply before searching or using AI.

Key Methodology / Practical Instructions

A) Run “five good questions” as an assessment (replace answer-focused exams)

• Have learners/experts demonstrate understanding by producing five strong questions about the topic.
• Evaluate:
  • conceptual grasp,
  • ability to frame problems,
  • readiness to apply knowledge (not just recall it).

B) Reframe learning for the AI age (“teach enough, then get out of the way”)

• Teach enough knowledge for learners to form good questions.
• Avoid overload (“no firehose”).
• Shift from:
  • knowledge deposits → question withdrawals
• Then:
  • make time for thinking and application, so learners can use what they retrieve (from AI or other sources).

C) Use “How might we?” to generate actionable questions

• Start prompts with: “How might we?”
• Add a specific constraint/tension/challenge, e.g.:
  • “How might we help new employees feel confident, clear, and useful in their first 10 days?”
• Strengthen further, e.g.:
  • “How might we do this without increasing any management time?”
• Treat question quality as “discovery power”:
  • weak questions → shallow outputs,
  • good questions → better exploration of solutions.

D) Time compression strategy that preserves thinking (course redesign)

• Iteratively shorten a course:
  • 5 days → 4 days → 1 day (too exhausting; worsened results)
  • return to 2 days as the best balance
• Add structured follow-up pause time (“soak time”):
  • schedule next-week half-day check-ins,
  • let learners organize thoughts and return ready to apply.

E) Micro-training model (for high-performance, time-starved learners)

• Convert longer training into:
  • 10-minute trainings over 10 days
• Each day:
  • provide just enough information,
  • require learners to form and apply a good question,
  • move on rather than trying to “remember everything.”

Evidence / Studies / Reported Outcomes

• Company exercise (senior leaders + AI prompt):

  • 12 senior leaders created five questions to prove they understood what was covered in a strategy discussion.
  • Result: they produced not even five good questions, revealing a training/development gap in question-forming.

• BCG / Harvard / Wharton (AI productivity experiment):

  • With AI and good questions: work quality increased by ~40%.
  • With over-trust in AI without questioning: errors increased by ~19%.

• Training effectiveness:

  • With question-training: success up to ~78%.
  • Without training: ~45%.

• Application timing (important distinction):

  • “Applying” learning within 10 days is emphasized over mere remembering.
  • Reported figure: ~76% applying what they learned within 10 days (in one compressed-course run).

Story Examples Used to Reinforce the Lesson

• Narrator’s own mistake:

  • Built a 5-day innovation course and assumed the issue was content quality.
  • The real problem (“unasked question”) was: fit the course to clients’ schedules and provide time to think.
  • Fixing those questions improved course structure and outcomes.

• His son’s learning path:

  • Repeatedly “failed” traditional measures (kicked out of school, later dropped out).
  • Learned through experience and question-driven experimentation.
  • Example journey:
    • Asked: “How might I introduce other people to this?” (psilocybin context)
    • Built a retreat business in Mexico → moved to Jamaica → later evolved into a bat guano supplier with a successful global business.
  • Lesson: persistent, strong questioning turns challenges into opportunities (“figureoutable” with the right questions).

Central Lesson / Takeaway

• In the age of AI, the key teaching goal is not more answers or more content.
• Teach people enough to ask good questions, then provide time and opportunity to apply.
• The “one thing” to teach: how to ask better questions.

Speakers / Sources Mentioned

• Bryan Cassady (main speaker)
• Paulo Freire (creator of the “banking model of education” concept)
• Boston Consulting Group (BCG)
• Harvard
• Wharton
• Einstein (quote about spending time on the question vs. solution)