An academic friend of mine once said something about how the knowledge she gained during her studies was emancipatory.
This feels like it could almost be a truism, but I would probably add 2 things that knowledge acquisition needs to successfully be "emancipatory".
-->coherent body of knowledge (clear how different pieces of knowledge link together) -->practical body of knowledge (applicable to the real world as you know it)
I think the AI safety/EA communities does this exceptionally well (and maybe has a significant head-start considering the type of people that are drawn to this space(?)) while it is grossly under-looked in classical educational spaces and, I'd argue, especially STEM spaces (my bias speaking).
What should happen is: a student gets a problem and works backwards. All knowledge gained will be part of a coherent body (the problem at hand) as well as practical (the problem at hand exists in the real world).
Most classical curricula seem to be built up this way, except for the crucial fact that they miss the problem. I'd love the problem to be formulated more along the lines of:
-> Lots of people get sick -> Some of these diseases are transmissible -> You can use epidemiology to study the transmissible diseases -> Modelling is an important part of epidemiology -> Bayesian statistics are an important of this modelling
That's why you're sitting through a class of incomprehensible numbers and that, to me, makes sense.
But way too often the problem is given in an abstracted form and the pipeline ends up looking like:
-> Lots of people get sick -> I made the decision to study biomedical sciences 3 years ago ..... -> Bayesian statistics (???)
Is this entire post just me lamenting not having a better mathematical understanding than I do? Maybe, but I think my larger point is that the AI safety space has been the single best space I have been in for me to acquire this type of knowledge. I recently came across a post arguing field building might not be the most important thing to focus on in very short timelines and I would probably agree.
However, if you see field building as having the obvious goal of increasing the number of FTE's working on AI safety, but also as a way to get more people to realize that you can really just do stuff (and I'd argue the latter happens way before the former), it's still useful. Assuming RSI does happen in the next 12 months and that it doesn't immediately kill us all, it would probably be good to have more people that believe they can do things during a post-stability era instead of being stuck in useless, but stable, corporate software jobs.
An academic friend of mine once said something about how the knowledge she gained during her studies was emancipatory.
This feels like it could almost be a truism, but I would probably add 2 things that knowledge acquisition needs to successfully be "emancipatory".
-->coherent body of knowledge (clear how different pieces of knowledge link together)
-->practical body of knowledge (applicable to the real world as you know it)
I think the AI safety/EA communities does this exceptionally well (and maybe has a significant head-start considering the type of people that are drawn to this space(?)) while it is grossly under-looked in classical educational spaces and, I'd argue, especially STEM spaces (my bias speaking).
What should happen is: a student gets a problem and works backwards. All knowledge gained will be part of a coherent body (the problem at hand) as well as practical (the problem at hand exists in the real world).
Most classical curricula seem to be built up this way, except for the crucial fact that they miss the problem. I'd love the problem to be formulated more along the lines of:
-> Lots of people get sick
-> Some of these diseases are transmissible
-> You can use epidemiology to study the transmissible diseases
-> Modelling is an important part of epidemiology
-> Bayesian statistics are an important of this modelling
That's why you're sitting through a class of incomprehensible numbers and that, to me, makes sense.
But way too often the problem is given in an abstracted form and the pipeline ends up looking like:
-> Lots of people get sick
-> I made the decision to study biomedical sciences 3 years ago
.....
-> Bayesian statistics (???)
Is this entire post just me lamenting not having a better mathematical understanding than I do? Maybe, but I think my larger point is that the AI safety space has been the single best space I have been in for me to acquire this type of knowledge. I recently came across a post arguing field building might not be the most important thing to focus on in very short timelines and I would probably agree.
However, if you see field building as having the obvious goal of increasing the number of FTE's working on AI safety, but also as a way to get more people to realize that you can really just do stuff (and I'd argue the latter happens way before the former), it's still useful. Assuming RSI does happen in the next 12 months and that it doesn't immediately kill us all, it would probably be good to have more people that believe they can do things during a post-stability era instead of being stuck in useless, but stable, corporate software jobs.