If you're seeing this message, it means we're having trouble loading external resources on our website.

If you're behind a web filter, please make sure that the domains *.kastatic.org and *.kasandbox.org are unblocked.

Main content

Welcome to the course!

An overview of Khan Academy's new AI for education offering.

What is this course?

There are so many exciting developments in the field of artificial intelligence (AI), and the future remains unwritten.
This course is somewhat unwritten, too!
This course is a work-in-progress!
We've partnered with other thought leaders in educational technology—including Code.org, Common Sense Education, aiEDU, and Professors Ethan and Lilach Mollick—to develop a set of resources to help the world better understand some of the ways that generative artificial intelligence can be used to improve human intelligence and learning outcomes.
As our understanding evolves, so will the course.
We'll learn how some educators are beginning to use generative AI technology with their students, we'll answer some questions, make some suggestions, and we'll attempt to explore two overarching questions:
What potential do generative artificial intelligence tools like large language models (LLMs) hold to transform worldwide education for the benefit of humanity?
How can we maximize the benefits of this emerging technology while minimizing its risks?

AI's potential in education

We believe AI has the potential to usher in a new age of education and revolutionize the way we learn. We believe that by working with AI, humans will be able to more quickly generate ideas to solve problems, brainstorm and create engaging learning experiences, and find exactly what they are looking for on their learning journeys.

Risks and responsibilities

In this course, we also will discuss some of the risks inherent in working with this rapidly evolving technology. For example, the text generated from LLMs should not be taken at face value, but should rather be independently validated whenever possible. This means that human judgment will remain as important as ever, if not more so!

Evolving our understanding

AI-powered tools are exciting, but they can also be alarming and strange. As we learn more from our learners, teachers, and parents about what they would like to better understand about AI in education, we’ll continue to build out this course so that you can use AI-powered tools with confidence.

Embracing change

New developments are calling into question what a formal education should include. By becoming more fluent in the variety of ways AI can be used ethically and responsibly, we can more confidently embrace the change it represents.
We are all still learning about these technologies, but join us—let's learn about them together!

Want to join the conversation?

  • blobby green style avatar for user lfey
    Was the course (or the above text) written by AI?
    (59 votes)
    • piceratops ultimate style avatar for user Dave Travis (Khan Academy)
      Nope. That's 100% genuine human content right there. :-) One of the things you'll learn in this course is that it generally is pretty inefficient to ask an LLM to create content about stuff that its training data wouldn't help it to know, like how Khan Academy thinks about AI, or about what is in a course that it has never seen. It would just make stuff up, and while it might sound good, a lot of it would be inaccurate. This course is human-generated!
      (87 votes)
  • orange juice squid orange style avatar for user jakob
    what about people using chat gpt for their ela assignments,
    does sal support that?
    (18 votes)
    • piceratops ultimate style avatar for user Dave Travis (Khan Academy)
      From the author:Hi Squid! No, we don't support that—unless it is part of the assignment and your teacher knows about it! But we are encouraging teachers and school administrators to have conversations about how to adapt their academic honesty policies to cover the use of AI, and to discuss ways that the technology can be used to drive engagement and learning. Read the rest of the articles in this course, and you'll see what I mean :-).
      (39 votes)
  • aqualine ultimate style avatar for user knot126
    I know this probably isn't the best place to ask, but it seems like a lot of the things that Khan Academy has been advocating for (like mastery learning) would actually go much better with AI.

    For example, with mastery learning, it is much harder to just have an AI do your work for you because you expect the work to be very, very good. And, if you're needing to master the content anyways (because it's not just about how good your work is - it's also about understanding) then you will probably natrually care more and thus won't want to cheat as much. (Because if you are passionate, or at least care, cheating is really cheating - you understand the impact on an emotional level.)

    Are there any plans to look into this link more in the research and communication around AI at KA? While I think KA is (or will be?) on the right side of history in terms of AI, I do worry that other important things might take a back seat when in reality it would be more beneficial to focus on using them together.
    (12 votes)
  • marcimus orange style avatar for user Sasha Nibbs
    How is ai programmed if it's always learning? Can you give an example of the code? I mean you give it information but how does it use it to evolve to that information, if it has no solid structure? Can it even rewrite its code, for adaption?
    (3 votes)
    • stelly blue style avatar for user Amitoj Shahi
      AI programming, especially in the context of machine learning and deep learning, is fundamentally different from traditional programming. Instead of explicitly defining every rule and decision-making process, AI systems are trained to learn from data and improve their performance over time.

      Let's dive into the process:

      1. *Data Collection*: The first step is to collect relevant data that the AI model can learn from. For example, if we want to create an AI model that can classify images of cats and dogs, we would gather a large dataset of labeled images containing cats and dogs.

