AI in 2019: the current state of affairs
This translation is part of The Real-World AI issue .
The well-known fact is that if you repeat any word many times, it will eventually lose all meaning, turn into phonetic nothing. This is the reason why for many of us the term “artificial intelligence” has long ceased to mean something. AI is now almost everywhere, from your TV to the tooth cheek, but this term has never meant so little.
Something went wrong.
Despite the fact that the term “artificial intelligence”, no doubt, is used incorrectly, the technology associated with it is still developing very actively - and this is both good and bad. Technology penetrates healthcare and military affairs , it helps people composemusic and writing books , she checks your resume and judges your creditworthiness , and also improves photos taken on your mobile phone. In short, technology changes your life, whether you like it or not.
At the same time, it is rather difficult to understand what the AI is being discussed by technology companies and marketers. Take, for example, the Oral-B Genius X toothbrush, which, among many other devices with “AI added,” was introduced at CES. If you carefully examine the press release of the company, it becomes clear that this brush simply helps to determine the correct duration of the procedure for brushing your teeth, processing exactly where you need it. Yes, there are several “smart” sensors in the brush, which provide the functionality of the device, but to call it artificial intelligence is just silly.
Even in the case when there is no excessive hype, there is a misunderstanding of the essence of the term. For example, coverage of a certain event in the press is accompanied by an illustration with the Terminator., and it happens quite often. Sometimes authors get confused about what artificial intelligence is. All this is difficult for someone who is not an expert, so most people define AI as a computer with a mind that is many times smarter than a person. Experts call this hypothetical artificial intelligence a strong form of AI, but its creation is a distant future. In the meantime, do not exaggerate.
What is AI? Clockwise, starting at the top: character from the movie Metropolis, Oral-B toothbrush, autonomous robot courier.
It is much better to call what we have “machine learning” rather than AI. This technology has a very strong influence on the world (along with deep learning). These terms have nothing mystical, like artificial intelligence, and they describe what technology is much better.
How does machine learning work? Over the past few years I have read and looked through many explanations. And one of the most useful is the following: machine learning allows computers to learn on their own. But what does it mean? Another big question.
Let's start with the problem. Let's say you want to create a program that recognizes cats (always cats for some reason). You can try to implement this "old-fashioned" method, asking the machine rules like "fluffy seals". But what will the program do if you show her a picture of a tiger? Developing each rule manually is a long process, and you will have to describe all the nuances that can come to mind. It is best to give the machine the opportunity to learn the most. So, you give her a huge collection of photos of cats, and the car already determines the patterns, and independently. At first this is a primitive process, which is then complicated.
Actually, this is so clear. In fact, you’ve probably already read similar explanations before, and please forgive me for that. But what is important is to understand how it works. What could be the consequences of learning such a system?
The main thing is that you do not need to program it. Yes, of course, there is a lot to adjust, optimize, it is necessary to improve the efficiency of data processing and all that, but you don’t tell the machine what to do. And the machine really works, highlighting the patterns that a person misses. Well, since the machine only works with ones and zeros, there are many areas where you can apply this technology, because the modern world is full of data. With a machine learning hammer in hand, you will be able to cope with the solution of many problems that need to be solved immediately.
But there are disadvantages here, think about them. If you can’t train a car, how do you know how it makes decisions? Machine systems do not know how to explain their course of solving a certain task. So it may well be that the algorithm is not valid. The car has no awareness of what it does - what we expect from a person. You can make the best car for recognizing pictures of cats in the world, but the machine will not explain to you that kittens do not know how to ride motorcycles, or that the seal is likely to be named Vaska and not Megalort Immortal.
Teaching a computer to learn on your own is great. But in AI systems there is no rationality. It is not a mind at all, not an organic mind, and there will be no machine playing by the rules of the people. You may ask the question “How smart is a book? What can clever fried bread do? ”
So what stage of artificial intelligence are we at? After all these years, with editorials that talked about the “imminent big breakthrough,” experts believe that we have reached the plateau .
However, progress is not a hindrance. With regard to research in the field of AI, there are a huge number of opportunities for studying, now we are only at the very top of the algorithmic iceberg.
Kai-fu Lee, a venture capitalist and an expert in artificial intelligence, describes the current moment as the “age of realization” when technologies start to get out of laboratories to the outside world. Benedict Evans compares machine learning with relational database technology — a type of enterprise software that revolutionized entire industries. But now it is commonplace. Most likely this will happen with machine learning - it will be everywhere, and no one will worry about it. But all this is a distant future.