This is an interesting article. It made me think of the socio-political consequences of the move towards neural networks.
Many of us, children of the personal computer revolution, were attracted to computers for its empowering and democratizing effects. Just with a computer you could build anything! You are as powerful as any of "them"! The proprietary software model was developed to monetize end consumers by enchaining them, but the free software model and the fertile communities of the early Internet showed that the model of personal computation could not simply be displaced by legalese.
But if we take the author's position on the impact of deep networks, the model of computation is changing again.
If teaching deep networks is the new way to write useful programs, our brains and personal computers are not enough... they are obsolete. We now need clusters and, most importantly, massive access to data! This gives extraordinary leverage to the hoarders of data and servers of the Internet, the googles and the facebooks.
I'd like to think that maybe we are just at the beginning, the early "mainframe neural networks" era. That we just have to wait until there is enough technology and new markets discovered to build the "personal neural network". That consensual, open and distributed ways of sharing data will emerge and that new massively parallel computers will become affordable by the masses. That the models of neural networks will become well divulged and simplified and kids will be able to program them with their "Neural Basic"...
But at the moment the prospects don't look good. The Internet is becoming more and more centralized, personal computers harder and harder to program and there is a general "war on general computation" [1]. Even universities seem to bedisplaced by the Internet Lords in driving neural network development...
Maybe it is already a good time to start thinking about what "Libre Neural Networks" look like. And how can we get there.
Many of us, children of the personal computer revolution, were attracted to computers for its empowering and democratizing effects. Just with a computer you could build anything! You are as powerful as any of "them"! The proprietary software model was developed to monetize end consumers by enchaining them, but the free software model and the fertile communities of the early Internet showed that the model of personal computation could not simply be displaced by legalese.
But if we take the author's position on the impact of deep networks, the model of computation is changing again.
If teaching deep networks is the new way to write useful programs, our brains and personal computers are not enough... they are obsolete. We now need clusters and, most importantly, massive access to data! This gives extraordinary leverage to the hoarders of data and servers of the Internet, the googles and the facebooks.
I'd like to think that maybe we are just at the beginning, the early "mainframe neural networks" era. That we just have to wait until there is enough technology and new markets discovered to build the "personal neural network". That consensual, open and distributed ways of sharing data will emerge and that new massively parallel computers will become affordable by the masses. That the models of neural networks will become well divulged and simplified and kids will be able to program them with their "Neural Basic"...
But at the moment the prospects don't look good. The Internet is becoming more and more centralized, personal computers harder and harder to program and there is a general "war on general computation" [1]. Even universities seem to bedisplaced by the Internet Lords in driving neural network development...
Maybe it is already a good time to start thinking about what "Libre Neural Networks" look like. And how can we get there.
[1] https://www.youtube.com/watch?v=HUEvRyemKSg