PODCAST: explore a YouTube video from Veritasium, explores the impact of AI on education. The speaker argues that while new technologies, including AI, are often touted as revolutionary for learning, historical patterns show they often become tools rather than fundamentally changing the educational system. He introduces Daniel Kahneman’s System 1 (fast, intuitive thinking) and System 2 (slow, effortful thinking) to explain how genuine learning, which builds long-term memory and “chunking” abilities, requires sustained, effortful engagement (System 2). The primary concern with AI is its potential to reduce this essential effortful practice, hindering true knowledge acquisition, and the talk concludes by emphasizing the irreplaceable value of human teachers and social interaction in fostering deep learning and accountability.
The expectation of a “revolution” in education has been a persistent theme throughout history, often accompanying the introduction of new technologies. However, as Derek Muller explains in the Veritasium video, this word might not mean what people think it means when applied to education.
Historically, various technologies have been proclaimed as the next big thing to revolutionize learning, but these predictions have largely failed to materialize:
- Motion Pictures (1920s): In 1922, Thomas Edison declared that the motion picture was “destined to revolutionize our educational system” and would “supplant largely if not entirely the use of textbooks”. He even claimed that education from a textbook was 2% efficient, while from a motion picture it was 98% efficient, though the origin of these numbers is unknown.
- Radio (1930s): In the 1930s, people believed radio would revolutionize education by allowing experts to “radio in to a thousand classrooms,” theoretically eliminating the need for many teachers and achieving an economy of scale.
- Television (1950s): Academics in the 1950s conducted studies on the impact of TV, comparing live lectures to those delivered via closed-circuit TV. They found “no significant difference” in learning outcomes, suggesting the experience was fundamentally the same.
- Interactive Computers (1980s): MIT academics in the 1980s thought interactive computers, particularly through programming activities like “turtle programming,” would revolutionize education by improving general reasoning skills. However, while students became proficient at programming the turtle, those skills “did not transfer to other sorts of reasoning,” and overall thinking skills did not improve.
- Video Discs (1990s): In the 1990s, the increasing use of video discs was also seen as promising to “revolutionize what will happen in the classroom of tomorrow”. Despite their presence, they did not lead to a revolution.
- Massive Open Online Courses (MOOCs) (around 13 years ago): More recently, MOOCs were “all the rage” and heralded as a major disruptor for higher education, with headlines suggesting they would “revolutionize” it. Yet, many sources later questioned or outright stated that the MOOC revolution “drifts off course” or “may not be as disruptive as some had imagined”.
The speaker highlights several reasons why these anticipated revolutions haven’t occurred:
- Inertia of Educational Institutions: One reason suggested is that educational institutions have “a lot of inertia” and are resistant to change, being “set in our ways”.
- Technological Overhype: Another factor is simply “technological overhype”. The speaker notes that the “magic” of early AI tutors is akin to how motion pictures seemed magical in their time.
- Misunderstanding the Core Problem of Education: Crucially, Muller argues that the technological hype stems from a fundamental misunderstanding: “the problem of education is not being able to get the information to the student.” Information has always been available, for instance, through books. The real issue lies elsewhere.
- The Social and Effortful Nature of Learning: The speaker emphasizes that education is fundamentally “a social activity” where “people care about other people”. He draws an analogy to physical fitness: the world is full of heavy objects and fields for exercise, but people aren’t inherently fit without guidance and accountability. Teachers, like personal trainers, are essential for holding students accountable, energizing them, and ensuring they put in the necessary “reps” of effortful practice. This social experience and the community of learners are what truly drive results.
When considering AI’s role, the speaker sees a positive potential in its ability to “provide timely feedback,” which is essential for learning any skill through repeated practice. An example of an AI tutor providing immediate feedback on identifying parts of a triangle demonstrates this potential.
However, the speaker’s biggest concern regarding AI in education is its “opportunity to reduce effortful practice”. He worries that if generative AI can perform tasks like writing essays or creating art for students, they might be prevented from engaging in the “painful effortful process” that is “the core process of learning”. This effortful engagement, using limited “system 2” (slow, methodical thinking) resources repeatedly to build “long-term memory,” is what allows for mastery and automaticity (“system 1” thinking). Without this practice, students may not develop the deep, connected knowledge structures in their brains necessary for insights and command of subjects.
