The Deep Mind of Demis Hassabis
medium.comThis article is part 4 of a series by Steven Levy about Google research. I rather like Levy's writing so I wanted to find parts 1, 2, 3 as well.
The navigation to earlier parts of the series seems broken to me (Iceweasel/Debian Sid/noscript set to allow page) so I backed up to the front page and found the links...
Part 1
https://medium.com/backchannel/how-google-search-dealt-with-...
Part 2
https://medium.com/backchannel/googles-secret-study-to-find-...
Part 3
https://medium.com/backchannel/google-search-will-be-your-ne...
Part 4
https://medium.com/backchannel/the-deep-mind-of-demis-hassab...
I do not have a GPS, most of my friends do. I notice something with everyone who has a GPS when I am with them: they cannot find their way back. They have to use the tool again.
Why am I saying this? I do not know whether these technologies are making people smarter or less smart. It certainly helps to have something that can help do the tasks you just would not be able to do on your own without much toiling, however I think a lot about the day when your AI enabled digital assistant makes every single decision for you.
It's a perception problem. GPS is an artificial organ that happens to reside outside of your brain at this stage of evolution. Most people have to use vision as well, they cannot just use their old trusty sense of smell to get around. So what it's biological?
Saying that we are now controlling the evolution (in biological sense) is incorrect.
I find the complete opposite personally, in that GPS has given me a better appreciation for my surroundings and proximity. Maybe I just look at maps too much?
Once you trust the machine to get you back, you can retake those brain cycles previously used to track that information to do something more productive, e.g. think about what you're going to do at the destination, engage in social banter, or plan the rest of your day.
You don't realize it, but you're actually less productive than your friends with GPS.
Not only do you use a different part of your brain for navigation from most everything else, it grows to accommodate its use: http://en.wikipedia.org/wiki/Hippocampus#Role_in_spatial_mem...
I never use a GPS, and generally never have to spend time thinking about directions. They just come to me, and eventually I'm at my destination and wonder how I got there. Compare to futzing with the terrible UI of most GPSes and interrupting conversations to listen to them tell me how far until the next slight bend in the road.
The banner is really frustrating. It says "Part 4" and has four dots, but I can't click the dots to go back to Part 1 and there aren't any header links to Part 1... is there even a Part 1 to go to?
parts 1-3 linked at bottom of the post.
He is yet to produce something useful. Why celebrate him now? Google PR?
That was also my feeling, and i have it pretty much every time i hear about "deep learning" : it's all PR and people trying to get as much funding as they can while the trend is hot. I have a suscipicion that it will all collapse like a bubble once again.
The only thing that really impressed me so far was that ibm computer playing jeopardy, but the fact that no other application became public after all this years make me wonder if once again all people built was a manually tuned specialized system.
Note : i'm not in the field so my feelings are just based on the communication around the subject and decades of claims about building an intelligent algorithm with no success.
Deep Mind Atari demo - http://www.youtube.com/watch?v=EfGD2qveGdQ
Do we know what reenforcement method they used ? Did the training on one level of breakout had the algorithm perform well on other levels of the same game without any new training ?
Did those games had any kind of random behavior or does the same things happen all the time at the same time ?
It is a progress, i agree, but all those games are just about issueing sequences of "left right" commands to maximize the time spent playing the game.
Things would be a lot different if they could somehow analyze the structure of the network's "conceptual" layer to identify functions over areas ( like " this is where ball trajectory is identified, and we can see it rest and activate depending on the ball's motion" or something similar). But the slide on his presentation shows a big question mark there, which isn't really reassuring.