ColdFusion - Google's Deep Mind Explained! - Self Learning A.I.

Starting point is 00:00:00 You are watching Cold Fusion TV. Hi, welcome to another Cold Fusion video. For the first time in history, we're seeing the successful development and demonstration of general purpose AI. Its name is AlphaGo and it was made by Google's DeepMind Division. The AlphaGo AI algorithm beat a human at the world's hardest game. This is a historic moment in computer science and artificial intelligence. So what exactly is Google DeepMind? Where did it come from? And what can the company's artificial intelligence actually do? In this video, we'll find out.

Starting point is 00:00:46 A computer program beat a human brain at the ancient Chinese board game of Go. The final score 4-1, a triumph for artificial intelligence equipped with human intuition. The victory of a program over a human in the ancient board game Go has sparked intrigue and, in some cases concern. It shows that a machine has approximated human intuition and outsmarted the best human brain in the game. It's something that scientists hadn't expected to happen for at least another decade and it's a giant leap for artificial intelligence showing that machines can learn on their own. I think maybe the game ended. No I don't think so because it looks like a

Starting point is 00:01:33 V is still counting. No I think he resigned. Wow I think you're right. Wow, wow. This was the moment a computer called AlphaGo beat a master of the ancient Chinese game Go. It's not the first time a grandmaster's been humbled by a machine. But what makes AlphaGo different is that it's the first demonstration that machines can truly learn and think in a human way. AlphaGo's victory shocked experts in the artificial intelligence community. Many thought such an event was at least a day So firstly, a few questions. Why is this important and what's all the fuss?

Starting point is 00:02:17 AlphaGo shows that machines can really learn. How so? Well, instead of using brute force to calculate all the moves it could make like previous AIs, AlphaGo used reinforcement learning and neural networks to mimic the learning process of a human brain. Keep in mind that the ancient Chinese game, Go, has as many more possible moves than chess as their are atoms in the entire universe. So there's no way of just calculating every possible move on the board. That's practically impossible. For this reason, Go is the holy grail of AI, and learning to do such a task from scratch is a huge feat. Further to this, Deep Mind's creators say that the algorithm can learn many more things without alteration or guidance. In other words, the AI is general purpose.

Starting point is 00:03:05 We always used to talk about, well, if we could eventually crack Go and have a program that could be the world champion, then we must have invented some generic general purpose algorithm. So maybe we're on the cusp of all of that. If you ask a great Go player why they played a particular move, sometimes they'll just tell you it felt right. So the one way you can think of it is that Go is a much more intuitive game, whereas chess is a much more logic-based game. So what is DeepMind? Google DeepMind is a British artificial instrument. intelligence company that was founded in September of 2010 as DeepMine Technologies. It was renamed when it was acquired by Google in 2014 for $500 million.

Starting point is 00:03:45 Interestingly enough, this was just after Facebook had just finished negotiations with them in 2013. And another fun fact, Elon Musk is actually an investor in the company, just to keep an eye on them, of course. DeepMine received the Company of the Year award by the Cambridge Computer Laboratory in 2014. So where did DeepMine come from? DeepMind was first a startup co-founded by Demas Hassebis in 2010. Demis was a child genius of sorts and went from being a chess prodigy and reaching master level at age 13, onto becoming the lead programmer of Lionhead Studios for groundbreaking games such as Black and White.

Starting point is 00:04:25 After this, he went on to start his own gaming studio before leaving to complete some extra studies at the University College of London. This is where he would meet the future co-founders of DeepMind in 2010. So what is the goal of deep mind technologies? According to their website, their goal is to solve intelligence. They are trying to achieve this by combining the best techniques from machine learning and systems neuroscience to build powerful general purpose learning algorithms. Or in other words, they want to formalize intelligence, find out what it is and how it works.

Starting point is 00:04:59 The end game is not just to implement this into machines, but also understand the human brain. Demas explains further. Quote, attempting to distill intelligence into an algorithmic construct may prove to be the best path to understanding some of the enduring mysteries of our minds." The key here is that this machine is general purpose. This is very important, but more on this later. Okay, so how does this all work? The AI from DeepMind uses a technique called Deep Reinforcement Learning, which makes it very

Starting point is 00:05:33 different from other AIs such as IBM's Watson or the primitive Siri or Google Google Now. The AIs just mentioned were only developed for a predefined purpose and only function within their scope. DeepMind claims that their system is not pre-programmed and it learns from experience using only raw pixels as data input. So the way we start off training AlphaGo is by showing it 100,000 games that strong amateurs have played that we've downloaded from the internet.

Starting point is 00:06:02 And we first initially get AlphaGo to mimic the human player. Of course, ultimately we would like AlphaGo to be stronger than human amateurs and compete with the top professionals. So the way we do that is, after we take that first version that's learned to mimic human play, we then allow it to play itself 30 million times on our servers. And using reinforcement learning, the system learns to improve itself incrementally through avoiding its errors and increasing and improving its win rate against older versions of itself. and after all these games then you end up with a new version that can beat the old version, the original version, around 80, 90% of the time.

