15 thoughts on “Breakthroughs in Machine Learning – Google I/O 2016

  1. ha ha ha… at 18.44 he said even hinglish!!!! ๐Ÿ˜€๐Ÿ˜€๐Ÿ˜€ hindi+English… ๐Ÿ˜œ

  2. There's an English version of the Il Pulcino Pio (and a Spanish one and a Japanese one and… well you should get it by now, almost any language). The Italian version is still better tough ๐Ÿ˜€ and honestly? Who wouldn't love to watch at that video?! ๐Ÿ˜€ Specially after Gangnam Style ๐Ÿ˜›

  3. Check out and subscribe to our channel to hear more about the uses of AI in business from experts at the forefront of the 4th Industrial Revolution working with companies such as IBM, TCS and Publicis.Sapient.

  4. This is truely incredible, respects to the developers behind this work. Question from a non programmer, CS illterate: eventually we will have an AI as a collection of different algorithms that replicate our logic, and pattern finders / platform building thinking based on the data we produce for it on an ongoing basis and the senses we give it (way beyond our 5/6 if you think of all the instruments available).

    We will not be able to stop its potential "misreads" such as kdkdkd (but on a much more immediately impacting level), when that combination is say 95% properly put together… And that collection is slowly coming together, perhaps before a human takes claim or suggest a structure for the "universal code". The human mind is itself hard to tame…. even by its owner. I guess its just a message to the developers… Think ahead for potential lethal logic that could be a synergy from two or three that may be commonly used for simpler tasks.

  5. Are you sure its the ai I believe I have a neural connection and seem to be able to combine technology with visual augmentation and produce new apps instantly by quantum enhancement is this possible? although my computer identity comes back mismatched slightly, I am not a programmer by the way i,m an enthusiast with a personal computer, I seem to know stuff ? .x. (pip).x

  6. I have translater app,next couple of days will try and combine skype with games console, ??

  7. Yeah, at the end of the video is a very clear illustration of one of the downfalls of our current way of doing machine learning. It's not a problem with machine learning itself, but of the systems we've set up. Yeah, Il Pulcino Pio makes sense after Gangnam Style…. if you're potty training a 2 year old. What if you don't have kids, or you do have kids but they have respectable bowel control? Well, then you're boned. Because the machine learning is done on aggregate data on corporate servers that couldn't handle the load of training separate networks for every single user. And there are business motivations which prevent them from pushing the substantive work down into client devices (even though they could). The solution thus far has been, at best, to break the userbase up into rough categories and presume that every user is an average representative of one of those rough categories.

    This often has a tendency to turn a tractable problem into an intractable one. For instance, let's suppose you have a very simple habit. You hate jumping into the middle of a series. If you happen upon a video and it's part 2 or part 3 or part 18 or part 190 of a series, you always, without fail, want to start watching from part 1 and watch it in order. A very, very basic pattern that any machine learning system could pick up on if it were studying your personal history. But, keeping a log of every video you have watched and training on it? Impossible. The web offers extremely little (up until the past couple years, none) facility to store information on the clientside and storing a complete view history on the serverside would require too many resources. Then training on top of that? Forget it. So instead you get profiled as the kind of person who watches hedgehog videos and since a meme came through that related hedgehogs and hyperkinetic violence, the system recommends you watch some hyperkinetic violence instead of Hedgehogs in History Part 1.

Leave a Reply

Your email address will not be published. Required fields are marked *