Hacker News Reader

Comment by bertil | Hacker News Reader

I feel like we have not cracked the code on how to use software to automate teaching yet. Not sure why: NLP, complex representation of understanding, dynamic edition of video content, monitoring someone’s attention… All those are there and have been for a while.

View on HN

I feel we barely have a notion on how to teach other humans how to teach, let alone a computer.

As a former teacher, I find this very amusing. There's a lot more to it than uploading raw info into receptive brains. As a current software engineer, I think it's probably a lot more likely we'll automate parts of my current career.

Also a teacher-turned-dev, and I completely agree. Teaching is such a labor-intensive, poorly-understood process that the idea of effectively automating it anytime soon is ridiculous. We still don't really have any consistent metric that measures what a "good" teacher is - if you can't even kinda measure performance in an accurate and consistent way, how on earth is the process going to be automated effectively?

Pedagogical understanding is pretty good (given enough effort in a particular field).

But NLP's accuracy isn't high enough for anything more than basic tasks.

It's just not good enough at understanding mistakes, or giving well calibrated probabilities of complex meanings given a context.

(I work in NLP professionally, and I've worked in the education space previously).

NLP is generally garbage. Everything you described sounds really amazing but is twenty years away in conjunction

Yes, all those parts are there, and some of them have been around for decades, but adding these parts and building a teaching AI will be more complex, and I really think that if we are ready to build it we won't need to train humans that much anymore.

We don’t yet have a substitute for asking complex questions about the source material.

Reply on news.ycombinator.com