In essence, the speaker concludes that these technologies, including AI, are unlikely to “revolutionize” education in the way often predicted. Instead, they are best viewed as “tools in the hands of educators”. The genuine learning process still relies on human connection, accountability, and sustained effort, which technologies can augment but not replace.
The “two systems of thought” framework, detailed in Daniel Kahneman’s book “Thinking, Fast and Slow,” provides a powerful lens through which to understand human learning and critique traditional educational approaches. This framework posits that our brains operate using two distinct systems: System 1 and System 2.
The Two Systems of Thought and Human Learning
- System 1: Fast, Automatic, and Intuitive
- Description: System 1 is a rapid-fire, automatic process that operates in the background, without conscious awareness. It quickly collects information from our senses, pulls out relevant details, and gets rid of extraneous stimuli. It’s associated with long-term memory, allowing it to work quickly and effectively.
- Function in Learning: When faced with a problem, System 1 serves up immediate answers, often based on patterns and associations stored in long-term memory. For example, in the “bat and ball” problem (where a bat and ball cost $1.10 together, and the bat costs $1.00 more than the ball, and you’re asked for the ball’s cost), System 1 quickly suggests “10 cents” because it’s the number that intuitively comes to mind, even though it’s incorrect. Similarly, when asked how long it takes for the Earth to go around the sun, System 1 might immediately suggest “a day”.
- Role of Long-Term Memory (Chunking): As we gain experience, practice, and interact with information, we develop long-term memory that allows us to “chunk” information. Chunking enables System 1 to see disparate bits of information as a single entity, significantly expanding our ability to process complex situations. This is why a chess master can recall 16 pieces after a 5-second glance at a chessboard in a game scenario, while a novice can only recall about four. This mastery means System 1 can solve problems at a glance, like Magnus Carlsen playing chess.
- Domain Specificity: The complex web of long-term memory that System 1 works with is highly specialized based on individual experiences. This means there is generally no general thinking or problem-solving skill; an expert in one field is not necessarily an expert in another unless the patterns transfer.
- System 2: Slow, Effortful, and Deliberate
- Description: System 2 is the conscious, effortful thinking process, often perceived as the “voice in your head”. It is slow, methodical, can work through step-by-step processes, catch mistakes, and think about thinking.
- Function in Learning: System 2 is crucial for processing new or complex information, engaging in reasoning, and deliberate problem-solving. When System 1’s quick answer is insufficient or incorrect, System 2 needs to “snap into action” to analyze and find the correct solution, as seen when the person in the video re-thought the Earth’s rotation time.
- Limitations: System 2 has a very limited working memory capacity, initially estimated as “seven plus or minus two” new items, but now revised to about four. When System 2 is engaged in hard thinking, there are physiological responses like increased heart rate, sweating, and pupil dilation, indicating cognitive load.
- Cognitive Load Categories:
- Intrinsic Cognitive Load: The inherent difficulty of the task itself (e.g., complex physics concepts).
- Extraneous Cognitive Load: Distractions or poor teaching methods that unnecessarily tax System 2’s limited resources (e.g., uncomfortable seats, illegible text, distracting accents).
- Germane Cognitive Load: The mental effort invested in deep processing and building understanding, like observing patterns or thinking about thinking, which is beneficial for learning.
Challenges to Traditional Educational Approaches
The two systems framework provides key insights into why traditional and overly simplistic educational “revolutions” have failed and what effective learning entails:
- Focus on Effortful Practice:
- Necessity of System 2 Engagement: Learning effectively requires careful and repeated use of System 2’s resources to store information in long-term memory, eventually allowing System 1 to perform tasks automatically. This process is often “painful” and “effortful”.
- Mastery is Key: Achieving mastery in basic skills (e.g., multiplication tables) means they become System 1 domains. This frees up System 2 to tackle more complex problems that build upon those foundational skills without being overloaded.
- Managing Cognitive Load:
- Eliminate Extraneous Load: Education should eliminate distractions (extraneous cognitive load) to allow System 2 to focus optimally on the learning task.
- Limit Intrinsic Load: Teachers must limit the amount of novel material introduced in a single lesson, keeping it “bite-sized”. Overloading working memory with too many new concepts at once leads to students being “lost”. This can be done by starting where students are, or for music, having students play songs they already know.
- Slow Down: Deliberate, slow practice allows System 2 to fully engage and build robust long-term memory structures.
- Increase Germane Load (Carefully): While not a general strategy, sometimes making a task slightly harder to read (e.g., using a difficult font) can force System 2 into action, leading to more correct answers, as seen in the Cognitive Reflection Test example. This is because the immediate System 1 answer isn’t readily available, prompting deeper thought.
