How to Hype AI in Education
Brave New Words: How AI Will Revolutionize Education (And Why That’s a Good Thing) by Salman Khan
OR
How to Hype AI in Education
By Jacob Pleasants
Brave New Words by Sal Khan has been out since May 2024, but I only recently became aware of it. When I found out that Khan had written a book, I was curious to know more. It’s not that I wanted to be “enlightened” by his take on AI in education. I had very few hopes that the book would offer many well-researched claims or well-reasoned arguments. I think part of me was just looking for some chum, some grist for the critical mill. My reasons were admittedly not all that great, but regardless, I got a copy from my library and started in.
Partway through the book, I actually had a pretty powerful insight. This book is instructive, but not in the way that it intends to be. What I gradually realized is that this is a playbook for how to hype AI in education. As a hype piece, it really is exemplary. Across 200+ pages, Khan wields a wide-ranging arsenal of overly optimistic claims about the past, present, and future. Many of his techniques are classic and well-worn, but his execution is spot-on.
Let me be clear: If you want to learn something about AI in education, I do not recommend that you read this book. True to my expectations, the claims and arguments in this book are unsubstantiated (there are essentially no citations in this book, though there are plenty of quotes from industry insiders) and mostly self-serving (it is largely an advertisement for Khanmigo). But if you want to learn something about how hucksters hype AI in education, it is actually rather informative.
On the other hand, you might find reading this book to be pretty painful, so I’ve gone ahead and saved you from the slog. I have pulled together the greatest hits from the book and crafted what I believe to be a pretty solid user manual for “How to hype AI in education.”
If you are looking to further inform yourself about the phenomenon of AI hype, I do suggest the following:
Tackling AI Hyping by Sloane, Danks, and Moss (2024) in AI and Ethics
AI Snake Oil by Narayanan and Kapoor (2024). Here’s an annotated example of a piece of EdTech journalism in which they identify a slew of common hype moves.
How to Hype AI in Education: A Playbook
Lay Some Groundwork
Sometimes people criticize technology as being “a solution in search of a problem.” Therefore, you want to be sure to lay some groundwork for the AI solutions you’re going to be presenting. Fortunately, when it comes to schooling, problems are not hard to find. Pretty much everyone agrees that there is lots of room for improvement. You can introduce some specific problems as needed, but be sure to stick to the big picture as much as possible and paint with a broad brush.
Create a caricature of current schooling practices that portray it as antiquated, one-size-fits-all, inefficient, and ineffective. Basically, current schooling is a failure and is badly in need of disruption. Teachers lecture, students listen, with little opportunity for individual student feedback and attention. Feel free to remind people of this early and often.
“A decent education is expensive anywhere in the world. In the US, Louisiana spends roughly $10,000 per student per year; New York spends $40,000. In India, government schools might spend anywhere between $500 and $1,200 per student per year: Despite the range in resources, the fundamental model is the same. Students are moved lockstep through curricula, oftentimes feeling lost or bored. If a student doesn't keep pace in understanding a foundational concept, the class keeps moving. Limited support exists for personalization or for revisiting gaps, much less for one-on-one tutoring. This is despite the fact that many classrooms have students at a wide range of preparedness- some may be ahead of pace while others may be two or three grade levels behind.” (p. 172)
***
“I believe all human beings are born highly creative and entrepreneurial. Unfortunately, our Industrial Revolution-designed education system unintentionally suppresses both traits. Kids learn to sit in rows, make no noise, and take notes. They are spoon-fed knowledge and forced to learn in lock-step. Both academically and socially, nonconformity is punished.” (p. 205)
2. Invoke the promise of the 1-on-1 tutor. Be sure to bring up the study by Benjamin Bloom and the “Two Sigma Problem:” If only every student could have their own individual tutor, they’d be performing two standard deviations above where they are now. Lament the fact that the methods of individual tutoring are completely incompatible with the caricature of schooling communicated previously.
“The resulting paper on what Bloom described as the two-sigma problem framed the benefits of one-on-one tutoring in a mastery learning context. In this paper, Bloom wrote that if a student works with a tutor to master a topic or skill, the student would gain a two-standard-deviation improvement - a massive upgrade that takes someone from the 50th percentile to roughly the 96th percentile.
