The 2016 match between Lee Sedol 9 and AlphaGo, which ended in AlphaGo’s victory, changed the perception of the development of artificial intelligence. This article explains the present and future of artificial intelligence and the possibility of superintelligence.
In March 2016, Lee Sedol 9 and AlphaGo played a match that ended with AlphaGo winning 4 games to 1. Before the match began, the public’s opinion was that it was “still a long way off,” but after the match, the mood changed to one of satisfaction with Lee Sedol’s succession in game 4 and recognition of the advances in artificial intelligence. Now, people are worried that scenes from science fiction movies will become reality. This is because we don’t understand exactly how far AI has come, and it’s hard to predict how fast it will continue to advance. However, contrary to many people’s fears, the era of AI has already started a long time ago. The search engines, navigation systems, spam filters, calculators, and even the thermostats in our refrigerators are all forms of artificial intelligence, but they’re far from what we think of when we think of artificial intelligence. Here’s a quote from John McCarthy, who first coined the term. “People stop calling something artificial when it becomes artificial.” In other words, there is a lot of artificial intelligence all around us, we just don’t recognize it. So what exactly is artificial intelligence?
Artificial intelligence is the ability of computers to mimic intelligent human behavior, so it’s a pretty big concept, and people divide it into three stages. The first is Artificial Narrow Intelligence, which is AI that’s only good at certain things. It’s amazingly good at certain things, but the downside is that it can’t solve anything outside of a certain range of problems. Some of the most common and currently only implemented AIs include autopilot on airplanes, voice recognition on cell phones, and automatic temperature control in air conditioners. The aforementioned AIs are weak AIs, and so is AlphaGo, the AI that won the game against Lee Sedol. Even if AlphaGo is designed according to new techniques and has fast processing speeds, it will be able to play Go, but it will not be able to measure the room temperature and automatically adjust the temperature of the air conditioner.
If the current state of the art is weak artificial intelligence, and even AlphaGo is weak artificial intelligence, doesn’t that mean that artificial intelligence will surpass all of humanity’s capabilities? No, it won’t. What makes AlphaGo’s achievement so great is that it demonstrates one of the paths from weak AI to strong AI (Artificial General Intelligence). A strong AI is one that can compete with humans in many ways. In other words, it should be able to do all the intelligent things that humans can do. It should be able to look at a painting and judge the quality of the work, or look at a picture of an animal and recognize it, and it should be able to think both abstractly and intuitively. Humans are capable of intuitive thinking because the brain is a remarkable organ that can perform more than 1 quadrillion (16 to the power of 10) calculations per second. We’ve already built computers with a processing power of 3.4 cps (calculations per second), so is intuitive thinking possible? The answer is no. Another reason why humans are capable of abstract and intuitive thinking is the accumulation of tons of experience. The machine learning techniques used in the design of AlphaGo are a way to increase the performance of AI based on experience. Machine learning, a technology that evolves based on the big data provided, has been used as a model for creating weak AI, but in the future, it is expected to be at the forefront of research into the development of strong AI, accelerating the emergence of strong AI.
Thanks to AlphaGo, humanity has seen the possibility of AI reaching human levels, but is it possible to create an AI with intelligence beyond humanity? Artificial Super Intelligence is a term that captures this concept. It refers to an artificial intelligence that is far superior to human intelligence in all areas, including science, knowledge, and social skills. What does it mean to surpass human intelligence? Nick Bostrom, the creator of the definition, categorizes superintelligence into three main roles. The first is Oracle mode, where it can give a precise answer to almost any “what if” question. The second is Genie mode, which can understand and carry out advanced instructions. Finally, there’s Sovereign mode, where the AI is capable of carrying out missions and making its own decisions along the way. However, these are just projections, and there are still no answers to the question of what AI is capable of, and it’s generally accepted that human intelligence is not capable of understanding AI. A common question is whether we should turn it off if it becomes a threat, but this is like saying that spiders don’t understand humans, so we shouldn’t give them webs and let them starve to death. Unlike the emergence of strong artificial intelligence, there is a wide range of opinions within the academic community about the possibility of such an AI. However, if we follow the logic that artificial intelligence cannot be created, it becomes inexplicable, so let’s follow the logic of the camp that believes that artificial intelligence will emerge. In their view, the emergence of AI is driven by recursive self-improvement, which means that once a strong AI emerges, it can be programmed to improve itself, and it will quickly reach AI. According to a 2013 survey by Nick Bostrom and Vincent C. Muller, AI experts, on average, believe that AI will be realized around 2040, with a 10% chance that the time required to make the leap from AI to superintelligence will be two years and a 75% chance that it will be reached within 30 years. What is certain, however, is that once AI is realized, it will eventually replace human intelligence. It will be our final invention and challenge, and we need to be very prepared for it.
Human technological progress is very fast, and it is increasing exponentially. We are already surrounded by weak artificial intelligences that have surpassed humanity in the scope of a given problem. Strong AIs with human intelligence levels in all areas are expected to emerge in about 15-20 years. Finally, superintelligence, which transcends humanity in every way, is still debated, but it is coming sooner than we expect. Rather than becoming numb to the pace of AI advancement as we become accustomed to it, we should keep an eye on it and prepare for a future pace of advancement that is far different from today.
