ColdFusion - Elon Musk’s A.I. Destroys Champion Gamer!

Starting point is 00:00:00 You are watching Cold Fusion TV. Hi, welcome to another Cold Fusion video. It's no secret that artificial intelligence is progressing rapidly. It seems like every couple of weeks there's an unexpected development that takes people by surprise. It's starting to become clear that we're near the start of a new error that will include artificial intelligence. Those people that aren't paying attention are likely to be taken aback by the things that are going to be possible in a few years. Some voices have spoken out about the dangers and need for safety measures that must go along with developing powerful AI. One of the loudest voices of concern was that of Elon Musk.

Starting point is 00:00:47 In a strange twist, Elon's non-profit artificial intelligence startup, which is called OpenAI, has just achieved a pretty remarkable feat. In this video, we'll take a look at what's going on. So what's the background story? Every year, the game developer, Valve, hosts a competition for expert players of the game. Dota 2, a popular online multiplayer game. The competition hosts professional players from all around the world all fighting for a 24 million dollar grand prize. This year there was a special guest competitor who wasn't human. It was an AI trained by the engineers of Elon Musk's startup Open AI. When put up one-on-one against one of the world's

Starting point is 00:01:29 best players of Dota 2, a crowd favorite, Dendi, the artificial intelligence won. Even a year ago, nobody was sure if this kind of thing was possible. Dendi was surprised that an AI could outplay a human. He said that the AI, quote, felt like a human, but a little like something else, end quote. This feat was achieved with just two weeks of real-time learning by the AI. The engineers state that during this training period, it accumulated lifetimes of experience. They also state that the rules of Dota are so complicated that if you just wrote down pre-programmed rules and programmed rules and programmed to follow it, the end result wouldn't even be as good as an average player. The artificial intelligence was trained from scratch with no knowledge of the game.

Starting point is 00:02:13 It played against itself over and over again until it had mastered the game. How on earth does a computer learn to play Dota? Yeah, so this bot is quite unlike anybody you've seen before. So we've coached it to learn just from playing against itself. So we didn't hard code in any strategy. we didn't have it learned from human experts. It's just from the very beginning, it just keeps playing as a copy of itself.

Starting point is 00:02:44 It starts from complete randomness, and then it makes very small improvements, and eventually it reaches the pro level. These are pro players. These are human brains. So are you just telling me this robot has failed so many times? It's actually better than professional Dota players. It's played for really lifetimes of experience,

Starting point is 00:03:03 and it's played so many games of Dota, It's explored many different strategies, learn to bait, learn to exploit other people who bait, and it's just played far more into the strategy space than any human has. Musk is hailing this achievement as the first time that artificial intelligence has been able to beat professionals in competitive e-sports. With e-sports being considered for the 2024 Olympics, this is certainly interesting. So this actually isn't the first notable achievement by OpenAI. They've actually done some pretty cool stuff in the past. The company actually invented a method where humans can interact with a robotic AI and teach

Starting point is 00:03:42 it just like you would a human. Here's how it works. First a human wears a VR headset, does a task and the robot watches and then imitates the task in real time without ever having done it before. In this case, it's learning how to stack some Lego blocks. That task sounds very simple for a human, but it's actually extremely hard for a machine to do this. The AI manages to do this by having its visual neural network trained on a large set of images

Starting point is 00:04:09 of what blocks in simulated environments could look like. This first visual neural network then feeds the data to another neural network called the imitation network. After this training and after just one single original demonstration of what to do, the robot can now stack blocks even with different colored blocks placed in different positions every time. This means that the robot has to and does perform different actions to the scenarios that it's already seen. The end goal is to create an AI that can adapt to new and unpredictable

Starting point is 00:04:40 environments. Okay, now, back to Open AI's recent achievement. Elon has tweeted that his feat of beating some of the best Dota players in the world with an AI is a task much more complicated than chess or the board game Go. Now I'm not so sure that I agree with Elon's statement that an e-sport is more complicated than Go. Chess, yes, definitely, but go? Perhaps not. Some things are you. state that Go is the most brilliant game ever made. Go is a 3,000-year-old Chinese board game in which an AI called AlphaGo from the company DeepMind recently beat the world champion in a series of playoffs. This event was hailed the biggest moment in artificial intelligence, not expected

