The 2016 Go match between AlphaGo and Lee Sedol put deep learning in the spotlight. Deep learning enables AI to learn on its own and has revolutionized many fields, including image recognition, speech recognition, and more. AlphaGo won the game through unsupervised learning, and advances in deep learning technology will positively transform industries such as healthcare, education, and finance.
In March 2016, the Go match between AlphaGo and Lee Sedol shocked the world. AlphaGo, an artificial intelligence that utilizes deep learning technology, defeated the 9th ranked Go player Lee Sedol. Since then, the term “deep learning” has become a household word. In fact, deep learning is an old theory that’s been around for more than a decade. However, many people don’t know much about it. In this article, we’ll explain what exactly deep learning is, what makes it different, and how AlphaGo applies it.
First, let’s take a quick look at the history of artificial intelligence. The history of artificial intelligence starts with Alan Turing in the 1950s. In his book “Computing Machinery and Intelligence”, Alan Turing talked about the possibility of artificial intelligence, saying that the day could come when artificial intelligence could think like humans. Since then, many people have been working on AI, but the limits of AI development were reached in the 1980s. For about 20 years, no one tried to develop AI.
However, in 2006, Professor Hinton introduced deep learning techniques in his paper “A fast learning algorithm for deep belief nets”, and AI research gained momentum. Deep learning refers to machine learning techniques based on artificial neural networks (ANNs) that allow computers to learn on their own from a large amount of data, just like humans. Based on the motif that humans think and act with hundreds of layers of neurons, artificial neural networks are computationally constructed to make computers learn.
Artificial intelligence has been studied in many fields, and deep learning has revolutionized many areas, especially image recognition, speech recognition, and natural language processing. For example, Google Photos’ facial recognition technology, Apple’s voice recognition feature in Siri, and Amazon’s recommendation system are all examples of deep learning. These technologies have made our daily lives easier, and deep learning is expected to be applied in many more fields in the future.
But what makes deep learning different from traditional methods? Artificial neural networks have been studied in the past. However, there are two main problems with existing artificial neural networks. The first is the local minimum problem, which is the problem of outputting the wrong answer before finding the true answer. The second is overfitting, where the training data and the real data are different, resulting in errors.
However, deep learning technology has enabled unsupervised learning, prioritizing the first problem. Deep learning allows computers to constantly feed back the correct answer, which has led to an exponential increase in accuracy. The second problem was solved with the help of big data. In the traditional method, training data was processed to make it easier for the computer to understand, but with the development of big data, a large amount of information can be provided to the computer at once. Based on this, unsupervised learning was implemented. In other words, the computer was actually learning unsupervised from real data. With both of these problems solved, deep learning ushered in a new spring in the field of artificial intelligence.
So how did AlphaGo incorporate deep learning? AlphaGo was trained unsupervised based on amateur Go players. Through unsupervised learning, AlphaGo updated its information about the probability of winning. Eventually, it was able to use probability distributions to determine which move was better at each moment in time. AlphaGo used an algorithm to identify 2-3 starting points with a high probability of winning. The algorithm we’re talking about here is a combination of the value network method and the rollout method. The former is a method that instantly evaluates the value of each starting point based on Monte-Carlo search techniques, and the latter is a method that extracts high-probability starting points based on the statistics of all AlphaGo’s stored notations. By utilizing these two methods in combination, AlphaGo reduces the complexity of the computation by avoiding counting all the cases of Go. This is how AlphaGo was able to defeat Lee Sedol 9 in Go.
In the previous sections, we discussed the evolution of artificial intelligence, what deep learning is, its benefits, and how it was applied to AlphaGo. Deep learning is a revolutionary technology in itself. However, the most important significance of deep learning is that it has shown us that it is possible to replicate human self-organization in machines, which makes it possible to envision human-like artificial intelligence that is not just science fiction. If deep learning technology can be further developed and applied to technologies that benefit humans, humans will be able to live in a more convenient world if deep learning AI can solve problems that humans can’t easily solve.
Therefore, the development of deep learning technology is not just a technological innovation, but has the potential to revolutionize society as a whole. Deep learning can be utilized in a variety of industries, including healthcare, education, finance, manufacturing, and more, which will further improve our quality of life. For example, in healthcare, deep learning will enable precision medicine, which will allow for personalized treatment, and in education, it will allow for personalized education based on individual learning levels. These are all positive impacts of advances in deep learning technology.
As such, deep learning will continue to evolve, and our lives will become more convenient and enriched by its advancements. I hope this article has given you a better understanding of deep learning and prepared you for the changes ahead.
