Education Theory & Learning Technology study guide · Lesson 6 of 10

Brave New Words: AI & Education

Khan's vision · Khanmigo & the debate

This is the lesson's study guide from the FlashFeed app, plus its key facts and a self-test.

What you'll be able to do

  • Explain the personalization problem that Khan argues AI tutoring can finally solve.
  • Match Khan's three sci-fi inspirations to their educational vision and identify the work that is NOT an inspiration.
  • Explain Khanmigo — its GPT-4/OpenAI basis, Socratic design, and the 'conmigo' pun behind its name.
  • Recognize the book's stakes beyond K-12 and John Warner's contrarian critique of technology optimism.

Khan's vision

Salman Khan's 2024 book reframes AI not as an academic integrity crisis but as the first credible answer to a problem education has never solved: how do you give every student the benefits of a personal tutor?

Brave New Words

Brave New Words: How AI Will Revolutionize Education (and Why That's a Good Thing) (2024) is part memoir, part technology manifesto. Khan, who built Khan Academy from YouTube math tutorials, argues that conversational AI is not a threat to learning but potentially its greatest democratizing force. The book's title signals its thesis: the disruption is real, but the direction is positive. Khan opens with three works of speculative fiction that shaped his thinking — each imagining a world where personalized, adaptive instruction is available to any child regardless of wealth.

After November 2022

When ChatGPT launched in November 2022, the dominant response among educators was panic about cheating. Khan's reframing challenges that reaction directly. Rather than a weapon for academic dishonesty, he frames conversational AI as a patient, available-at-midnight tutor capable of following a student's reasoning, asking Socratic questions, and adjusting to their pace. The technology doesn't replace the teacher's judgment; it fills the hours between classes that previously went unsupported. The cheating panic, Khan argues, misidentifies the threat because it misidentifies the opportunity.

AI as equity engine

The deepest claim in the book concerns equity. Wealthy families have always been able to purchase one-on-one tutoring; students in under-resourced communities have not. This gap is not a minor inefficiency — Benjamin Bloom's 2 Sigma Problem (1984) demonstrated that individually tutored students outperform classroom peers by two standard deviations. No systemic intervention has closed that gap at scale. Khan argues that AI can: a tutor that understands where a specific learner is, what they know, how they learn, and what engages them has historically been available only through expensive private instruction. Making that tutor free and ubiquitous is, in his framing, the most consequential thing technology could do for human potential. Khan's three literary touchstones — The Diamond Age (an AI tutor transforms a poor girl's life), Ender's Game (challenge-driven simulation learning), and Asimov's The Fun They Had (a future child finds a classroom of peers exotic) — each illuminate a different facet of this vision. Notably, Snow Crash does not appear as an inspiration in the book's opening.

Khanmigo & the debate

Khan didn't just write about AI tutors — he built one. The second half of the book's argument lives in the product, the scope beyond K-12, and a genuinely mixed critical reception that reveals the fault lines in today's edtech debates.

Khan + conmigo

Khanmigo fuses "Khan" with the Spanish word conmigo — "with me." The name encodes the pedagogical stance: a tutor that accompanies rather than lectures. Built on GPT-4 in partnership with OpenAI, Khanmigo serves as proof of concept for the book's thesis: a conversational AI that guides students toward answers through Socratic dialogue rather than supplying them directly. The design choice is deliberate and philosophically important — a system that gives the answer on demand would undermine the productive struggle that makes learning stick. Khanmigo is meant to provoke thinking, not replace it.

Beyond K-12

Brave New Words refuses to limit the argument to school-age children. Khan extends the case for AI-assisted personalized learning to college admissions (navigating essays, fit, and financial aid), the workplace (continuous upskilling without the friction of formal enrollment), and civic participation (an informed citizenry able to engage complex policy). The framing treats AI tutoring as infrastructure — not a school feature but a lifelong utility, the way electricity or the internet are not just for students. This broadens the book's stakes considerably beyond what most edtech writing attempts.