- Critique of “Discovery Learning” and Insufficient Scaffolding:
- The framework challenges pure constructivism or “discovery learning” where students are expected to “figure it out” without adequate guidance. While students must be active constructors of their own knowledge (System 2 engagement), pulling away scaffolding too early overtaxes their limited working memory.
- Worked Example Effect and Fading Assistance: Effective teaching involves providing scaffolding and gradually fading assistance, starting with fully worked examples, then partially completed problems, and finally, problems for students to solve independently. This approach reduces intrinsic cognitive load by not taxing System 2 with too many simultaneous demands.
- Expert-Novice Divide: Professors, whose System 1 is highly developed in their domain, often struggle to see a problem from a student’s novice perspective, making it hard to provide appropriate scaffolding unless they consciously account for this divide.
- Concerns with AI and Learning:
- While AI can be beneficial for providing timely feedback and offering scaffolding (e.g., hints, practice questions), the biggest concern is that AI can reduce the need for effortful practice. If AI can write essays or create art, students might not engage in the repeated, difficult practice necessary to develop their own skills and build rich System 1 knowledge networks. This could prevent the development of deep insights and command over a domain.
- There’s a risk that students will use AI to “do the work without doing the work,” which is detrimental to learning. Therefore, educational settings may need to adapt by designating some tasks where AI aids are allowed and others where they are strictly prohibited, similar to how calculators are used in exams.
- The Social Aspect of Education:
- The speaker argues that education is fundamentally a social activity. The problem in education is not a lack of information (which is readily available in books or online). Instead, the challenge is getting students to engage in the effortful practice and hold them accountable.
- Teachers are likened to personal trainers who energize students, provide accountability, and foster a community of learners. This social connection and motivation are crucial for results and cannot be easily replicated by technology, which is why previous “revolutions” (motion pictures, radio, TV, computers, MOOCs) have not materialized as predicted. Technology, including AI, serves primarily as a tool in the hands of educators, rather than a revolutionary replacement for the core human and social elements of learning.
The “two systems of thought” framework, popularized by Daniel Kahneman in his book Thinking, Fast and Slow, describes how our brains operate using two distinct cognitive systems.
Here’s a breakdown of each system:
- System 1: Fast, Automatic, and Intuitive
- Description: System 1 is a rapid-fire, automatic process that works in the background, often without conscious awareness. It’s the “guy” that immediately suggests “10 cents” for the bat and ball problem, or “a day” for the Earth’s rotation around the sun. It’s described as being in a “library setting,” because it is associated with all of your long-term memory.
- Function: System 1 quickly collects information from your senses, pulls out relevant details, and gets rid of extraneous stimuli. When faced with a problem, it serves up immediate answers based on patterns and associations stored in long-term memory.
- Long-Term Memory and “Chunking”: This system’s speed and effectiveness come from its connection to long-term memory. As individuals gain experience, practice, and interact with information, they develop long-term memory that allows them to “chunk” information. Chunking enables System 1 to see disparate pieces of information as a single entity, significantly expanding the ability to process complex situations. For example, a chess grandmaster can recall 16 pieces after a 5-second glance at a chessboard in a game scenario, whereas a novice can only recall about four, because the grandmaster chunks the board into recognizable patterns. This mastery means System 1 can solve problems “at a glance”.
- Domain Specificity: The complex web of long-term memory that System 1 utilizes is highly specialized based on individual experiences. This implies that there is generally no general thinking or problem-solving skill; an expert in one field is not necessarily an expert in another unless the patterns transfer. For example, a chess master performs no better than a novice when pieces are arranged randomly on a board because the familiar patterns their System 1 relies on are absent.
- System 2: Slow, Effortful, and Deliberate
- Description: System 2 is the conscious, effortful thinking process, often perceived as the “voice in your head”. It is slow, methodical, and can work through step-by-step processes. It’s the system that “snaps into action” when the quick, intuitive answer from System 1 is insufficient or incorrect, such as when the person in the video re-thought the Earth’s rotation time to a year.
- Function: System 2 is crucial for processing new or complex information, engaging in reasoning, and deliberate problem-solving. It has the ability to catch mistakes and think about thinking. For example, multiplying 13 by 17 requires System 2 to engage in a series of steps.
