No, AI is not a threat
We have all seen movies depicting AI as a threat. In Avengers 2, Ironman created Ultron so that it can help avengers to defend people from enemies. But Ultron went out of control and wanted to dominate people, becoming an enemy. But as a professional machine learning engineer, I do not think AI will dominate us. There are a lot of pessimistic opinions making you think that AI and machine learning are turning modern society into a dystopia. But in this article, I would like to demystify AI and show you its optimistic side.
The reason why people are so pessimistic is because they are unfamiliar with this new concept. When people are shocked by how strong AI can be, they also want to question whether it is a threat to us. So, this kind of fear spreads out in mass media very fast and even stirs up anxiety in mass media. Moreover, not just for AI, there are opinions and conspiracy theories demonizing every kind of advanced technology, such as “Covid vaccines are dangerous because microchips are inserted in your body” and “GMO is a threat to our health because it contains the genes of virus”. All modern technology falls victims for people’s anxiety. Mass entertainment can play around with it. For example, the Wachowskis, who made the Matrix series, are not experts in computer science but professional entertainers. They can make an entertaining cyberpunk story and need a villain for that. So, the anxiety-arousing “intelligent machines” fit that role. But please do not be afraid of AI even though they are pictured as villains because when you know the nature of AI, you will find that AI will never hold the grudge against mankind.
First, I would like to explain machine learning from my perspective. Unlike humans, machines cannot grow their intelligence while learning because they only do statistics. The term “machine learning” was very confusing to me at the first glance. When I was taking the data structures course for computer science, I need to learn sorting algorithms that make computers know how to sort data fast. I was wondering, “Why is this not called machine learning? Programmers are teaching computers to sort data.” In fact, all the programs we designed for computers are for them to learn to solve problems. However, sorting algorithms are not considered “machine learning”. So-called “machine learning” algorithms should be able to do predictions with statistical modeling by using large amounts of data. But I would rather say “statistical learning” instead of “machine learning” as a more accurate term because the essence of “machine learning” is still statistics. It is the statistics behind those algorithms that uses large amounts of data to calculate probability out of it. Statistics is not a threat; it has been invented long time ago and serving us alongside with human civilizations’ evolution. For example, weather forecast has been with us for a long time. Climatologists use data measured from weather stations to predict the probability of rainfall in the following days. Statistical learning’s fundamental principle is just like weather forecast. With the advance of technology, there are more data and better computing power so that people can use statistical learning for better accuracy, higher speed, and more diverse purposes.
Second, I would like to explain deep learning from my own perspective. Again, the recent deep learning is not going to threaten human because its nature is calculation rather than thinking. Deep learning is considered as a sub-field of machine learning by using artificial neural networks, which is inspired by human brains. Artificial neural networks, or just neural nets in short, have drastically improved the performance of AI, but also spawned fear among the public. People fear that AI will be developed to be super-intelligent. Eventually, AI can think like humans and defy humans. This kind of concern is widely depicted in mass entertainment such as the movie I, Robot; and TV series Westworld. However, deep learning is still way different than human brains. We are not using artificial neural networks to build an electronic brain. The “neurons” in artificial neural networks are not electronic neurons resembling humans’ neurons, but merely the place where linear models and non-linear models are combined so that the calculation can approximately approach any kinds of functions according to the Universal Approximation Theorem. To explain it in a simpler way, artificial neural networks are rounds of matrix multiplication. AI cannot think like humans.
Does AI really outsmart humans? Yes, but only because humans want AI to be smart so that it can serve people. AI will not learn the evil thoughts itself as in those anxiety-arousing movies and dramas because there are always engineers behind these algorithms. There are claims that because the god created humans, and humans also want to be creators like the god; humans created AI. The horror movie series Prometheus & Alien: Covenant tells exactly this story. In this movie series, humans created the AI stronger and more cunning than humans, killing both humans and humans’ creators. However, not all algorithm engineers believe in Creationism. The reason why farmers raise pigs is not because farmers want to be pigs’ gods, but because pork can feed people. And we algorithm engineers are training AI for the same reasons as farmers. We use machine learning not because we want to be creators, but because we want those algorithms, which can do calculation for big data, to solve our problems. For example, I work in e-commerce field. I design my algorithms to predict hot-selling items for my marketing colleagues. There are millions of items on our platform. Because our people cannot look at all the information for millions of items one by one. I used statistical algorithms to rank those items, so that my marketing colleagues only need to check the top 1000 items on the chart. It is because processing large amounts of data is an exhausting and repetitive work, we let statistical learning algorithms share our burden and work efficiently. This is just like that we use trucks and cranes to deliver large amounts of materials in modern era to build skyscrapers, instead of using large amounts of human slaves to build pyramids like Egyptian pharaohs in old days.
Although AI itself is not a threat, it can be used for vile purposes when it falls into wrong hands. The recent deepfake is a bad example of this, which can be used for replacing one’s face with a computer-generated face in videos. Machine learning makes those videos convincingly real, but also hard to identify whether they are trustworthy. Jordan Peele’s video is a wake-up call for Deepfake’s detrimental way of use. Deepfakes of politicians can be a threat to the democracy. Deepfakes of celebrities can be a threat to their reputations. Deepfake perjuries can be a threat to the justice. But I will not blame AI but blame the ones who use deepfakes for ill purposes. Before deepfakes, people use Photoshop to make fake images of others. And before Photoshop people use paint and brushes to forge fake paintings. Fakes has been along with our civilizations for long. And Deepfakes bring it to a new level. Moving forward with modern technology, we need to be vigilant with new modus operandi. We should always question if sources are trustworthy.
There are also a lot of other things I want to cover with AI such as recommendation algorithms and employment. These topics are huge so I will make stand-alone explanations for each of these in future. But anyway, AI will not dominate us because its nature is statical learning, which will significantly boost our efficiency in modern society. Please stay optimistic about our future with AI.
