3 ms·
(I work at Khan Academy.) > The Khan Academy developers have devoted a lot of thought to hope to computerize problems rather than exercises, and they are still
by spicyj 12y ago
(I work at Khan Academy.)
> The Khan Academy developers have devoted a lot of thought to hope to computerize problems rather than exercises, and they are still working on building new problem types for the mathematics lessons. I'm actually quite impressed with some of the problem types that show up in more advanced mathematics lessons on Khan Academy that go beyond the eighth grade level implied by current United States curriculum standards.
Thank you.
> Common Core, and good mathematics curricular around the world, expect reasoning and explanation from pupils in mathematics lessons already at elementary school age. That currently is indeed difficult to automate.
This is definitely true. By no means is Khan Academy a complete solution for mathematics instruction, nor is it likely to become one in the next few years. Our site should ideally be used alongside other materials that test reasoning and explanation in a non-automated way, as well as projects that allow students to apply their knowledge. We have early experiments that take a much more interactive approach to teaching certain concepts and encourage more experimentation, but we're probably years away from developing them into a shipping product.
> The author's comments on development of Smarter Balanced Assessment (SBC) as a new platform of online mathematics instruction is of course especially interesting. I will be curious to see how this program continues to develop, and I would be glad to hear from developers of that program in this thread too.
If I'm not mistaken, many of the SBAC problems (especially free-response ones) are hand-graded, which can work if you charge for each test taken, but can't work for us when we have millions of problems done every day, all graded automatically. That said, we're exploring strategies like peer grading in other subject areas (particularly our programming curriculum) and would be excited to bring them to math as well.
- zanny 12y ago> that test reasoning and explanation in a non-automated way > early experiments that take a much more interactive approach > we're exploring strategies like peer grading Are you guys working on any genetic / learning algorithm implementations of AI sufficient enough to evaluate responses to arbitrary prompts? It sounds both like something that should be feasible with modern tech, especially when time to response can be large (a student can wait hours for an answer to a complex essay response) and the scale can be arbitrary (as many server clusters as "its worth" yet also feasible (we can see similar scale literary processing in other spaces). I can imagine such an algorithmic implementation where you take some common query given to students across hundreds of schools, with all the teacher graded responses, and maybe after a few years it can get within an accuracy confidence threshold to start grading the prompt, and after another few years providing critique on that grade. That would be the most revolutionary educational tech since the Internet itself, even if it were pigeonholed in several years studying the same single problem for now. Sounds awesome either way. You can even use that peer grading tech to build up a library of problem / solutions to build your causal relations table from. I just hope you guys have someone in a back room working on this, because to me it seems as real world usable computational linguistics happen more often, there arise a ton of useful applications even in the minimum case.