3 ms·

I feel like the article is severely overestimating the state of the art and the current capabilities of the "big 5". All of them are working on the most basic

I feel like the article is severely overestimating the state of the art and the current capabilities of the "big 5". All of them are working on the most basic fundamental components that are prerequisites to a Primer and every prerequisite is still pretty far off: no one can currently build a useful conversational agent (anything beyond query-and-response of Alexa), no one can really extract useful semantic meaning from text comments with surrounding context, no one can construct good automatic customized lessons for 5th graders based on feedback loops. There is no meaningful version of a primer that can exist without all of those things just being solved problems. Given the inability to meaningfully execute in any of those areas, the challenges listed in the article seem laughably meaningless to me: the problem of "they won't use it 24/7, so the system wouldn't be able to learn everything it needs to" is meaningless if we can't make something learn something from 24/7 use. Similarly, a system diagram showing that it would be a book with sensors and smarts doesn't really seem to be providing any insight at all.

inimino 9y ago
I think you're missing the point, which is to develop something that can help people learn better today and be a starting point for something like the primer, not to build the primer today. Obviously the technology for the latter doesn't exist, which is why some creativity is needed in approaching the problem. The state of the art in guided learning is still textbooks and lectures, and most applications of technology beyond those have been failures, but that doesn't mean there are no wins to be had with current technology.
- esrauch 9y ago
  I think it is the exact opposite of that. There is some version that can be done with current technology, but the challenges for that version are fundamentally much much more basic than what is discussed here: the article is basically jumping ahead imagining we had 90% of the technology of the primer solved and what problems you might face in that hypothetical world where the remaining constraints being "people don't talk to it enough, otherwise it would work" The article mentions several times the concerns about getting enough user specific data and that it can't be easily bootstrapped by a smaller company that isn't the "big 5" because others don't have enough data. Except the big 5 can't solve related and easier problems yet with the data they have and a ton of engineering effort invested. The article ponders whether a primer for the hardest topics (educated adults) could have enough usage while we can't even do "make something that usefully, intelligently adapts to teach users even trivial topics or targeting a single grade of very young children", or "something that can hold a conversation better than a toddler".
  - inimino 9y ago
    I think the author knows and is assuming a lot of what you're pointing out. The technology for the Primer as such isn't there and won't be until we have general AI. However, we could build something today with current tech, like a Kindle that teaches you to read, or teaches you calculus, and it wouldn't be hard to make a compelling product. There are many possible starting points and the technology isn't really the problem if the goal is just to make a start. The real problem is much deeper and the key point is that the Primer was subversive. How do you organize a consumer product company around a subversive product? You don't. So how do you fund such a project? These are much bigger than the technical problems. There's a reason why Stephenson had the Primer commissioned by an infinitely rich eccentric and built by a rebel engineer. What I suspect the author is hinting at by talking about data isn't that the data is hard to get, it is that for any company in our society with access to that level of resources to start making machines to mold young minds would be a pretty short road to a pretty boring dystopia.

inimino 9y ago

I think you're missing the point, which is to develop something that can help people learn better today and be a starting point for something like the primer, not to build the primer today. Obviously the technology for the latter doesn't exist, which is why some creativity is needed in approaching the problem. The state of the art in guided learning is still textbooks and lectures, and most applications of technology beyond those have been failures, but that doesn't mean there are no wins to be had with current technology.

esrauch 9y ago
I think it is the exact opposite of that. There is some version that can be done with current technology, but the challenges for that version are fundamentally much much more basic than what is discussed here: the article is basically jumping ahead imagining we had 90% of the technology of the primer solved and what problems you might face in that hypothetical world where the remaining constraints being "people don't talk to it enough, otherwise it would work" The article mentions several times the concerns about getting enough user specific data and that it can't be easily bootstrapped by a smaller company that isn't the "big 5" because others don't have enough data. Except the big 5 can't solve related and easier problems yet with the data they have and a ton of engineering effort invested. The article ponders whether a primer for the hardest topics (educated adults) could have enough usage while we can't even do "make something that usefully, intelligently adapts to teach users even trivial topics or targeting a single grade of very young children", or "something that can hold a conversation better than a toddler".
- inimino 9y ago
  I think the author knows and is assuming a lot of what you're pointing out. The technology for the Primer as such isn't there and won't be until we have general AI. However, we could build something today with current tech, like a Kindle that teaches you to read, or teaches you calculus, and it wouldn't be hard to make a compelling product. There are many possible starting points and the technology isn't really the problem if the goal is just to make a start. The real problem is much deeper and the key point is that the Primer was subversive. How do you organize a consumer product company around a subversive product? You don't. So how do you fund such a project? These are much bigger than the technical problems. There's a reason why Stephenson had the Primer commissioned by an infinitely rich eccentric and built by a rebel engineer. What I suspect the author is hinting at by talking about data isn't that the data is hard to get, it is that for any company in our society with access to that level of resources to start making machines to mold young minds would be a pretty short road to a pretty boring dystopia.