In this blog post, we will look at the advances in artificial intelligence revealed through the match between Lee Se-dol and AlphaGo, and what it means for human intelligence and the future of society.
In March 2016, the match between Lee Se-dol 9th Dan and AlphaGo ended with AlphaGo winning 4 games and Lee losing 1. Before the match began, the majority of the public believed that we still had a long way to go, but after the match ended, people were satisfied with Lee’s victory in the fourth game and began to acknowledge the advancement of artificial intelligence. Now, people are worried that scenes from science fiction movies may become a reality. This is because we do not fully understand the current level of AI development, and it is difficult to predict the speed of future developments. However, contrary to the concerns of many, the age of AI began long ago. The search engines, navigation systems, spam filters, calculators, and even the automatic temperature control in our refrigerators that we use every day are all forms of AI. However, these are far from the images that come to mind when we think of AI. John McCarthy, who first coined the term AI, said, “When people stop calling something artificial intelligence, it has become intelligent.” “People don’t call it artificial intelligence when it’s implemented.” In other words, there is already a lot of artificial intelligence around us, but we are simply unaware of it. So what exactly is artificial intelligence?
Artificial intelligence is the ability of computers to mimic human intelligent behavior. Therefore, artificial intelligence is a fairly broad concept, and people divide it into three stages. The first is artificial narrow intelligence, which refers to AI that performs well only in specific fields. It performs amazingly well in specific fields, but its disadvantage is that it cannot solve any problems outside of its given scope. The most easily accessible and currently only implemented examples of AI around us are automatic navigation in airplanes, voice recognition in mobile phones, and automatic temperature control in air conditioners. The AI around us mentioned above is also narrow AI, and AlphaGo, which defeated Lee Sedol in a game of Go, is also narrow AI. No matter how AlphaGo is designed with new techniques and has fast processing speeds, it cannot perform commands such as measuring the indoor temperature and automatically adjusting the air conditioner.
If weak AI is what we have now, and even AlphaGo is weak AI, won’t AI eventually surpass all human abilities? No, it won’t. The reason AlphaGo’s achievement is so great is that it proved one of the ways in which weak AI can lead to strong AI (Artificial General Intelligence). Strong AI is AI that can compete with humans in all aspects. In other words, strong AI must be able to perform all the intelligent tasks that humans can perform. It must be able to judge the quality of a painting, identify animals from photographs, and engage in abstract and intuitive thinking. Humans are capable of intuitive thinking because the brain is an extraordinary organ that can perform more than 10^16 calculations per second. Humans have already created computers with a processing capacity of 3.4 quadrillion cps (calculations per second). So, is intuitive thinking possible? No, it is not. Another reason why humans are capable of abstract and intuitive thinking is because of their vast experience and accumulated knowledge. The machine learning technique applied to the design of AlphaGo is a method of improving AI performance based on experience. Machine learning, a technology that develops based on big data, has been used as a model for creating weak AI, but in the future, it is expected to lead the development of strong AI and accelerate its emergence.
Thanks to AlphaGo, humanity has seen the possibility of AI reaching human levels. So, is it possible to create AI with intelligence that surpasses that of humans? The term that encompasses this concept is “artificial superintelligence.” Artificial superintelligence refers to AI with intelligence that far surpasses that of humans in all areas, including science, knowledge, and social skills. What does it mean to surpass human intelligence? Nick Bostrom, who defined superintelligence, categorizes the role of superintelligence into three main types. The first is Oracle mode, which can provide accurate answers to almost all questions about “possibilities.” The second is Genie mode, which can understand and execute advanced commands. The last is Sovereign mode, in which AI can perform tasks and make decisions on its own in the process. However, these are only predictions, and there is still no answer as to what superintelligence can do. It is generally accepted that human intelligence cannot understand superintelligence. People often wonder if we can simply turn off the power if AI becomes so advanced that it poses a threat. However, this is like saying that we should starve spiders to death because they do not understand humans and cannot give them cobwebs. Unlike the emergence of strong AI, there are various opinions within academia regarding the possibility of superintelligence. However, following the logic that superintelligence cannot be created makes it impossible to explain, so let’s follow the logic of those who believe that superintelligence will emerge. The emergence of superintelligence, as they say, will be achieved through recursive self-improvement. In other words, if strong AI appears and is programmed to improve itself, it will reach superintelligence in an instant. According to a 2013 survey by Nick Bostrom and Vincent C. Muller, artificial intelligence experts responded that strong artificial intelligence would be realized around 2040 on average, and that there was a 10% chance that it would take two years to make the leap from strong artificial intelligence to superintelligence, and a 75% chance that it would be achieved within 30 years. However, what is certain is that once superintelligence is realized, it will replace human intelligence. This will be humanity’s last invention and challenge, and we must prepare ourselves thoroughly before diving in.
Technological advancement is progressing rapidly and exponentially. Within a certain scope of problems, we are already surrounded by weak AI that has surpassed humanity. Strong AI with human-level intelligence in all aspects is expected to emerge within the next 15 to 20 years. Finally, although there are still differing opinions on the feasibility of superintelligence that surpasses humanity in all aspects, it is approaching faster than we expected. There are countless AI around us, and rather than becoming accustomed to them and insensitive to the speed of AI development, we must keep a close eye on them and prepare for future developments that will be on a completely different level from today.