The ethical framework

Khan is not uncritical of the technology. He explicitly names the risks his own tool inherits: hallucination (AI generating confident falsehoods), bias (training data embedding historical inequities), and privacy (sensitive student data flowing through third-party models). His response is an argument for responsible deployment — transparency, human oversight, ongoing evaluation — rather than a call to wait until the risks disappear. He positions this as the same bargain educators made with earlier technologies: imperfect tools adopted carefully, not avoided indefinitely.

The contrarian review

Reception was split. John Warner, in Engaged Education, argues the book "makes no real attempt to grapple with the implications of its own argument." His critique is that Khan substitutes technology optimism for rigorous pedagogical evidence — believing transformation will happen because the tools are impressive, not because controlled research shows they produce learning gains. Warner's review is not uniformly enthusiastic; it is a pointed challenge. The tension he names — between what edtech promises and what it demonstrably delivers — remains unresolved and is the central argument of the field.

Key terms

The personalization problem
Great education requires understanding where a specific learner is, what they know, how they learn, and what engages them — historically only possible with expensive private tutors. Khan argues AI can democratize this at scale.
Khanmigo
An AI tutoring assistant built on GPT-4 in collaboration with OpenAI. The name plays on "Khan" and the Spanish "conmigo" ("with me"). It uses the Socratic method, guides step-by-step, and never simply delivers answers.

Key concepts

The Diamond Age

  • One of three works of speculative fiction Khan credits with shaping his vision of personalized instruction for any child.
  • In its story, an AI tutor transforms the life of a poor girl.

Ender's Game

  • One of three speculative-fiction works Khan names as inspiration for his 2024 book on AI and learning.
  • He cites it for its vision of challenge-driven simulation learning, where students grow through immersive trials.

The Fun They Had

  • An Isaac Asimov short story Khan lists among the fiction that shaped his vision of universal tutoring.
  • In it, a future child finds a classroom full of human peers exotic and strange.

Key facts

Brave New Words

Brave New Words: How AI Will Revolutionize Education (and Why That's a Good Thing) (2024) by Salman Khan is part memoir, part technology manifesto. It frames AI not as a cheating threat but as the most powerful potential equalizer in education history — a way to give every student a world-class personal tutor.

After November 2022

When ChatGPT launched in November 2022, many educators' first response was panic about cheating. Khan's reframing runs against that reaction: rather than a threat to academic integrity, he casts conversational AI as a tutor that can finally personalize learning for everyone.

AI as equity engine
AI as equity engineAI-generated (gpt-image-1.5)

AI as equity engine

Khan argues AI can address historical education inequity by giving students in under-resourced communities access to individualized tutoring equivalent to what wealthy families buy in private instruction. This directly targets Bloom's 2 Sigma Problem — the gap between tutored and classroom learners.

What a Tutor Must Know
What a Tutor Must KnowAI-generated (gpt-image-1.5)

What a Tutor Must Know

Khan argues an effective tutor must understand four things about a learner: where they are, what they know, how they learn, and what engages them.

Personalization's Price Tag

Historically, the kind of individual attention a great tutor provides was only available through expensive private tutoring — affordable for wealthy families but out of reach for most students.

Khan + conmigo
Khan + conmigoAI-generated (gpt-image-1.5)

Khan + conmigo

Khanmigo is a play on "Khan" and the Spanish word conmigo — "with me." The pun captures the pitch: a tutor that learns alongside the student rather than lecturing at them. It was built on GPT-4 with OpenAI as Khan Academy's proof of concept.

Beyond K-12

Khan does not confine the argument to schools. Brave New Words explores AI's implications for college admissions, the workplace, and civic participation — treating personalized AI as infrastructure for learning across a whole life, not just the K-12 years.

The ethical framework

Khan does not pretend the technology is finished. He explicitly acknowledges current AI's limits — hallucination, bias, and privacy risks — and argues for a responsible deployment approach rather than uncritical adoption.

The contrarian review
The contrarian reviewAI-generated (gpt-image-1.5)

The contrarian review

Reception was mixed. John Warner, writing in Engaged Education, argues the book "makes no real attempt to grapple with the implications of its own argument" — leaning on faith in technology rather than rigorous evidence. The tension between technology optimism and pedagogical rigor sits at the center of today's edtech debates.

