IELTS Reading Practice Test 6: Free, with Answers

AHuman beings recognise faces, read handwriting and judge the distance of a moving car without any conscious effort, so it is easy to assume that seeing is a simple task. Even a small child can tell a dog from a cat in an instant, yet this was something that defeated computers for decades. For computers, it has proved to be one of the hardest problems in the whole of artificial intelligence. Today, phones unlock when they recognise their owners and cars can detect pedestrians, but these achievements are the result of more than fifty years of research, much of which ended in disappointment.

BThe early optimism is captured in a famous document. In July 1966, Seymour Papert, a researcher at the Massachusetts Institute of Technology, proposed that a group of students should spend the summer building 'a significant part of a visual system', one that could pick out objects in a photograph and name them. The project was intended as a practical exercise for the students, and it was assumed that most of the work could be done in a few months. Nearly sixty years later, researchers are still working on some of the problems it set out to solve.

CThe approach taken at the time was to describe the world to the computer in the form of rules. Programmers wrote instructions for detecting edges and corners, and then for combining them into shapes. As early as 1963, Larry Roberts, also at MIT, had shown that a computer could work out the three-dimensional shape of simple blocks from a photograph, by identifying their straight edges. But rules that worked for neat blocks on a table failed completely in the real world, where objects are partly hidden, lit from different angles and seen from countless points of view. A cat, after all, does not have straight edges. Each new situation required new rules, and the list of exceptions never seemed to end.

DA different idea had been around for decades: instead of telling the computer what to look for, let it learn from examples. In the late 1980s, Yann LeCun and his colleagues at Bell Laboratories in the United States built a network loosely modelled on the visual system of animals and trained it on thousands of examples of handwritten numbers. The system learned to recognise digits it had never seen before, and by the end of the 1990s a version of it was being used by banks to read more than ten per cent of all the cheques written in the United States. Few people outside the field noticed, however, and for many years most researchers regarded such networks as too slow and unreliable for more complicated images.

EOne reason they performed poorly was a lack of data. A system that is to recognise thousands of kinds of object needs to see millions of labelled pictures, and no such collection existed. In 2007 the computer scientist Fei-Fei Li began to build one. Her view, unfashionable at the time, was that the size of the collection mattered as much as the cleverness of the method. Many colleagues doubted that such a collection would be worth the enormous effort and expense involved. Her team gathered images from the internet and paid tens of thousands of people around the world, working online, to label them. The result, a database called ImageNet, was released in 2009 and eventually contained more than fourteen million images sorted into over twenty thousand categories. From 2010, research teams competed each year to see whose system could identify the objects most accurately.

FIn 2012, the competition produced a result that changed the field. A team from the University of Toronto, led by the graduate student Alex Krizhevsky, entered a network of the kind LeCun had pioneered, but much larger. Its best guesses were wrong only about fifteen per cent of the time, compared with twenty-six per cent for the nearest rival, a gap so large that it took many experts by surprise. Part of the team's success came from an unexpected source: they trained their network on graphics processors, chips originally designed to produce realistic images in computer games, which could carry out the necessary calculations far faster than ordinary processors. Within a few years, almost every research group had adopted the same approach.

GMachine vision is now part of everyday life, but its weaknesses have become clearer too. Researchers have shown that a few small stickers placed on a road sign can cause a system to misread it, a worrying discovery for the makers of self-driving cars. Other studies have found that some face-recognition systems make far more mistakes with women and with people with darker skin, largely because the images used to train them included too few such faces. A system, in other words, can only be as fair as the examples it has learned from. Researchers are now looking for ways of making such systems harder to fool, and of checking training collections for gaps before they are used.

A typical large supermarket now stocks tens of thousands of different products, and anyone buying a phone, a pension or even a pair of jeans faces a bewildering range of options. Over the past two decades, an influential idea has taken hold: that this abundance is not a blessing but a burden, and that beyond a certain point, more choice makes people less likely to choose at all and less satisfied with what they eventually select. The argument was popularised by the psychologist Barry Schwartz, whose book The Paradox of Choice appeared in 2004, and it has since become a commonplace of business advice and popular psychology. It is an appealing idea, partly because it matches the frustration many of us feel in front of a crowded shelf. The scientific evidence, however, is considerably less straightforward than its popularity suggests.

The study most often cited in support of the idea was carried out by Sheena Iyengar and Mark Lepper and published in 2000. On different days, shoppers in an upmarket grocery store in California were offered the chance to taste jams from a display containing either six or twenty-four varieties. The larger display attracted more attention, with sixty per cent of passing shoppers stopping to taste, compared with forty per cent for the smaller one. Yet only three per cent of those who stopped at the large display went on to buy a jar, compared with thirty per cent at the small one. The study itself was carefully designed, and its authors were cautious about their conclusions; the difficulty lies in what others have made of it, for it has often been presented as proof that offering fewer options is always better.

Subsequent research has not been kind to that simple version. In 2010, Benjamin Scheibehenne and his colleagues combined the results of fifty experiments on the effects of choice, and found that, on average, the effect of having more options was close to zero. Such a result does not prove that the original study was mistaken; it suggests instead that the effect varies considerably from one situation to another. Some studies had found that larger selections discouraged people; others had found the opposite, with people who were given more options buying more and reporting greater satisfaction. I see no reason to dismiss these more encouraging results, which are at least as reliable as the famous one.

This does not mean that the problem is imaginary. A later review, by Alexander Chernev and his colleagues in 2015, suggested that choice overload is real but appears only under particular conditions: when the options are complicated and difficult to compare, when people do not have a clear idea of what they want, and when they must decide quickly. A shopper who knows exactly which jam she likes is not troubled by a hundred jars; a first-time buyer comparing twenty mortgage offers, each with different fees and conditions, may well be. The useful question, in other words, is not how many options are too many, but in what circumstances choice becomes a burden.

Personality appears to matter too. Decades ago, the economist Herbert Simon coined the term 'satisficing' to describe the strategy of accepting the first option that is good enough, rather than searching for the best possible one. Schwartz and his colleagues later distinguished between 'satisficers' and 'maximisers', people who feel compelled to examine every option. In one study, university students who were maximisers found jobs with starting salaries about twenty per cent higher than those of satisficers, yet they felt less satisfied with the jobs they had chosen and with the process of finding them. By any objective measure, then, maximisers did not choose worse; they simply enjoyed their success less. Maximising, in this sense, depends less on the number of options available than on the standard a person sets for themselves.

There are also practical lessons for organisations that design choices for others. In one well-known case, an American company changed its pension scheme so that new employees were enrolled automatically unless they chose to opt out, instead of having to apply. The proportion of new employees saving for retirement rose from thirty-seven per cent to eighty-six per cent. Nobody was forced to save, and nobody's options were reduced; the company had simply decided what would happen to people who did nothing. Similar automatic arrangements have since been introduced for workplace pensions in several countries, including the United Kingdom. Such defaults are powerful, and for that reason they must be chosen with care, since a poorly designed default can trap people just as effectively as a good one can help them.

The conclusion I draw from this evidence is not that choice should be reduced, which would deprive many people of options they value, but that it should be better designed. Choice is valuable precisely because people differ, and a range of options allows each of them to find something that suits their own needs. Fewer, clearly described alternatives where the decision is complicated, sensible defaults where people are likely to do nothing, and a cultural willingness to accept that 'good enough' is often good enough: these are more promising responses than nostalgia for a world in which there was only one kind of jam.