Your Brain On Distributed Deep Learning

September 4, 2018

Your Brain On Distributed Deep Learning

Brains are wonderful and strange. The human brain is a great greyish-pink mess of neurons. It sits in your skull, rippling with tiny shocks of electricity, constantly creating and destroying neural connections. Some areas are specialized: there are neural networks for memory, for feelings, for vision, for language, for problem-solving. It’s a nest of networks. It’s incredible. And it has a copycat.


Imitation is the Highest Form of Flattery

IBM’s announced that they’ve taken another step towards robo-brains: they’ve made substantial developments in deep learning. This time, they figured out how to distribute a deep learning task across multiple servers. Maybe those words alone don’t stop you in your tracks, but it’s a very big deal. What they’re doing enables AI to dramatically accelerate its self-education. That means the structural and computational gap between “us” and “them” is getting smaller. It’s another rope thrown across the canyon between AI computation and the human brain.


Not So Far-Fetched

Saying that even IBM’s brightest AI looks like human consciousness is a leap too far. But if you compare each deep learning machine to a neuron, capable of building and breaking connections… you can see that a network of these machines forms a crude approximation of a human brain.


A Thought Exercise

While you’re busy staring out the window lost in thought, ponder what this could mean for the workers of tomorrow. Let’s try a thought exercise: how about we all hold hands and take three hypothetical steps forward into the unknown.


  1. Deep learning gets faster: This means data gets cheaper and more bountiful. Companies will be sure to take advantage of that, and jobs will bloom accordingly.
  2. Deep learning gets deeper: We’ll be able to extrapolate larger and larger meanings from all this data. That means automation of a lot of analysis jobsbut not necessarily unemployment. White collar jobs will be transformed into blank collar ones, where employees concentrate on semantics rather than crunch numbers.
  3. Deep learning gets brainier: Computers start to mimic the human brain. Don’t call Skynet just yet! That just means the interface of deep learning machines will start to mimic human mental processing. AI software will be able to get the same output from input as a human brain would. This one could get interesting: combined with the right hardware, we might be able to hook it up directly to our brains. When Elon Musk talks about neural lacing that’s the future he’s referring to. It’s hard to imagine the effects something as incredible as that would have on the workforce.


You can’t believe anyone who says they can predict the future. But thanks to distributed deep learning, it’s a future we can picture a little clearer today.

P.S.: If you’re still reeling after hearing about neural lacing, check out this great detailed article on WaitButWhy.