He framed this a “problem,” however, because existing educational systems were unable to realistically scale group instruction this way, leaving the two-standard-deviation increase out of reach for most students.” (p. 14)
Create an Imagined Future
Your primary goal is to establish an imagined future with AI at the center and where the problems of schooling have been addressed. The power of AI as a technology is that we can imagine its ability to do just about anything. At the same time, it will help to lay out some concrete predictions and hypothetical ways that it might solve the thorny problems of schooling. The key is to give people just enough detail so that they can imagine the wonderful future that AI will bring forth.
You will need to give a general description of what AI is and what it does, but it’s best to avoid getting into highly detailed descriptions of the technology. If you keep things broad and general, it’s easier to speculate about what it might be able to do. Use analogies and metaphors instead of precise technical language. “Neural nets” is a great piece of terminology, as it lets you use the metaphor of the human brain.
“Large language models such as GPT-4, short for Generative Pre-trained Transformer, are essentially big - albeit digital - “word brains” trained on a colossal amount of information from books, articles, websites, and all sorts of written material.” (p. xvi)
***
“I now do the same thing when I face a tough problem while leading Khan Academy. I have faith that my brain, or someone else's, will come up with a creative solution by morning. What are our brains doing subconsciously while our consciousness waits for an answer? Clearly, when you "sleep on a problem," some part of your brain continues to work even though "you" aren't aware of it. Neurons activate, which then activate other neurons depending on the strength of the synapses between them. This happens trillions of times overnight, a process mechanically analogous to what happens in a large language model. When a plausible solution presents itself, the subconscious then surfaces it to the conscious as a flash of insight.” (p. 44)
2. Position AI as revolutionary, at least on par with if not even more powerful than previous revolutionary technologies (personal computers, the internet). Linking it to those technologies establishes its transformative potential.
“By bringing artificial intelligence into the classroom, educators could tackle entrenched problems in education that we had not been able to solve with existing technology and resources. Soon, students might be able to learn faster and retain more information than ever before, proving Al to be the ultimate learning tool for accelerating human intelligence and poten-tial. Al might hasten learning globally and even get us closer to realizing a world in which every person on earth had access to affordable world-class learning. This technology had the potential to revolutionize how we communicate, create, and consume information the same way that, twenty years earlier, we marveled at the educational possibilities of the internet, and twenty years before that, the personal computer, and twenty years before that, the calculator.” (p. xxvii-xxviii)
3. As you start telling stories about what AI has the power to do, language is key! Focus on the potential of AI to solve the most challenging problems of teaching and learning. It will do this by revolutionizing learning, making it more efficient, and thus vastly increasing access. The key educational use cases to highlight are the ways that AI can make learning personalized by acting as an intelligent tutor and assistant.
“When a hot new technology like GPT-4 comes out, it is important to not use it simply because it is "cool." We have to think about what important problems this technology might be able to help solve. Could it help close learning gaps or provide access to quality education regardless of geographic constraints, economic limitations, or social circumstances? Could it help meet the diverse needs and learning styles of each student where they are instead of the typical one-size-fits-all approach? Could it help address the limited availability of high-quality resources in education systems globally, especially in underserved or remote areas, or help with problems of student retention of learned material? Could it help save teachers time and support them better, preventing overwork and attrition in the process?” (p. 8)
***
“I also believe, however, that with proper care, transparency, and guardrails, we can mitigate the risks and develop assessments that are far richer, more accurate, and fairer than those that we have today. This will have positive consequences for the education system as a whole, reopening the aperture of what makes a quality education. By measuring skills long thought to be immeasurable, such as communication, creativity, and curiosity, it will naturally motivate the system to care a lot more about developing the whole person.” (p. 184)
***
“Mozart, Einstein, and da Vinci weren't just innately gifted. They had access to opportunities and resources that the bulk of humanity didn't have access to. Technology has generally lowered the cost of access to world-class tools and learning. Our mission of free, world-class education for anyone, anywhere would have seemed delusional without computers and the internet. Al is going to be the next technological wave that empowers future creatives in art and science. The Al, along with feeding us information on nearly any topic, becomes a companion in art, aiding in this practice. Not only does it allow students to produce more polished, finished works, but it can model the creative process with them. It can riff with students and ignite their curiosity, spark their imagination…” (p. 50)