      2. *Data Preprocessing*: Raw data is often noisy and unstructured, so it needs to be preprocessed to make it suitable for learning. This step involves tasks like data cleaning, normalization, and feature extraction.

      3. *Model Architecture*: AI models have architectures that define how they process data and make predictions. For example, in a neural network, the architecture includes layers of interconnected nodes where data is processed.

      4. *Learning/Training*: This is where the AI model starts to "learn." During the training process, the model is presented with the preprocessed data, and it adjusts its internal parameters iteratively to minimize the difference between its predictions and the actual labels in the training data.

      5. *Backpropagation*: In deep learning, a popular technique called backpropagation is used to update the model's parameters based on the difference between its predictions and the true labels. This process is performed iteratively on batches of data until the model's performance converges to an acceptable level.

      6. *Evaluation and Tuning*: After training, the model is evaluated on a separate set of data called the validation set to see how well it performs on unseen examples. The model's hyperparameters (parameters that determine the model's structure, like the number of layers or learning rate) may be adjusted to optimize performance.

      7. *Deployment*: Once the AI model achieves satisfactory performance, it can be deployed to make predictions on new, unseen data.

      Now, to answer your question about AI evolving and potentially rewriting its own code:

      AI models, especially those based on deep learning, can indeed adapt and evolve in response to new data and changing circumstances. However, it's important to distinguish between "rewriting code" in the traditional sense and model adaptation.

      The actual code that constitutes the AI model is not modified during the learning process. The model's architecture remains fixed once defined by the programmer/researcher. However, the model's internal parameters are adjusted during training, allowing it to adapt to the data it's exposed to.

      Once the model is deployed, it can continue to learn and improve over time by being fed more data or by implementing techniques like online learning. This ongoing learning process helps the AI system to adapt to new patterns and changes in the underlying data distribution.

      In summary, AI is programmed through the design of the model architecture and the implementation of the learning algorithm, but the model's parameters change during training to adapt to the provided data. After deployment, the model can further improve its performance through continuous learning on new data.
      (27 votes)
  • area 52 green style avatar for user Karl Hansen
    Can Artificial Intelligence understand "gaps" in a learners understanding and pathway, and help to remediate those learning "gaps". For instance, a student who struggles understanding number place value could recieve intervention guidance to understand what ones, tens and hundred values mean?
    (7 votes)
    • piceratops ultimate style avatar for user Dave Travis (Khan Academy)
      From the author:Hi Karl! Adaptive diagnostic assessments are indeed one use case that we are exploring with our research with Khanmigo. At the moment, the best way to "find your learning edge" and remediate gaps is to take "course challenges" in the courses you're considering, and bumping up against skills you don't know yet and getting those questions wrong. The Mastery system will flag those skills you missed in your practice experience.
      (9 votes)
  • blobby green style avatar for user Sarb
    Is there any way that AI can phisicly harm us humans why and why not and does humanity need to be worried about this rapid technology improving every day and one day even be in a phisical form?
    (4 votes)
    Default Khan Academy avatar avatar for user
    • starky tree style avatar for user Limbo
      Not at this scale, not it the way you're thinking. AI today, is nowhere near advanced enough to harm humans physically, but you could use AI to commit crime. It all depends on how we use and restrict it. Eventually AI will become much more advanced, to the point where we will have to make sure it doesn't become too much more intelligent than us, but even then, the chances of AI becoming physically dangerous is extremely low.
      (2 votes)
  • hopper cool style avatar for user Mrs Pieciak
    Could AI be used to begin a project such as creating a step-by-step plan?
    (3 votes)
    • female robot grace style avatar for user Jme
      Perhaps. You'd have to feed it details about what you want from your project, and there's no guarantee that it'll even be correct, but it could help you out if you're stuck.
      (5 votes)
  • blobby green style avatar for user coolbros61720
    what is the circumfrerence of the earth
    (3 votes)
    Default Khan Academy avatar avatar for user
  • female robot grace style avatar for user Kendean
    Quite a few people say that there are some downsides to AI but frankly I don't see any. Can someone tell me a few downsides?
    (1 vote)
    • leafers ultimate style avatar for user Makoyaki
      As AI advances, many people are worried that AI might take over their jobs. For instance, an AI that can easily write functional code could remove the need for real, human programmers. As things become more automated, we lose the need for human involvement. Hope this helps :)
      (6 votes)
  • leaf blue style avatar for user The Will
    What is A.I.?
    (3 votes)
    • female robot grace style avatar for user Jme
      Artificial intelligence! There are many, *many* different forms, but for the most part, you input some training data (input examples and the correct/possible correct outputs), and after being trained, the computer will try to ""guess"" what the answer should be for a given input. I highly reccomend this series by 3blue1brown (don't click the link, copy the URL): https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi
      (2 votes)