Students don’t have to (and cannot) understand backpropagation and stochastic gradient descent unless they have learned Math in high school. But they should understand what backpropagation and stochastic gradient descent are doing —— “trying to adjust weights to minimize the errors”. However, the teachers should understand how backpropagation and stochastic gradient descent work. Otherwise they would not be able to simplify the concepts as such for students, and would not feel uncomfortable in teaching students. To understand backpropagation and stochastic gradient descent, one must understand the math first. Math is a prerequisite to understand algorithms but it does not mean all Mathematicians understand AI. Also, it does not mean AI professionals are good at teaching k12 students. In fact,  It is the educators who are good at teaching K12 students.

Students don’t need to understand binary representations or how circuits work in computer in order to understand AI, those are the concepts for understanding computer. AI is just a highly math-intensive computer software. A common mistake is to think that many computer related or programming are part of AI. Python is a tool for AI but Python is not part of AI; Most robots are not doing AI (so when you teach students on robotic programming, you are not teaching AI). Automatic control is not AI...

For educators, The good thing is that, if you understand how the “cat recognition” works, you will feel much easy to understand many other algorithms if an AI expert tells you the similarities.

Learning from Al Professionals is the quickest way (not the only way) for educators to understand AI, and after that, educators will be able to think about how to transfer the knowledge to students in a good way for different level of students.

Best,

Frank

Sent from my iPhone

> On Jul 28, 2019, at 12:56 AM, Ken Kahn <xxxxxx@gmail.com> wrote:
>
> backpropagation