- Limitations and Cognitive Load: System 2 has a very limited working memory capacity, initially estimated as “seven plus or minus two” new items, but now revised to about four. When System 2 is engaged in hard thinking, there are physiological responses like an increased heart rate, sweating, and noticeably, pupil dilation, indicating significant cognitive load.
- Categories of Cognitive Load:
- Intrinsic Cognitive Load: The inherent difficulty of the task itself (e.g., complex physics concepts).
- Extraneous Cognitive Load: Distractions or poor teaching methods that unnecessarily tax System 2’s limited resources (e.g., uncomfortable seats, illegible text, distracting accents).
- Germane Cognitive Load: The mental effort invested in deep processing and building understanding, like observing patterns or thinking about thinking, which is beneficial for learning.
Interaction Between Systems: The goal is to optimize cognitive effort by allowing System 1 to handle everything it can, and only handing off tasks to System 2 when necessary. However, System 2 can be “lazy” and accept System 1’s immediate answers without checking them. Learning effectively often requires careful and repeated engagement of System 2’s resources to store information in long-term memory, ultimately allowing System 1 to perform tasks automatically and efficiently.
Cognitive load is a measure of how much mental effort you are investing in something. It describes the demands placed on an individual’s limited working memory capacity (System 2) during a cognitive task. When System 2 is highly engaged in thinking, there are physiological responses such as a faster heart rate, increased sweating, and noticeably, pupil dilation.
The concept of cognitive load can be broken down into three main categories:
- Intrinsic Cognitive Load: This refers to the inherent difficulty of the task itself. For example, teaching complex physics concepts like F=MA to a new student has a large intrinsic cognitive load because each component is new and sophisticated. It’s about the number of novel concepts or elements that need to be processed simultaneously. Strategies to limit intrinsic cognitive load include starting where students are, keeping work “bite-sized,” and slowing things down. For instance, in music, getting students to play songs they already know can limit intrinsic load, allowing them to focus on new concepts like reading music without being overwhelmed by rhythm and notation at the same time.
- Extraneous Cognitive Load: This type of load is caused by distractions or poor teaching methods that unnecessarily tax System 2’s limited resources. Examples include uncomfortable seats, illegible text, distracting accents, or someone chewing next to you. The goal in education is to eliminate extraneous cognitive load to allow students to focus their mental effort more effectively. Subtitles can sometimes help, especially with accents, by reducing this load.
- Germane Cognitive Load: This is the mental effort that is beneficial for learning, specifically the effort invested in deep processing and building understanding. It involves engaging System 2 to do “thinking about thinking,” observing patterns, or noticing helpful information for future use. This type of cognitive load is considered “great” if it can be achieved.
How Cognitive Load Relates to the Two Systems of Thought and Learning:
- System 2’s Limitation: System 2, our conscious, effortful thinking system, has a very limited working memory, initially thought to handle about seven new items, but later revised down to about four. When System 2 is overloaded, it struggles to process information, and learning becomes very difficult.
- Expert vs. Novice: For an expert, like a physics professor, complex concepts may seem perfectly clear because their System 1 is fully developed and can “chunk” information, reducing the perceived intrinsic cognitive load. However, for a novice, the same problem will appear complex and demand significant System 2 resources. This is known as the “expert-novice divide”.
- Mastery and Automation: The goal of effective learning is to use System 2’s resources carefully and repeatedly to store information in long-term memory. This process allows for “chunking”—seeing disparate pieces of information as a single entity—and eventually enables System 1 to perform tasks automatically and efficiently. When a skill is mastered and becomes a System 1 domain (like basic multiplication tables), it doesn’t overload System 2 when used in more complex problems.
- Danger of Discovery Learning: Some educational approaches, like certain implementations of constructivism or “discovery learning,” can be dangerous because they might pull away scaffolding too early. This forces students to figure out problems entirely on their own, which can overload their limited System 2 resources, especially in complex domains. Providing assistance and gradually reducing it (like in worked examples) is a way to reduce intrinsic cognitive load and support learning.
- Increasing Domain Load to Activate System 2: Interestingly, in certain situations, increasing the cognitive load can be beneficial. For example, when a cognitive reflection test (like the bat and ball problem) was printed in a hard-to-read font, the error rate dropped significantly. The difficulty in reading kicked System 2 into action, forcing people to think more deliberately rather than relying on the immediate, incorrect System 1 answer. While not a general teaching strategy, it illustrates how engaging System 2 through appropriate challenges can lead to better outcomes.