michael_nielsen

(I get occasional emails from people looking to cite the book, telling me some journal editor wants to know where "Determination Press" is located. I've so far resisted replying "Everywhere", and just demurely tell them "San Francisco".)

To sum up: open access makes material freely available to people who would otherwise never even hear about it. This amplifying effect is not small, it is enormous. And it applies in parts of the world woefully underserved by the existing publication system.

To finish: a couple of short, standalone essays within the book that you may enjoy. Written for a general audience. On whether there is a simple algorithm for intelligence: http://neuralnetworksanddeeplearning.com/sai.html 

On whether deep learning will soon lead to truly general artificial intelligence (scroll down a bit): http://neuralnetworksanddeeplearning.com/chap6.html#on_the_future_of_neural_networks …

Some additional calibration data: an editor at a major academic press tells me great sales figures for a similar technical textbook in a "hot" field are typically about 5,000-10,000 a year. So open access has a factor 200x or more here.

That said, deep learning is super hot, and I expect figures may be even higher there for the absolute best sellers. Even so, I've no doubt the benefit of being open access is ~100x.

Thanks for sharing these stats! Have you ever considered giving Zenodo DOIs to your publications? It's super simple with GitHub integration. Would help with citability and with looking like a legit publication (and archiving) ...