A.I. is progressing faster than you think!
You are watching 12th Fusion TV. Hi, welcome to another Cole Fusion video. At the end of last year, I made a video about 5 new technologies which most people didn’t expect to exist. The episode included an AI that could look at a static image and dream up a video of what it thought would happen in the next few seconds. Yes, it created a video from a still image. I went on to describe another AI that could guess what’s happening in a scene just by listening to the audio. It was consistently better than humans at doing this. In the same video, I talked about yet another AI that can predict human behavior with training only from watching sitcoms. If you haven’t seen this video yet, you’re missing out. The link to the video will be the first one in the description. Today, we’ll be looking at another interesting AI. Imagine typing a descriptive sentence of a scene and having an artificial intelligence generate a convincing photorealistic image just from your text input. This is just being created and that’s what we’ll be taking a look at today. After this, I’ll share some of my wider thoughts on artificial intelligence. Let’s get straight into it. Before we dive straight into the AI, we need to build a bit of context. The type of AI that we’re about to look at is called a neural network. That’s basically a computing system that’s modeled after the human brain. There are processing nodes that act as neurons and the neuron layers behave as segments of the brain. This concept and idea is nothing new and it has been around since the 1980s, but it’s only become feasible in the last five years due to GPUs with hundreds of cores being cheap and accessible. In addition to this, the available open source tools for neural networks are making it easier to create, pushing progress parabolically faster. So some examples of neural networks include Google’s Alpha Go and also things like this, an AI that’s capable of describing images in text form. Another interesting example of neural networks is WaveNet, an AI that can generate raw sounds such as speech and music. Here are some examples. This first example is speech generated by WaveNet with a twist. WaveNet wasn’t specified what to say, so it had to make up its own language. Notice the breathing and natural lip sounds. Hey, the two that are sitting out all sweaty, I should see the cellar’s eyes. They’re loud and goo. Here’s the head sheen with going to a few cells apart. Do you ever want to share that study given as well? I can jump right out of the cell, I just wrote a game title, all of the others are to be watched on my first AI. This next clip is some music that WaveNet generated. Again, it wasn’t told what to play, there was no musical score. It just played whatever it wanted. For this piece, WaveNet was trained by listening to classical music pieces. So this is great, but what if you want to take things further? What if we combine two neural networks together and made them compete against each other so that they could train and improve themselves without human intervention? That’s what our featured AI called StackGan is doing. It uses one neural network to generate images and another neural network within the same system to decide if the images generated a real or fake. What ends up happening is that the generative neural network improves itself at generating images based on the feedback given by the deciding network. And in the same stride, the deciding network gets better at distinguishing what’s real and fake. This creates a feedback loop of continuous improvement without human intervention. The end process of this is a creation of a low resolution synthesized image. After this, a second stage takes place. In the second stage, the AI is told to clean up any defects in the original picture, and the results are nothing short of stunning. Yes, what you’re looking at are images created just by a written text description. This type of dual network AI is called a generative adversarial network, and it’s the same type of system that allowed the AI mentioned in my previous video to dream up those videos when shown a still image. This form of artificial intelligence is actually brand new and was only invented in 2014. And already, I’m beginning to think that this is one of the most powerful methods for machine learning. The text to image AI comes hot on the heels of an artificial intelligence from Carnegie Mellon University beating the world’s best players of Texas Holden Poeker. I remember some people commenting on my deep mind video telling me that this exact event was something that artificial intelligence has been struggling with for some time. There’s also an AI that’s been released by Gamalon, a Boston company. This one can rewrite its own code based on experience and probabilities rather than hard variables. It’s created to say that it could make the tedious part of coding AI completely automatic. To give you an idea how far some areas of AI are progressing, here’s a clip of Jeff Deane, a senior fellow at Google’s research group, giving a presentation at a TED talk. To take an example, the field of computer vision. Every year there’s a contest where teams compete to see who can give the