Built on GPT-4
Built on GPT-4AI-generated (gpt-image-1.5)

Built on GPT-4

Khanmigo was built on GPT-4 in partnership with OpenAI, serving as Khan Academy's proof of concept for the book's thesis.

Matched pairs

Each item on the left goes with the one on the right.

The Diamond Age (Neal Stephenson)
An AI tutor for a poor girl becomes her greatest educational resource
Ender's Game
Personalized, challenge-driven simulation learning
Asimov's "The Fun They Had"
A child finds a classroom of other children an exotic concept
conmigo
Spanish for "with me" — the root of Khanmigo's name

Self-test

Answer each question from memory, then check your work against the answer key.

  1. 1. According to Khan, what four things must an effective tutor understand about a learner?

    1. ATheir grade level, test scores, attendance, and school district
    2. BTheir reading speed, memory span, motivation, and screen time
    3. CWhere they are, what they know, how they learn, and what engages them
    4. DTheir IQ, learning style, family income, and career goals
  1. 2. Historically, why has the personalization problem gone unsolved for most students?

    1. AThat level of individual attention required expensive private tutors
    2. BThat level of individual attention arrived only with adaptive software
    3. CThat level of individual attention was cheap but confined to elite boarding schools
    4. DThat level of individual attention exceeded what even wealthy families could buy
  2. 3. Benjamin Bloom's 1984 finding, which Khan invokes, showed individually tutored students outperform classroom peers by how much?

    1. AHalf a standard deviation
    2. BTwo standard deviations
    3. COne standard deviation
    4. DThree standard deviations
  3. 4. Which is NOT part of Khan's argument about AI and the personalization problem?

    1. AAI can make individualized tutoring free and ubiquitous
    2. BAI can fill the unsupported hours between classes
    3. CThe tutoring gap is an equity problem, not a minor inefficiency
    4. DAI tutoring should replace the teacher's professional judgment
  4. 5. A district gives every student 24/7 access to an AI tutor that adapts to their pace. In Khan's framing, this most directly attacks which problem?

    1. AThe gap between what learners know and what tests can measure
    2. BThe gap between what tutored and classroom learners achieve
    3. CThe gap between what schools teach and what employers later need
    4. DThe gap between what parents expect and what schools report home
  5. 6. Which of the three works opening Brave New Words imagines an AI tutor transforming a poor girl's life?

    1. ASnow Crash
    2. BEnder's Game
    3. CThe Diamond Age
    4. DThe Fun They Had
  6. 7. In Neal Stephenson's The Diamond Age, what educational vision inspired Khan?

    1. AAn AI tutor becomes a poor girl's greatest educational resource
    2. BA gifted boy hones his strategic talent through escalating simulated battles
    3. CA shared virtual classroom lets children from every district learn together
    4. DA girl schooled by a machine at home marvels at the idea of a classroom
  7. 8. Which work supplies Khan's image of personalized, challenge-driven simulation learning?

    1. AThe Diamond Age
    2. B"The Fun They Had"
    3. CSnow Crash
    4. DEnder's Game
  8. 9. Asimov's "The Fun They Had" contributes which idea to Khan's opening?

    1. AA future child attends a school built inside a shared virtual world
    2. BA future child is lifted from poverty by a machine tutor
    3. CA future child finds a classroom of other children exotic
    4. DA future child finds a gift for command through simulated battles
  9. 10. What do all three of Khan's literary touchstones have in common?

    1. AEach imagines personalized, adaptive instruction available to any child
    2. BEach imagines a child's gifts drawn out by a devoted human mentor
    3. CEach imagines instruction that a whole community shapes together
    4. DEach imagines adaptive instruction tuned to a whole class rather than one child
  10. 11. Which work is explicitly named as NOT one of Khan's three sci-fi inspirations in Brave New Words' opening?

    1. AThe Diamond Age
    2. B"The Fun They Had"
    3. CEnder's Game
    4. DSnow Crash
  11. 12. Khanmigo's name fuses "Khan" with a Spanish word meaning what?

    1. A"Come along"
    2. B"With me"
    3. C"My friend"
    4. D"Teach me"
  12. 13. Khanmigo was built on which model, and with which partner?

