At the end of the series, we look back at why artificial intelligence challenges for humans were possible, and also think about the future with AI in the future. If you haven't watched past content in the series yet, I recommend checking it out first.
[AI Story] Artificial Intelligence Challenging Humans (1) Deep Blue (Deep Blue)
[AI Story] Humans vs. Artificial Intelligence (2) Watson (Watson)
[AI Story] AlexNet (AlexNet) opened the era of human vs. artificial intelligence (3) deep learning
[AI Story] Humans vs. Artificial Intelligence (4) Pepper (Pepper), a robot that reads emotions
[AI Story] Humans vs. Artificial Intelligence (5) AlphaSGo (AlphaSGo) calculating intuition
AlphaGo's victory in 2016 shocked us greatly. This was due to the nature of the game Go, which is known to have an almost infinite number of situations that can unfold to be called the “universe above half.” Until now, it was thought that it was impossible for a machine to surpass a human in Go battles that required high-level thinking such as intuition and reasoning.
Also, until a while before AlphaGo appeared, there were many people who were skeptical about AI. In fact, it was a time of widespread disappointment called the winter of artificial intelligence. At the time, AI research was in a difficult situation, as financial investments were cut off because the results people expected were not achieved.
However, a few pioneering scholars did not give up, and in the end, they achieved today's remarkable results. Of course, it wasn't just because the researchers had a high sense of scientific obligation. If so, what changed in the meantime so that AI, which once seemed to have fallen into a dark period, was able to achieve a reversal?
Increased computing power
Artificial neural networks that train AI are known to mimic the way the human brain works. Specifically 'By imitating biological structures, computing devices are used as neuron inputs, and weights indicating the strength of synaptic connections are applied. Learning occurs by adjusting the weight assigned to each input. '*It says. In other words, each connection point interacts with the surrounding connection points to understand the signals received by the neural network.
The principles of neural network algorithms, sources
This “As the layers get deeper, the amount of weight to be learned increases, and the amount of computation increases exponentially.”It means that it is**. However, until now, a typical computer processor only processed one at a time. In other words, the amount of computation to be processed has become so large that it is impossible to rely only on existing CPUs.
However, the use of GPUs enables a revolutionary leap forward. Alexnet (Alexnet), which I introduced a while ago, was the starting point of this innovation. “Using GPUs that are advantageous for parallel computing, it is now possible to quickly and accurately process large-scale computations required in deep learning.”***
AlphaGo too “It's hard to guess, so many computational volumes were possible because there was a high-performance system with 176 GPUs. Since the computation speed was more than 30 times faster than a typical CPU system, it was possible to perform effective computation in a shorter time, and there was no need to mention that power consumption was also greatly reduced. '***
Securing massive amounts of big data
No matter how great intelligence is, if it's not learned, it's like natural wisdom. What's more, in the case of artificial intelligence, this is especially true. And it is the information in our human world that is necessary for learning this kind of artificial intelligence. Extensive digitized databases of search results, Wikipedia, web cookies, and tracking data accumulated over decades since the advent of the Internet make artificial intelligence smarter. ****
In particular, with the advent of deep learning technology, the importance of large-scale data is attracting more attention as a key resource for improving AI performance. It is now possible to obtain results that are more accurate as the amount of data increases.
Furthermore, in recent years, emphasis has been placed not only on the amount of data but also on securing optimized data suitable for AI learning. For that reason, governments and major agencies are also working hard to build high-quality databases. Big data platforms and data construction projects for AI learning, such as the Digital New Deal project being promoted by our government, also fall under this category.
An AI integrated platform operated by the Korea Institute for Intelligent Information Society (NIA), sources
What's more “The data we generate is doubling every year, and it is predicted that in 10 years there will be 150 billion network sensors, more than 20 times the Earth's population.”I say: ***** And this huge amount of data will make AI learn more, faster, and become more advanced. AI, which has now begun to learn on its own, is rapidly improving its ability to understand the world and interact with humans.
A better algorithm
Alexnet, which opened a new breakthrough in AI research that once fell into a dark period, is a prime example.
Alexnet appeared “ILSVRC (ImageNetLarge Scale Visual Recognition Challenge) is a competition that evaluates the accuracy, speed, etc. of an algorithm by recognizing images from a given large set of images using data. Algorithms with a shallow architecture (shallow architecture) won the championship until 2010 and 2011, but they showed an error rate of about 26%. There was a saying that reducing the error rate by even 0.1% would win, so it was very difficult to reduce the error rate with an algorithm based on a shallow structure. However, in 2012, deep learning experienced a revival when Jeffrey Hinton's team's AlexNet won the championship with overwhelming accuracy, reducing the error rate from about 26% to 16% due to deep architecture (deep architecture). '******
Changes in error rates since Alexnet appeared in 2012, sources
Since then, to today “Most algorithms are shared through global platforms (GitHub, Kegle). It creates synergy among members participating in the ecosystem by increasing the convenience of additional research through algorithm and code sharing. 'It's *******. Through this, AI models are being studied in a wider variety of ways, applied to the actual market, and advanced.
OpenAI GitHub, sources
As mentioned above, the current AI era was made possible by computing power, securing vast amounts of big data, and better algorithms. For the time being, AI technology will continue to evolve based on the synergy of these three key elements. In fact, new AI-based services that were previously only imaginable, such as AI interpreters, AI robots that interact with humans, AI facial recognition, and deepfakes, have become a reality.
Finalizing, into the 'human with artificial intelligence' era
AI has been studied for a long time and has achieved revolutionary development thanks to advances in computing power, big data, and algorithms. It is now used all over the world, and it has become an essential presence. It's hard to even accurately predict what else will become a reality in the future through AI.
However, at one point, the appearance of Deep Blue and AlphaGo raised doubts about human needs. However, as you know, regardless of the outcome of the confrontation with AI, humans are still playing chess and Go. This is because I saw AI not as a competitor to humans, but as a tool that can further enhance human abilities.
In fact, since then, in chess as well as Go, AI has been used to train players at a higher level. Also, I don't think this is a story limited to chess or Go. Even now, many companies and researchers are working to develop various AI technologies, and related industries are also developing at a rapid pace.
Finally, chess champion Kasparov, who lost against Deep Blue, once again quotes a remark to finish the series. “Don't be afraid of artificial intelligence machines; cooperate with them. (Don't Fear Intelligent Machines. Work with them.)” I think now is the time to prepare for a better future where AI coexists rather than having vague fears about AI.
References
[1] The activation of AI depends on data, algorithms, and computing power https://www.koit.co.kr/news/articleView.html?idxno=79799
[2] The Three Breakthroughs That Have Finally Unleashed AI on the World https://www.wired.com/2014/10/future-of-artificial-intelligence/
[3] How has artificial intelligence (AI) developed, and the history of artificial intelligence https://blogs.nvidia.co.kr/2016/03/13/history_of_ai/
[4] https://ko.wikipedia.org/wiki/인공지능#역사
Good content to watch together
[AI Story] Artificial Intelligence Challenging Humans (1) Deep Blue (Deep Blue)[AI Story] Humans vs. Artificial Intelligence (2) Watson (Watson)[AI Story] AlexNet (AlexNet) opened the era of human vs. artificial intelligence (3) deep learning[AI Story] Humans vs. Artificial Intelligence (4) Pepper (Pepper), a robot that reads emotions [AI Story] Humans vs. Artificial Intelligence (5) AlphaSGo (AlphaSGo) calculating intuition