***
“Consider reading comprehension: Today, students read a passage and then answer a few multiple-choice questions based on it. These questions might ask something about, say, the author's intent, followed by four choices. In the coming years, we will increasingly see assessments use generative Al to engage students about their views or the author's intent without the need for multiple choices. It will ask students to just write or speak their thoughts, and the Al will be able assess that response in a consistent way. Even better, it will be able to engage in a conversation with students about why they feel that way and discuss the evidence they are drawing on. The entire assessment will resemble a fluid, wide-ranging conversation with a thoughtful, empathetic, and fun mentor. Parts of it might involve role-playing or trying to work through a simulation. It wouldn't necessarily even have to be separate from learning. The same AI tutor that is there to help you would also build up evidence of what you know and don't know.” (p. 183)
***
“The benefits for teachers go beyond the planning and administrative work of writing, creating, and crafting lessons; grading papers; and communicating with parents. Als will be able to eventually facilitate classroom breakouts between students, give teachers realtime help and feedback on ways to better engage students, and let teachers know which students likely need their attention the most.” (p. 153)
4. Emphasize the ability for AI to do amazing things at scale. It will allow us to implement those time-consuming best practices that are just too difficult to pull off in the current context.
“For example, most educators intuitively recognize that allowing students to give free responses and engage in dialogue about a text would likely create deeper readers. They also see that pairing reading comprehension with writing is an ideal way to practice both. Unfortunately, these types of activities are hard to standardize and evaluate at scale.” (p. 38)
***
“For decades, future teachers have been taught that one of the best practices in education is differentiation and active learning. Differentiation is the idea that different students need different things. When Seldon and I talk about personalization, this is what we are referring to. Active learning is the notion that students don't learn best when sitting passively, pretending to pay attention to a lecture, but instead do so when they are actively engaged in discussion, games, projects, and problem-solving. Many new teachers aspire to this but quickly discover that these ideals are difficult to meet in a thirty-person class. The few teachers who can pull this off do so by spending countless hours tweaking lesson plans and creating personalized problem sets. Generative Al tutors support students by answering nuanced questions more fully. At the same time, perhaps the biggest game changer from the teacher perspective is that Al can now help craft lesson plans in minutes.” (p. 150)
***
“Educators have found short-term, narrow, and local solutions to some of these problems, but few provide the kind of equalizing force we want to be working toward in providing equal-access education globally. "The problem remains one of scaling, and that's where technology can be helpful," she says.” (p. 169)
5. As you paint this story of the future, acknowledge that the technology might not quite be there yet. But insist that the technology will only continue to improve, as technology always does. As it becomes more powerful and efficient, we will gain what is promised.
“The good news is that the computation will become cheaper and we will get better at using it more efficiently. These two trends should help bring the cost down by a factor of ten in the next few years. If we can reduce the costs by a factor of one hundred, which should happen in the next five to ten years, it will become comparable to the cost of using nongenerative web-based applications today.” (p. 175)
6. Science fiction can help your imagined future come to life.
“Few people may view the Star Trek universe through an economic lens, but doing so provides a window into a world that might soon be upon us. All of classical economics is based on the notion of scarcity-namely, that there isn't usually enough of anything to give everyone everything they want or need. Because of that, we use markets and pricing to allocate those goods, services, and resources to where they might result in the highest benefit. In Star Trek, however, there isn't much scarcity. Technology has allowed that society to replicate any food they want, transport themselves thousands of miles in the blink of an eye, communicate over light-years, and travel among the stars. All of humanity in that world has been fully educated so that they can participate in this bounty. Everyone is an explorer, researcher, engineer, artist, doctor, or counselor. Generative Al has the potential to allow many dimensions of our own society to be similarly low scarcity or highly abundant. Do we have the will to take us to the utopia of Star Trek?” (p. 219)
Present Some Actual Uses
Your audience might need some convincing that the hypothetical futures you’ve imagined are actually within reach. And so, you’ll need to present some examples of current use. Those examples don’t need to be quite as impressive as the future you’ve described. They just need to be proximal so that your audience can see how, as the technology inevitably improves, it will get where you’re saying it will go.