    1. AGemini, in partnership with Google
    2. BGPT-4, in partnership with Google
    3. CGemini, in partnership with OpenAI
    4. DGPT-4, in partnership with OpenAI
  13. 14. Which teaching method does Khanmigo use by design?

    1. AAnki-style spaced repetition — scheduling each review at an expanding interval
    2. BDirect instruction — modelling answers through worked examples
    3. CThe Socratic method — guiding toward answers through questions
    4. DThe flipped classroom — assigning lessons at home and problems in class
  14. 15. Why does Khanmigo deliberately refuse to hand students the answer on demand?

    1. AIt would undermine the productive struggle that makes learning stick
    2. BIt would deprive the teacher of the correction that fixes a misconception
    3. CIt would reward the quickest students at the expense of the rest
    4. DIt would push students past the pace that their mastery data supports
  15. 16. A student asks Khanmigo for the solution to an equation. Consistent with its design, it should do what?

    1. ALet the student pick either a hint or the full worked solution
    2. BAsk a question that moves the student toward the next step
    3. CWalk the student through the full solution one step at a time
    4. DSwap in an easier equation the student can solve on their own
  16. 17. In his contrarian review, critic John Warner argues that Brave New Words is flawed because it...

    1. Atreats hallucination, bias, and privacy as risks already engineered away
    2. Bsubstitutes technology optimism for rigorous pedagogical evidence
    3. Curges educators to wait out the risks instead of adopting tools carefully
    4. Dnarrows the case for AI tutoring to schoolchildren and their classrooms
  17. 18. Beyond K-12, Brave New Words extends the case for AI-assisted learning to which three domains?

    1. AAdult literacy, immigrant integration, and jury service
    2. BTrade apprenticeships, the gig economy, and public health
    3. CCollege admissions, the workplace, and civic participation
    4. DGraduate study, professional licensing, and local government
  18. 19. Khan frames AI tutoring as "infrastructure." What does that framing claim?

    1. AIt is an invisible background layer, like plumbing or wiring — not a visible tool
    2. BIt is a base that other tools build on, like a power grid — not an end product
    3. CIt is a collective investment, like highways or dams — not a private venture
    4. DIt is a lifelong utility, like electricity or the internet — not a school feature
  19. 20. Khan answers the risks he names with a case for responsible deployment. Which three practices does that case rest on?

    1. ATransparency, human oversight, and ongoing evaluation
    2. BInformed consent, algorithmic fairness, and open access
    3. CTeacher autonomy, gradual rollout, and open-source models
    4. DData minimization, content filtering, and equity audits
  20. 21. Warner's central charge is that Khan trades on technology optimism. Which evidence would most directly answer that charge?

    1. APeer-reviewed benchmarks showing the underlying model aces hard exams
    2. BLongitudinal usage data showing students return to the tutor daily
    3. CAn independent survey finding students rate the tutor as helpful
    4. DControlled research showing students using the AI tutor learn more