Starting point is 00:05:23 for another decade. Here's DeepMind CEO, Dennis Hasibis, talking about some of the complexity of Go. Now, the thing about Go is that it only has two rules. I could teach you the game in five minutes, but it leads to incredible complexity. It's probably the most elegant game that mankind has ever devised. What happens in Asia, in Korea, in Japan or in China, is if you show promise in the game of Go, at the age of sort of five or six or seven, you get taken out of normal school,

Starting point is 00:05:51 and you get put into Go school where you study Gohs a day, seven days a week, with your peers, who are also trying to become professional Go players. So this is taken really seriously, and it's been like this for hundreds of years. hundreds of years. Now one way to illustrate the complexity of the game is that there are more board configurations in the game of Go than there are atoms in the universe. So there's no way

Starting point is 00:06:15 that you can solve this game through brute force calculation. It's much too complex. Even if you took all the compute power in the world and you ran it for a million years, that wouldn't be enough compute power to calculate all the variations in Go. If children are literally pulled out of their school to spend half of their lives training to master this game, I can't really just dismiss it and say that Dota 2 is more complicated. Although, again, this could depend on your definition of complicated, but it's just something to think about. On the topic of DeepMind, I imagine that some of you would be interested in a few juicy updates about what DeepMind's AlphaGo is up to. Also, what has humanity learned from that moment when the Go World Champion was defeated by an artificial

Starting point is 00:06:57 intelligence. Well, while playing its winning match against the Go champion, AlphaGo played some very strange moves that ended up giving it the advantage to win the match. To be clear, in over 3,000 years of humans studying and playing Go, we've never thought of playing the way that AlphaGo did. AlphaGo's moves have now been used in Go schools to train students and expand their way of thinking about how to play the game. And further to this, DeepMeyn plans on using the AlphaGo algorithm them for more general purpose functions as it shows great promise in its ability to learn. And I think of AI as this incredibly powerful tool that will augment human ingenuity and unlock our true potential. In fact, one way you can think about AI and indeed AlphaGo is like, I think

Starting point is 00:07:44 of it as analogous to the Hubble telescope, a kind of ultimate tool to explore the universe. Of course, for Go players, AlphaGo was allowing them to explore their universe of the game of Go. And I think there are many other domains in the real world that suffer from this kind of combinatorial explosion that Go has. Now, obviously, as I said at the beginning, we test our systems on games because they're the most convenient way to develop our AI algorithms. But obviously, ultimately, we're not interested in just being good at games. We want to translate those algorithms into the real world and be useful and make huge impact on real world situations. And one reason we believe we can do that is because we're building general purpose learning systems. They're not being handcrafted for, say, like chess.

Starting point is 00:08:34 We've actually built, we believe, general purpose algorithms that could be taken from the games that we test them on and apply to real world. And we're applying these to all sorts of other areas, healthcare, robotics, and even optimizing data centers. So a variation of AlphaGo we took over last summer and we applied it to Google's data centers and we managed to save 15% of the power that was used in those data centers by controlling the cooling systems more efficiently.

Starting point is 00:09:06 So there you have it. As always, it's some very interesting times in the field of artificial intelligence. It seems that artificial intelligence playing games would be a great way to enable it to learn But further to this, the knowledge learned could be applied in other fields, making the active playing games much more important than it would first seem. Both deep mind and open AI are going with this strategy. So what's your view in the story?

Starting point is 00:09:31 Are you at all surprised by this? Or is this all just normal to you by now? If you're a subscriber with this channel, I'm sure you're pretty used to it. Anyway, that's the end of the video. Thanks so much for watching. If you've just stumbled across this channel, feel free to subscribe. This has been to go-go. You've been watching Cold Fusion.

Starting point is 00:09:46 I'll see you again soon for the next video. Cheers guys, have a good one.

ColdFusion - Elon Musk’s A.I. Destroys Champion Gamer!

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