Make sure to give examples of AI playing the role of “intelligent tutor.” Showcase the ability of the AI to give explanations and information on demand.
“Saanvi, a ninth grader in India who attends Khan World School, a full-time online school we started with Arizona State University, was reading The Great Gatsby for class. Her teacher gave her a video assignment to discuss several key questions about the novel, and Saanvi came to one that stumped her. "In The Great Gatsby, why does Jay Gatsby keep looking at the green light?"
Uncertain of the answer, she did what many industrious students do when they find themselves stuck: she started running a bunch of web searches on the topic. She wasn't really satisfied with what she found, so she took it a step further.
She turned to Khanmigo... and asked to talk to Jay Gatsby himself
JAY GATSBY: Ah, splendid to meet you, old sport. I'm now Jay Gatsby, the enigmatic millionaire from F. Scott Fitzgerald's classic.
SAANVI: Mr. Gatsby, why do you keep staring at the green light?
JAY GATSBY: Ah, the green light. It is a symbol of my dreams and desires. You see, it is situated at the end of Daisy Buchanan's dock, across the bay from my mansion. I gaze at it longingly, as it represents my yearning for the past and my hope to reunite with Daisy, the love of my life.
"We just started talking," Saanvi tells me. "There I was, having a great conversation with Jay Gatsby. At one point I felt bad that I was taking up all of Mr. Gatsby's time, and so I apologized to him! And Gatsby said, 'Oh, no, I'm not really Jay Gatsby, I'm just an artificial intelligence simulation."
This is exactly the type of interaction that engages our learners and broadens their minds. Today, you can talk to Victor Franken-stein, Hester Prynne, or Odysseus.” (p. 36-37)
***
“I can ask the AI to explain a complex concept like string theory to me:
KHANMIGO: String theory is a theoretical framework in which the fundamental building blocks of the universe are not particles (like electrons or quarks), but rather tiny, vibrating strings. These strings can vibrate at different frequencies, and the various vibrational patterns correspond to different particles we observe in nature. The idea behind string theory is to unify the two major theories in physics: general relativity, which describes gravity and the large-scale structure of the universe, and quantum mechanics, which deals with the behavior of particles at the atomic and subatomic level.
But what if this is a little too complicated for me to understand? I can simply ask GPT-4 to explain it to me as though I were a sixth grader:
KHANMIGO: Of course! Imagine that everything in the universe is made up of tiny, super small pieces of string, kind of like pieces of spaghetti. These strings are so tiny that we can't see them, but they're there, and they're always vibrating or wiggling around. Now, depending on how these strings wiggle, they can create different particles, like the ones that make up everything we see and touch. It is like playing different notes on a guitar by plucking the strings in different ways.” (p. 72)
2. Give examples of how AI can generate lessons as well as complete other teaching tasks. Don’t worry if some of the use examples you provide are actually weak. The bar for establishing the competence of AI in education is not actually all that high. As long as it isn’t suggesting wrong or obviously harmful things, you’re good.
“Perhaps a teacher needs a tangible example to help explain the complexity of the economics that led to World War II. Either the teacher can spend an hour or two looking it up in the library or online or they can simply ask the Al teaching assistant for help. Within seconds of receiving a prompt, Khanmigo responds, "Here's a tangible example using a baseball analogy."
AI TEACHING ASSISTANT: Try this! Imagine a scenario where two baseball teams, Team A and Team B, are competing against each other. Both teams rely on equipment and resources to play the game effectively. Team A produces high-quality baseball bats, while Team B specializes in manufacturing baseball gloves. Before the game, the teams engage in trade to exchange bats and gloves, ensuring that each team has the necessary equipment. However, at some point, Team A decides to impose heavy tariffs on the gloves imported from Team B. This means that Team B's gloves become much more expensive for Team A to purchase. As a result, Team B's economy suffers.
By using this baseball analogy, a teacher can now easily illustrate how trade imbalances, protectionist policies, and economic conflicts disrupted relations between nations, ultimately leading to more significant geopolitical conflicts, much like the complex economics that contributed to World War II.
