Algorithms are everywhere. They play the stock market, decide whether you can have a mortgage and may one day drive your car for you. They search the internet when commanded, stick carefully chosen advertisements into the sites you visit and decide what prices to show you in online shops. As Uber and Waymo will tell you, they can be the subjects of legal arguments; they cause regulatory worries too (earlier this month a group of luminaries called for a ban on battlefield robots running algorithms designed to kill people). PageRank—the algorithm that powers Google’s search results—has made its inventors very rich indeed. Algorithmically curated “filter bubbles” may even affect the way a country votes. But what exactly are algorithms, and what makes them so powerful?
An algorithm is, essentially, a brainless way of doing clever things. It is a set of precise steps that need no great mental effort to follow but which, if obeyed exactly and mechanically, will lead to some desirable outcome. Long division and column addition are examples that everyone is familiar with—if you follow the procedure, you are guaranteed to get the right answer. So is the strategy, rediscovered thousands of times every year by schoolchildren bored with learning mathematical algorithms, for playing a perfect game of noughts and crosses. The brainlessness is key: each step should be as simple and as free from ambiguity as possible. Cooking recipes and driving directions are algorithms of a sort. But instructions like “stew the meat until tender” or “it’s a few miles down the road” are too vague to follow without at least some interpretation.
Algorithms are closely associated with computers and code. They do not have to be. Alan Turing, a British mathematician who did a great deal of pioneering work on how to treat algorithms with mathematical rigour, once wrote a fairly complicated chess-playing algorithm on paper. He tested it in a match against a friend, scanning down the list of instructions with every move and doing what his instructions told him. But, as Turing’s opponent conceded, humans generally find such repetitive, mindless work boring and frustrating (there was so much paperwork and arithmetic involved that it reportedly took about half an hour to play each move). Computers, though, excel at quickly churning through dull, repetitive tasks such as “add these two numbers”, “decide if this number is bigger than that one” and “store the answer over there.” It is, in fact, the only thing that they are capable of doing.
For that reason, computers have allowed humans to build—and execute—ever bigger and more baroque algorithmical constructs. And it turns out that, like Lego bricks, piling up enough simple instructions allows you to build far more intricate and interesting things than is apparent at first. Every computer program, from Chrome to Call of Duty to a climate model, is, at its root, nothing more than a big pile of algorithms being executed at high speed. In a nice bit of symmetry, some of the most advanced algorithms are not written by humans at all, but by other algorithms. Machine learning is a fashionable artificial-intelligence technique used to teach computers to do things that people can do, such as decode speech or recognize faces, but which humans cannot explain in a sufficiently mechanical algorithmic fashion. So a machine-learning algorithm does the translation work for them. It ingests lots of examples of the thing in question—spoken language, say, or pictures of faces—which have been labelled by humans. It then produces another algorithm that recognizes them reliably. Brainlessness, in other words, is no impediment to intelligence.
Source: the Economist Blog