Starting point is 00:06:44 Technically, the reinforcement that DeepMind uses is model-free, meaning that it doesn't need a structure or set of rules to learn. According to MIT's take on the Google algorithm, the neural network, in theory, should be able to access an external memory in a manner that mimics the short-term memory of the human brain. In a way, it almost seems like our brain's design might be the most. efficient way for the creation of local intelligence. Who would have guessed? So that's all good. What about this general purpose term? What does that actually mean? Well, because the machine learns from more experience and data, it can perform well across a wide variety of tasks, straight out of the box. Okay, so that's enough talk. Let's watch the system in action, learning how to play breakout. And remember, the machine didn't have any foreknowledge as to what task it was going to

Starting point is 00:07:33 complete. It had to figure out how to do everything only from its own experience. Is the machine before training? So this is literally the first time the machine has ever seen this data stream and it's controlling the green rocket here and you can see it's losing its three lives immediately. So it's basically terrible at the game as you would expect. Now after training this is like overnight we leave it playing the game overnight on a normal computer so for about eight hours you come back in the morning and now it's superhuman level. So now the machine can play Space Invaders better than any human can. It's modeled the world, the game world so well,

Starting point is 00:08:08 it knows where the Space Invaders will be ahead of time. So as the Space Invaders speed up, when you get less of them on the screen, it does as far as the final shot is a kind of predictive shot of where the final Space Invader will be in a few seconds time. But the cool thing is, after understanding how to play a few games, it will be able to transfer this knowledge and understanding across more games. At the moment, Deep Mind is currently playing games from the 1970s and 80s, But work is currently being done on more complex 3D games such as Doom, which appeared in the early 90s.

Starting point is 00:08:39 So what could general purpose AI be used for? We'd love to use these types of algorithms for things like healthcare and science and improve the speed of breakthroughs in those areas by helping human experts achieve more. In a number of interviews, Dennis has talked about applications in healthcare, smartphone assistance and robotics as possible applications for general AI. Some other applications of such AI include online customer service, computer vision, finance, general computer science, and news publishing and writing.

Starting point is 00:09:12 General purpose AI actually could be huge. It can be thought of as another emerging science. Much like Newtonian physics laid the groundwork for technology such as rockets, a science founded in general purpose digital neural networks that use reinforcement learning could lay the groundwork for unimaginable things in the near future. Okay, so we're almost at the end of the video,

Starting point is 00:09:33 but there's one pressing question, should we be worried? In the thousands of years of meaningful civilization, our modern existence right now is an anomaly. We've overcome the limits of nature with technology that helps us facilitate such a lifestyle, but artificial intelligence is different. The reason being,

Starting point is 00:09:51 AI is the first thing humans have created that tends to function in a way that we can't predict. So, should we worry? My belief is that the fear of the unknown is mostly what drives the hysteria that arises from AI anxiety. There's not enough time in this video to go through all the theories of how things could turn out, but it's an interesting topic within itself. The website Wait But Why has a fairly interesting and comprehensive look at AI.

Starting point is 00:10:19 We don't have time to explore at all, but the article does raise an interesting point though. It's about how humans have an inability to perceive exponential change. For a long time, people aren't going to take AI seriously and it would seem like progress is very slow, but then suddenly it could overtake our wildest expectations. One of the methods that AI could use to achieve this outcome is by giving itself an ability for self-improvement, thus following the law of accelerating returns. Stephen Hawking and Elon Musk have both shown their reservations towards AI, but others like Paul Allen, co-founder of Microsoft, have different views. If you can forget about the movie Terminator for a second, there is a possibility that AI could be man's greatest aid, fixing fundamental problems on our planet, and allowing us to have technology and infrastructure that could have otherwise taken us centuries if we tried to figure

Starting point is 00:11:11 it out ourselves. But on the other hand, there is a possibility that AI could be malicious and would outwit us, but at this stage we really just don't know yet. And again, this is a story for another day. So at the moment, I'm not worried, but cautiously optimistic. As for Deep Mind, one of their conditions after Google's acquisition was that they established an AI ethics board, and I think most of us would agree that that's a prudent decision. All in all, we may be at a historic point, a critical point in artificial intelligence.

Starting point is 00:11:46 The future could see deep neural learning and general purpose AI in our smartphones and computers. Some say that we might even have to start thinking of a whole new economy. Reason being, AI could take over a lot of information intensive, specialized jobs, leaving only jobs that are creative or emotional in nature, something that AI might never achieve. I honestly can't say where all of this will lead with any certainty, but then again, I don't think anyone can. If you take a look at the macro picture in all of this, after many decades of trying to build smarter computers and algorithms, it turns out that the best blueprint,

Starting point is 00:12:24 was in our heads the whole time, the human brain. So naturally, our minds are capable of so much, it just depends on how we use it. So that's the end of the video. Thanks for watching guys. Don't forget to give this video a thumbs up if you liked it, leave a comment and subscribe if you knew. You might want to do that. If you want to see some more cool stuff, just hang around for the next 10 seconds and follow the prompts on this screen.

Starting point is 00:12:49 Cheers guys, thanks for watching, and I'll see you again soon for the next video. Have a good one.

ColdFusion - Google's Deep Mind Explained! - Self Learning A.I.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.