Based on materials from theverge.com
Each of us has ever encountered a psychological phenomenon: take any word and repeat it more often, and in the end it will lose its significance, it will simply turn into a sequence of letters. And for many of us, the combination of the words 'artificial intelligence' has met just such a fate. AI is an eyesore to our eyes wherever technology is present, from televisions to toothbrushes, and yet the expression 'artificial intelligence' has never been so devalued.
It doesn't have to be that way.
Tellingly, while the term 'artificial intelligence' itself is certainly undoubtedly misused, the technology itself is spreading like never before, for both good and bad purposes, in both health and arms. AI helps people create music and books, examines your resume, evaluates your credit history, and finally processes the photos captured with your phone. In general, he makes decisions that affect your life, whether you want it or not.
It's hard to get used to all the hype that tech companies and advertisers are creating around AI. Take, for example, the Oral-B Genius X toothbrush, one of many devices unveiled at CES this year and claimed to have artificial intelligence capabilities. But if you delve deeper into the press release, it turns out that the brush only provides you with simple information about whether you have been brushing your teeth for a long time and in the right places. Yes, smart sensors are built into the device, which do their job, determining exactly where the brush is in your mouth. But call it artificial intelligence? ..
And where advertising noise dies down, there is misunderstanding. The press is busy with their 'research', and every murky story involving artificial intelligence is invariably accompanied by a picture of the Terminator. And often all this is misleading even as to what artificial intelligence is. For amateurs, these are complex matters, and in the minds of people, the modern version of artificial intelligence often merges with the version that is most familiar to them: a conscious computer from a fantasy work, which is many times smarter than a person. Experts, in turn, call this phenomenon “strong artificial intelligence”, and if we ever manage to create something like this, it will happen in the distant future. Until then, we will continue to exaggerate the intelligence and capabilities of artificial intelligence.
So what is AI? An Oral-B smart toothbrush, an autonomous courier robot, or something else?
And so it makes more sense to talk about machine learning, rather than about AI. This is a branch of artificial intelligence, and it includes almost all methods that have a significant impact on our present life (including what is called deep learning). The term doesn't have the mystical fascination with the concept of 'artificial intelligence', but it is much more useful when it comes to explaining how technology works.
How does machine learning work? Over the past few years, many explanations have appeared, but the most comprehensive one lies in the term itself: machine learning is what allows computers to learn on their own. What this means is a much more important question.
Let's start with the problem statement. Let's say you need to create a program that should recognize cats (cats for some reason). You can do this the old-fashioned way by programming clear rules such as 'cats have pointy ears' and 'cats are fluffy'. But how will the program react to the picture with the tiger? Programming each criterion is very time consuming and you need to define each of these criteria, such as 'sharpness' and 'fuzziness'. It is much better to let the machine learn by itself. You provide her with lots of pictures of cats, and she looks through them in search of patterns in what she sees. At first, the machine connects the dots in an almost random fashion, but you try again and again with the best results. And over time, she is getting better at identifying what is a cat and what is not.
A predictable explanation, isn't it? You've probably read this before, so excuse me generously. But it is important not only to read, but also to understand the meaning. What could be the implications of training the decision-making system in this way?
So, the biggest advantage of this method is also the most obvious: you don't really need to program it. Of course, you have to do a lot of work improving the way the system processes data and find smarter ways to absorb this information, but you don't tell the system what to look for. This means that she can find ways that the human mind may not even suspect or not think about it in the first place. All a program needs is data – ones and zeros – and therefore it can be taught anything, because the modern world is filled with data. Relatively speaking, if you think of machine learning as a hammer, you can say that the digital world is full of nails that are ready to be driven in the right places.
Machines that learn on their own can achieve incredible success, as is the case with DeepMind's AI-based go system.
But let's also think about the costs. If you are not explicitly teaching a computer, how do you know how it makes decisions? Machine learning systems cannot explain how they think, and therefore your algorithm can work well for bad purposes. Likewise, since all that a computer has is the data that you have provided it, it may have a distorted view of the world, or it may only be useful for a narrow range of tasks that are similar to what it has done before. The machine does not have the common sense that we expect from a person. You can create the best cat recognition software in the world, but it never tells you kittens are unable to drive motorcycles.
Teaching a computer to learn on its own is a great way to make life easier, but like any simplification, it's not perfect. Yes, there is intelligence in AI systems, if you like to call it that, but it is not organic intelligence, and it does not live by human laws. Ask yourself: How smart is the book? What experience does a frying pan have?
So where are we today in the field of artificial intelligence? After years of headlines promising huge new breakthroughs (and they still do), a number of experts decided that we had reached something of a plateau. But this is not a hindrance to progress. From the point of view of science, the knowledge already existing in our country presupposes many ways of development. And in terms of product applications, we only see the tip of the 'algorithmic iceberg'.
Kai-fu Lee, a venture capitalist and former AI researcher, describes the current situation as an “era of adoption” where technology begins to “spill out of the lab into the world.” Benedict Evans, another venture capitalist, compares machine learning to relational databases, the type of enterprise software that took off in the 90s and changed the entire industry, but has now become so commonplace that it will surprise the reader to claim it was revolutionary. What both are pointing out is that we are living in a moment where AI will quickly become the norm. “Obviously, almost everyone will have machine learning somewhere inside, and no one will bother,” Evans says.
He's right, but we haven't got it yet. Here and now, artificial intelligence – that is, machine learning – is still something new, inexplicable, or insufficiently unexplored. However, technology is changing the world around us in order to become so mundane in the future that we stop noticing it.