From Alpha Go to Neural Network to Future of Artificial Intelligence and real life Skynet

Recently I’ve being following the game at Wuzheng about Alpha Go and its game with Kejie, the current world champion in the game of go. This game has a very significant effect on the development of AI for two reasons. One, the game of go has being considered one of the hardest games to solve for AI because it’s near impossible to do with brute force. The number of possible moves in go is magnitudes higher than the number of atoms in the observable universe. The other reason is that since the game with Lee Sedol last year, Alpha Go has improved significantly. It actually teaches itself how to play go (unfortunately there’s no human that can actually teach it anything), and it’s done using one tenth of the computation requirement.

For these of you interested, there's 208168199381979984699478633344862770286522453884530548425639456820927419612738015378525648451698519643907259916015628128546089888314427129715319317557736620397247064840935 possible legal moves in a 19 x 19 standard game board. source: http://tromp.github.io/go/legal.html

How Alpha Go works is by using deep reinforced learning. (If I’m wrong in the following explanation of neural network and machine learning then please let me know. All my knowledge of neural network is from Wikipedia, Waitbutwhy and 2 university lectures on introduction to neural network) Essentially it sets up multiple nodes that have which either links to a raw input, another node, or the system output. Each node filter the signal it receives and sends it off with some modifications. Through trial and error, it modifies the filter criteria of the value received as well as its modification to the value before passing it on. And it’s very like how our biological neuron functions.

The human race actually functions similar to a neural network as well. You have celebrities and politicians who have a heavy influence on the direction of society. You have people who don’t have as much of an influence individually but indirectly impact the society as a whole (most people are like this). And you have isolated native tribes who probably isn’t even aware that Earth is round and we have airplanes and probably do not influence the mass society at all. And we pass information to each other through oral and written methods like media, advertisement, blogs and reddit posts. We are actually a pretty terrible neural network in that the information we try to pass on is usually really slow, often lost in translation and most of the time gets lost.

One way of improving the efficiency of humanities’ neural network is to improve the method of passing down information. Thus one of the heavy influencers of our society, Elon Musk, recently decides to start a company called Neuralink to speed up and improve our communication. (Once again, everything I know about it is from this waitbutwhy article). If we can achieve better information flow, society as a whole will be capable of learning faster and hopefully make better decisions for the entirety of human race.

Now let’s go back to why Alpha Go is important. Alpha Go, and more broadly artificial intelligence, can solve the awful efficiency of our human neural network using a different approach. We can use machines to learn lessons and take them directly, by passing the low-effectiveness of the human neural network. We, humans, are actually terrible learners as well. It takes us many tries to break our habits and years of practice to become good at something. Grandmasters at Go or Chess still takes years of practice to get to where they are. Expertise is no more than a multiplication of efficiency and practice. Artificial intelligence however, have a much higher efficiency and can practice millions of times in the span of days! This means AI are much more capable of learning than humans. Any job that require expertise can be replaced by AI in the future. Based on the progress of Alpha Go, I forecast we’ll be able to achieve artificial general intelligence (AGI), AI that’s capable of learning non-specific tasks, within the next 10 years. By 2027, we might have android that is indistinguishable from humans. They’ll seem like a super smart person to the rest of us.

So one of the more common concerns for achieving AGI is whether this will cause a robot uprising and slave the human race under a robot overlord like Skynet. My view is that it is very unlikely unless someone tell it to do so. And the reason is that while machine learning algorithms like deep reinforced learning can learn better than humans do, it isn’t capable of handling tasks without a way to judge right and wrong, such as philosophy or morality. A society of artificial general intelligence wouldn’t attempt to fly to the moon even if they could, because there is no point of doing so. This is a human thing to do; we set illogical goals because we feel like it. So if we don’t tell it to take over the world, it won’t have a reason to. If the AI have a morality fail safe like the the three laws of robotics, it wouldn’t take over the world.

However, there’s another danger that I’m more worried about when it comes to AGI (beside the danger of it falling into the wrong hands), is that artificial intelligence lacks wisdom. The definition of wisdom used here is learning over a period of long time. We humans grow more wise, both as an individual and as an race, over time. It took us decades to even centuries to realize the effect of global warming, or the health effects of DDT, because they lack short term effects. The good news is, by the time we noticed, it was still a fixable problem. However that’s not quite true with AIs, they move so fast that by the time it realizes that something is wrong, it might already be too late. By the time the AIs notice the heat death of the universe is speeding up, they might have already sucked away hundreds of thousands galaxies worth of energy.

Hacker Noon is how hackers start their afternoons. We’re a part of the @AMIfamily. We are now accepting submissions and happy to discuss advertising & sponsorship opportunities.

To learn more, read our about page, like/message us on Facebook, or simply, tweet/DM @HackerNoon.

If you enjoyed this story, we recommend reading our latest tech stories and trending tech stories. Until next time, don’t take the realities of the world for granted!