With each back-and-forth with the AI, the stronger and more precise the lesson becomes. A teacher can push the Al further still and ask it to help come up with a creative student project based on the lesson. The AI teaching assistant replies, "Here are some ideas: Create or find an interactive timeline that highlights key events, leaders, and battles of the war. This will give students a visual representation of the chronology and the global impact of the conflict. Or provide students with primary source materials like speeches, let-ters, and photographs to analyze and interpret. This will help them develop critical thinking skills and gain a deeper understanding of the historical context." (p. 152-153)
3. Acknowledge the technical limitations that currently exist. No doubt, people have heard about the possibility that AI produces inaccurate information and hallucinations, which could obviously be a problem with AI tutors, so get out in front of that. Point out that this is a solvable problem. The technology has gotten a lot better recently and the technology will continue to improve. Eventually, these problems will vanish entirely.
“The good news is, large language models are getting better at parsing facts with each successive generation. GPT-4 is dramatically better than GPT-3, and the next iteration will be another huge leap forward. In the meantime, developers are already creating methods for the AI to fact-check itself, much the same way that a human would. When we ask a generative Al for data or facts, it too can search the internet, assess which sites are most credible, and then make sense of the information to validate or refine the answer.” (p. 125)
Head Off Concerns
You will probably need to respond to a wider array of concerns beyond the hallucinations problem. Instead of waiting for others to bring up various issues, it’s best to get out in front of them. There are a couple of common tools that you can use to defuse all sorts of criticism.
All technologies, not just AI, have the potential to be misused. Whether technology is good or bad depends on how it is used - so it is up to us to use it well. On the development side, we’ll of course put in some guardrails. But if people are going to be bad actors, well that’s just what people are gonna do.
“Skeptics argue that at its most essential level it produces content based on patterns encoded in the Al model from training on existing texts. Do the limits of its training data curtail the scope of creative expressions or ideas?
Even back in the late 1970s, Coppola saw how advancements in technology not only did little to hinder creativity but also improved the creative process. In similar ways, large language models have the potential to do just that, by sparking new ideas, saving time on tedious tasks, and providing valuable revisions to work - as long as it is used right.” (p. 43)
2. Critics often set unreasonable standards for AI, expecting it to be perfect. But it doesn’t need to be perfect - it just needs to be better than what we’re currently doing. The status quo is filled with human weaknesses and biases. So when people say that AI has problems, maybe they’re right, but it’s still way better than what humans can do. And unlike humans, we can fix the technology and make it better.
“I say all this not to give generative Al a free pass. But it is important to keep the problems of the status quo in mind when deciding how to best implement new technology. For example, regulators in the EU have already classified leveraging Al for evaluating job applicants or student performance as high-risk. This is because AT may introduce bias into these sensitive processes. Yet I believe the measuring stick shouldn't be that the Al is perfectly bias free (which may be impossible to even define). Instead, we should measure its risk relative to the bias that is already involved in subjective processes such as hiring and assessment. Likewise, generative Al can produce incorrect facts, but is it better or worse than what is already out there? Is it more or less manipulatable by folks with bad intent?
“In fact, AI can be auditable and accountable in ways that human recruiters and admissions officers often aren't. We can train Al, for instance, not to favor candidates by race, religion, gender, or age, and these prompts can be fine-tuned across thousands or even millions of test cases. Once the model is performing within reasonable bounds on an evaluation set of test applications, the Al can get much closer to evaluating every application on actual merit, according to the same standard, without favoring one group over another.” (p. 124)
***
“Of course, using AI for assessment can rightfully make folks wary. What if the Al has biases that are not immediately apparent? What if it makes mistakes? I try to compare this kind of hypothetical to the status quo. Current assessments are written by thoughtful but fallible human beings with their own biases. We already know that by not leveraging AI, we are limiting ourselves to a much narrower type of assessment that arguably magnifies a bias toward prioritizing easy-to-measure skills over ones that are harder to measure but perhaps more important. Historically, when we have been able to administer richer assessments, like in PhD oral thesis defenses or job interviews, they are inconsistent and rife with more bias than any current standardized exam. Generative Al allows us to capture the best of both worlds: standardization and scale with richness and nuance. Because of its potential accessibility, stakeholders will have a far easier time trying it out and auditing it themselves.” (p. 184)
3. People often worry that AI will replace certain professions (like teachers) or human interactions. Insist that this just isn’t the case, because AI will be an enhancer, not a replacer.