right categories out of a thousand different categories when given an image. And in 2011, before people were using neural nets, the winning team got an error rate of 26%. Which doesn’t sound too good when you think that humans are at 5% on this task. But fast forward five, just five years. And we’re now at 3% errors using deep learning and much more computational power. We’re actually better than humans on this task. So, in a way, computers can now see and recognize objects better than us for the first time ever. This has never happened before. In the same talk, he goes into give an example of how unreliable human optimologists were for diagnosing certain eye diseases. In experiments, any two human optimologists will only agree with each other’s diagnosis 60% of the time. And what’s worse, if you give any single optimologist the exact same image that they diagnosed a few hours earlier, they’ll only agree with themselves 65% of the time. In 2017, artificial intelligence image recognition has been proven to perform better than professional humans in this field. As you’re seeing in this video, things are progressing fast and may have taken some of you by surprise. Even the president of Alphabet, Sergei Brun, is being taken by surprise by the entire phenomena of AI in general. AI effort, but I didn’t pay attention to it at all. To be perfectly honest. And myself having been trained as a computer scientist in the 90s, everybody knew AI didn’t work. It’s not like people tried it. They tried neural nets. None of them worked out. Yeah, this kind of revolution in deep nets has been very profound and definitely surprised me even though I was like right in there. It’s an incredible time. And it’s very hard to forecast. What can these things do? We don’t really know the limits. Everything we can imagine and more. It’s a hard thing to think through and has really incredible possibilities. It seems like playtime is over in regards to AI, especially with techniques like deep learning and neural networks. I think such things will cause social discomfort as they begin to encroach upon jobs. There’s definitely both positives and negatives to this though. But the question is, what does all of this mean and what are we to do with all of these continual shifts and what we thought was possible? In a general sense, artificial intelligence can be seen as a continuation of the industrial revolution happening over the past 200 years. Even right now, artificial intelligence is helping in the medical field and making driving and other risky tasks safer. In the future, many mundane tasks can be handled by AI, giving more free time for people to do creative things. But of course, and I’ve heard this argument before, not everyone is creative. Life needs to have meaning and for some individuals, that meaning comes in the form of their day to day 9 to 5 job. I’ve been thinking about making a video about possible financial solutions to AI disruption. This includes concepts such as universal basic income based on blockchain technology and insights into how some researchers are starting to think about such things. Now of course, the sum of you’re going to be thinking of a SkyNet scenario, but right now it’s still far too early to predict. Predicting the fully developed AI of the future would be like trying to predict today’s computer industry from the point of view of the 1960s. Personally, I think what we should actually be worried about is who controls the AI. If the strong artificial intelligence of the future is open source and available to everyone, there’s a high probability that this will be a good thing. Here’s a small example of an open source AI being put to good use. A farmer in Japan used AI computer vision and robotics to enable his cucumbers to be automatically selected and categorized saving him hundreds of hours. The artificial intelligence driving his farm is called TensorFlow and its open source and was created by Google. Maybe in the future, instead of working for a company, there will be more common for people to run their own businesses with their own personal AI acting as a productivity aid. So that’s an example of open source AI being used in a good way. But conversely, if some of the best AI in the world is held in the hands of a few, the probability of a good outcome overall drops significantly. In the latter case, it may be a case of whoever owns the AI makes the rules. To stop AI getting into the wrong hands and to stop it becoming dangerous, institutions such as the future of life institute have been set up. It has tens of millions of dollars in funding and shows that people are already thinking about the next best steps for the future. So that’s pretty much the end of this video and I’m going to handball it to you guys. What do you think of the progress being made by AI in the past few years? Leave your thoughts in the comment section below. Anyway, thanks for watching guys, this has been Degogo, you’ve been watching Cole Fusion, subscribe to this channel if you just stumbled across it, don’t forget to check out my other video and the five new technologies that you wouldn’t think to exist yet. And of course, I’ll see you again soon for the next video. Cheers guys, have a good one.