Hacker News Reader

Comment by fizixer | Hacker News Reader

Some comments seem to be appreciating (or getting disgusted by) the aesthetics but I think the "inceptionism" part should not be ignored:

We're essentially peeking inside a very rudimentary form of consciousness: a consciousness that is very fragile, very dependent, very underdeveloped, and full of "genetic errors". Once you have a functioning deep learning neural network, you have the assembly language of consciousness. Then you start playing with it (as this paper did), you create a hello world program, you solve the factorial function recursively, and so on. Somewhere in that universe of possible programs, is hidden a program (or a set of programs) that will be able to perform the thinking process a lot more accurately.

View on HN

We're essentially peeking inside a very rudimentary form of consciousness

Blatant sensationalism. There is absolutely nothing here that would suggest consciousness. If you have a mask for matching images, you can reverse that mask and imprint it as an image. What we're seeing here is a more complicated version of the same process. Heck, look more closely. Some of those "building" images have obvious chunks of pedestrians embedded, probably because the algorithm was trained on tourist photos.

Is it interesting? Yes, from algorithmic point of view. Cool as hell. However, this has nothing to do with consciousness.

If anything, some of those images are just a more elaborate version of a kaleidoscope. It's not like they run a network and got a drawing. They were looking for a particular result, did post processing, did pre-processing and tweaked the intermediate steps (by running them multiple times until the image looked interesting). Finally, we as viewers do our share of pattern matching, similar to how we see patterns in Rorschach inkblots. And there are captions that frame what we see and "guide" us to recognizing the right objects.

> Blatant sensationalism. There is absolutely nothing here that would suggest consciousness.

Putting biological consciousness on a pedestal might be blatant sensationalism itself. By consciousness, I specifically mean the behavioral capacity of general intelligence, nothing more. If by consciousness you mean subjective character of experience then yes, there are serious issues with resolving the mind-body problem. But functionally speaking, our brains exist in a physical universe, are massively parallel, and do stochastic computations. Deep learning systems share all three of these traits except that the scale is about 3-5 order of magnitudes smaller (things like incorporating time, biological impulses are missing but if someone claims those features are going to be the dealbreaker then maybe we can have a discussion). And the scale difference is shrinking at lightning speed.

I'm not claiming DNN's are the end-all-be-all of an upcoming general electronic intelligence. But they seem to be doing mind-blowing stuff every few weeks, and it seems we've stumbled upon a radically new aspect of computation.

You did not address a single specific points I made.

I specifically mean the behavioral capacity of general intelligence

Any image-matching algorithm can be used to generate images. A simple color-matcher can be used to create very impressive things if plugged into a genetic algorithm, but no one claims it's conscious or intelligent. None of what you wrote here points out some fundamental differences between this and other image-generating techniques used before. You're simply trivializing what it means to be conscious or intelligent to the point that word is no longer useful.

I will consider an AI to be "generic" when it is able to apply training from one domain to an entirely different domain without any manual "mapping" from humans. For example, being able to decently play checkers after learning chess and being given a description of checker's rules. Applying training in image domain to image domain with tons of manual tweaking might be interesting and useful, but it's hardly qualifies as "mind-blowing".

> will consider an AI to be "generic" when it is able to apply training from one domain to an entirely different domain without any manual "mapping" from humans. For example, being able to decently play checkers after learning chess and being given a description of checker's rules.

Want to butt in to this discussion, and post this link here for anybody who hasn't seen/heard about this amazing development from a while ago: http://robohub.org/artificial-general-intelligence-that-play...

If anything, some of those images are just a more elaborate version of a kaleidoscope.

I put it to you that your 'self' is not much more complicated than any other elaborately iterated strange loop.

Reply on news.ycombinator.com