“Seldon had us thinking that robots might take over teaching, but the reality is far cooler than the stuff of science fiction. The future of Al in education is about teaming up with technology to make education even better. In other words, Al is not here to steal the show from teachers, it's here to help teachers steal the show. It's the trusty wingman that tackles the boring stuff, sparks creativity, supercharges lessons, and helps educators craft unforgettable learning experiences that light up students' minds.” (p. 153)
***
“We've also shown that rather than somehow being a substitute for the teacher, videos can off-load pieces of a lecture, freeing up more time for personalized learning, hands-on activities, or classroom conversation. This arguably makes the teachers more valuable, not less. And now it was time to see if generative Al could do the same-support students and let teachers move up the value chain.”
“Imagine the local school district suddenly discovering hundreds of millions of dollars and using it to offer every teacher the support of three bright assistants for their classrooms. These assistants would help create lesson plans and rubrics, grade papers, write progress reports, riff with teachers, and support their students. Every teacher on the planet would jump at this opportunity. These assistants do not threaten teaching jobs, but they would actually make teaching jobs sustainable. They would make the work more joyous. Most important, theyd help accelerate learning outcomes for millions of students, making them more prepared for college, careers, and life.
“Unfortunately, society does not have the resources to give every teacher three human assistants. The good news, however, is that we are now able to offer educators the Al equivalent. In some ways, this will be even more powerful than what human teaching assistants can do. These AI assistants are available around the clock and work one-on-one with every student in the classroom. They can also proactively engage students and hold them softly accountable. Even more, they can do all the grunt work involved in teaching-writing rubrics, giving students feedback on their essays, and drafting student narrative progress reports for parents” (p. 149)
***
“There will always be a space for parents, as well as for living, breathing tutors, motivators, mentors, and teachers. People provide all sorts of benefits that the Al is not going to be able to replicate in our lifetimes. We find that when we mix large language models into this equation, artificial intelligence makes learning quicker and frees up time for parents to connect with their kids about all the other things that make up a well-rounded person. In the future, we may even have a version of this artificial intelligence at our dinner tables or on car rides to facilitate family interactions with games and conversations. Technology is a vector, helping parents work with their kids to see the wonder and joy in knowledge together. The technology is so broad and so inviting that when you are using it, you really feel as if you were on an Al-guided journey that's designed for parents and kids to explore the world together.” (p. 118)
4. Explain that you and other developers know about the risks and have put up appropriate guardrails. Especially when it comes to working with kids, explain that parents and schools can trust that the good AI developers will make responsible choices. There might be some bad actors out there, because there always are. But you are one of the good guys.
“Parents fear that the Al models are gathering data about their children, and it might be used in the future to violate their privacy in some way. Companies that develop the major models, like Google, OpenAl, and Microsoft, are aware of this and seem to be putting good guardrails in place to avoid giving away any sensitive information about an individual. It is, however, plausible that nefarious users will find ways to get around those guardrails. In this case, the best practice might be to ensure that the base models refrain from any training on personally identifiable information, especially on data from children.
At the same time, developers might want to use the data to fine-tune a model for specific applications. We could, for instance, train our version of GPT-4 for use by Khanmigo, but only Khan Academy would have access to that fine-tuned model. Everyone else's version of GPT-4 would not be aware of that data or training. Even here, the most responsible approach to fine-tuning is one that avoids using personally identifiable information that might inadvertently undermine a user's privacy.
Then there is data that the application leveraging the model might retain. Khanmigo saves student conversations so that it can make them available for parents and teachers. The platform also has a sense of "memory," where the tool can "recall" aspects of previous conversations. If you ask Khanmigo why you should care about a subject, it will likely ask you what you care about, to make a personal connection with the topic. If you respond with "football," it will remember that about you. We do not use that data for training the underlying model, but the application can use it to help customize things in the future for you. This can really help with oversight, safety, and personalization, but transparency is important, as is an option to edit or reset these inputs.