Go deeper

Answer key

  1. 1. C Where they are, what they know, how they learn, and what engages themThe personalization problem is defined as understanding where a specific learner is, what they know, how they learn, and what engages them. The other options list demographic or diagnostic data, not the four dimensions Khan names.
  2. 2. A That level of individual attention required expensive private tutorsKhan's framing is economic: understanding each learner individually was historically only possible with expensive private tutoring, which wealthy families could buy and under-resourced students could not.
  3. 3. B Two standard deviationsBloom's 2 Sigma Problem (1984) found a two-standard-deviation advantage for tutored students — the gap Khan argues AI can finally close at scale. Mastery learning alone reached about one sigma in Bloom's comparison, and nothing in his ranking of alterable variables came near three.
  4. 4. D AI tutoring should replace the teacher's professional judgmentKhan explicitly says the technology does not replace the teacher's judgment; it fills the hours between classes. The other three are claims the book makes directly.
  5. 5. B The gap between what tutored and classroom learners achieveThis is Bloom's 2 Sigma gap — the tutoring advantage no systemic intervention has closed at scale. Khan frames free, ubiquitous tutoring as the first credible answer to it; the other gaps are real complaints about schooling, but none is the one Bloom measured.
  6. 6. C The Diamond AgeKhan opens with three touchstones: The Diamond Age, where an AI tutor transforms a poor girl's life; Ender's Game, for challenge-driven simulation learning; and Asimov's "The Fun They Had", where a future child finds a classroom of peers exotic. Snow Crash, also by Stephenson, is not among them.
  7. 7. A An AI tutor becomes a poor girl's greatest educational resourceThe Diamond Age features an AI tutor that transforms a poor girl's life — the equity vision at the heart of Khan's argument. The simulated-battle and marvelling-at-a-classroom visions belong to Ender's Game and 'The Fun They Had'; a shared virtual classroom appears in none of the three.
  8. 8. D Ender's GameEnder's Game supplies the challenge-driven simulation model. The Diamond Age supplies the AI-tutor-as-equalizer image, and 'The Fun They Had' the exotic-classroom image; Snow Crash is not a touchstone at all.
  9. 9. C A future child finds a classroom of other children exoticAsimov's story imagines a future child for whom a room full of peers is an exotic concept. The rescue-from-poverty and battle-simulation ideas come from The Diamond Age and Ender's Game; the virtual-world school belongs to none of Khan's three touchstones.
  10. 10. A Each imagines personalized, adaptive instruction available to any childKhan opens with three works of speculative fiction that each imagine personalized, adaptive instruction regardless of wealth. Two of the three put a machine rather than a human mentor at the centre, and each follows a single child rather than a class or a community.
  11. 11. D Snow CrashSnow Crash does not appear as an inspiration, while The Diamond Age, Ender's Game, and "The Fun They Had" are Khan's three cited works.
  12. 12. B "With me"The pun is on conmigo, Spanish for 'with me' — encoding a tutor that accompanies rather than lectures. The other glosses are plausible Spanish-sounding readings but not what conmigo means.
  13. 13. D GPT-4, in partnership with OpenAIKhanmigo is built on GPT-4 in collaboration with OpenAI, serving as Khan Academy's proof of concept for the book's thesis.
  14. 14. C The Socratic method — guiding toward answers through questionsKhanmigo guides students toward answers via Socratic dialogue and never simply delivers them. Direct instruction is closest to what the design deliberately avoids; the flipped classroom is a Khan Academy association, not the tutor's method.
  15. 15. A It would undermine the productive struggle that makes learning stickThe design choice is pedagogical: a system that supplies answers on demand undermines the productive struggle that makes learning stick — knowledge a learner constructs through guided questioning encodes more durably than knowledge merely received.
  16. 16. B Ask a question that moves the student toward the next stepKhanmigo is built to refuse direct answers and respond with guiding questions, because handing over the answer would undermine the productive struggle that makes learning stick.
  17. 17. B substitutes technology optimism for rigorous pedagogical evidenceWarner charges that Khan assumes transformation will happen because the tools are impressive, not because controlled research shows learning gains. The other options describe positions the book actually takes the opposite of.
  18. 18. C College admissions, the workplace, and civic participationKhan names college admissions, the workplace, and civic participation — treating AI tutoring as lifelong infrastructure rather than a school feature.
  19. 19. D It is a lifelong utility, like electricity or the internet — not a school featureThe infrastructure framing is about scope: AI tutoring is a lifelong utility the way electricity and the internet are not just for students.
  20. 20. A Transparency, human oversight, and ongoing evaluationKhan's responsible-deployment argument rests on transparency, human oversight, and ongoing evaluation — adopting imperfect tools carefully rather than waiting for hallucination, bias and privacy risks to disappear.
  21. 21. D Controlled research showing students using the AI tutor learn moreWarner's objection is that Khan expects transformation because the tools are impressive, not because controlled research shows learning gains.

Image credits

  • AI as equity engine: AI-generated (gpt-image-1.5)
  • What a Tutor Must Know: AI-generated (gpt-image-1.5)
  • Khan + conmigo: AI-generated (gpt-image-1.5)
  • The contrarian review: AI-generated (gpt-image-1.5)
  • Built on GPT-4: AI-generated (gpt-image-1.5)

Download all 10 lessons: PDF · EPUB