Nevertheless, real dangers with data still remain, but these dangers are the same ones we faced before the advent of generative Al. Where some might use personal data in healthy ways to measure impact or efficacy of a product, or to make the experience of using it more personalized, data also has very real value for ad targeting. Many organizations might initially desire to only use the data they collect for good, but when push comes to shove, and they need to improve their earnings for investors, there can be a strong temptation to walk into the gray area of data monetization. Some organizations, to save money, may not take appropriate precautions to safeguard data, leaving it vulnerable to hackers and data breaches. My advice to parents and educators is to ensure that any application, especially those that kids use, has the highest standards for what that data can be used for, and that it takes all reasonable precautions to protect that information.” (p. 129-130)
5. As much as you can, reframe things that might sound bad as actually good things. This especially holds for the case of surveillance. Don’t use that word, of course, because it has negative connotations. Instead describe how having an AI system closely monitor students will be beneficial. It will improve student safety and also promote transparency. If the AI is doing good things, then it cannot be called “surveillance,” because surveillance is bad.
“...educators are finding that these generative AI tools make our students far more skilled and efficient writers. They are also finding that, where producing essays was once seen as essential to helping students gain mastery in critical thinking and analytical and writing skills, the artificial intelligence provides students equal, and even better, opportunities to engage with a topic, gather and analyze information, and express their own ideas and arguments.” (p. 33)
***
“Now imagine if an AI tutor could "sit" next to students as they navigate the internet in general. Imagine if it were a browser plug-in. The same way that AI might help students better engage with online exercises or videos, it might also help them when they are browsing Wikipedia, Youtube, or the New York Times website. It might reformulate the news article they are reading closer to their grade level, potentially leaving out age-inappropriate details. While students are researching a paper, it might help zero in on material that actually addresses the issue they are investigating. It might also Socratically help a student engage with what they are reading or even provide context that the student needs to better understand the content.
“Having this functionality can also provide a valuable service for parents and teachers. As a parent, I want to maximize my children's constructive screen time (doing academic exercises online, coding, creating digital art, editing video, or writing a paper) and minimize their not-so-constructive time (stalking their friends on social media or watching other people play Roblox on YouTube). Even more, I want to ensure that the internet won't expose my kids to shady content. Ideally, I'd also get a report on what my children have been up to online. This would have seemed like a tall order only a few years ago, but it is very doable by the latest generation of AI.” (p. 136-137)
***
“In Khanmigo, we are developing the ability for a professor to create both an assignment and a grading rubric with the Al and then prompt students to complete tasks through the application. The professor can decide how much support the AI should provide. This could entail basic proctoring in which the application takes periodic snapshots of the paper as the student is writing it, or it could act like a full-fledged writing coach, riffing with the student on possible thesis topics, giving them feedback on their outline, and then providing initial feedback on the essay. This feedback could look at everything from grammar to vetting the quality of the references to estimating what grade the student will likely receive. Then, when the student is ready to submit the essay, the Al could send a report to the professor…
If the student completed the assignment using an essay-writing farm or ChatGPT and copied and pasted it into the assignment, Khanmigo could report it to the professor…
This type of transparency addresses many issues at once. It focuses on the process, helping the student while mitigating cheating. And even if a student enlists a friend (or Al) to try to engage with Khanmigo on their behalf, the final product would likely be inconsistent with the student's in-class, proctored writing samples. Teachers will receive a preliminary assessment, cutting down grading time, which allows them to devote more energy for themselves and their students. Last, but not least, students will get much more timely feedback and support to improve their writing.” (p. 162-163)
***
“There is even the possibility that Al agents can vouch for the student themselves, just like a teacher who knows the student well. Think about it this way: an Al platform like Khanmigo has been working with you for some period of time. Whether you have used it for a month or for many years of schooling, it knows your strengths and your passions and can plausibly render a dynamic picture of who you are. When it is time to apply to college, the Al can write a recommendation letter for you. The letter is standardized across every student who uses the platform, only it has different memories based on its experiences with each learner. Imagine if everyone in the country had the same teacher. This teacher would actually be a pretty good arbiter. If we wanted to take this to the extreme-and it is not clear that we do-the Al recommender could talk to the AI interviewer on the admissions side to see if there is a good fit.” (p. 193)