IELTS Academic Reading Practice Test 154

The heart beats because a small patch of specialised muscle in its upper right chamber, the sinoatrial node, fires an electrical impulse about once a second. That impulse spreads through the atria, the two upper chambers, which push blood into the ventricles below. It then crosses a narrow bridge of tissue, the atrioventricular node, and travels along conducting fibres to the ventricles, whose contraction drives blood around the body. If this pathway is interrupted by disease or by damage during surgery, the ventricles may beat too slowly, or miss beats altogether, and the brain is starved of blood, causing dizziness and fainting. A pacemaker watches for such gaps and fills them with a pulse of its own.

Electricity was used to revive stopped hearts long before anyone understood why it worked. In the nineteenth century, physiologists showed that a current applied to an animal's heart could make it contract, and a few clinicians tried the technique on patients who had collapsed. The experiments were crude: the equipment was bulky, the shocks were painful and nobody could control how the heart responded. For the first half of the twentieth century, a patient whose heart beat too slowly had little beyond bed rest.

The first machine designed for this purpose was built by Albert Hyman, an American cardiologist, in 1932. Hyman's apparatus was powered by clockwork: a hand-wound spring motor turned a small generator that delivered pulses through a needle inserted into the heart muscle. He called it the artificial pacemaker, coining a term that has survived, and claimed that it could revive patients whose hearts had stopped. His colleagues, however, were sceptical, and the machine was never widely used; it is remembered chiefly as a demonstration that the idea was possible.

In the late 1940s, surgeons in Toronto, including Wilfred Bigelow and John Callaghan, cooled patients to slow their metabolism during heart operations. The cold worked, but it also slowed the heart so much that it sometimes stopped. Bigelow turned to John Hopps, an electrical engineer at the National Research Council of Canada, who built a machine that delivered regular pulses to the heart. Hopps's pacemaker used vacuum tubes and drew its power from the mains, which made it the size of a filing cabinet. It was never implanted: electrodes were attached during an operation and the patient stayed tethered to the machine.

In 1952 Paul Zoll, a cardiologist in Boston, showed that the heart could be paced without opening the chest. He placed metal electrodes on the skin and passed short, powerful pulses between them. The technique could restart a heart that had stopped, and it was used in emergencies for years afterwards. The shocks, however, were extremely painful, they caused burns and muscle spasms, and the heart did not always respond. Pacing of this kind could only ever be temporary.

In 1957 a surgeon at the University of Minnesota, C. Walton Lillehei, was losing young patients after open-heart surgery because the conducting pathway in their hearts had been damaged. When the hospital's equipment failed during a power cut, Lillehei asked Earl Bakken, an electrical engineer who ran a small repair business in Minneapolis, for something more reliable. Bakken produced a pacemaker the size of a paperback book, powered by rechargeable batteries and using transistors instead of valves. Patients could wear it on a belt and walk about. His business later became Medtronic, now one of the world's largest makers of medical devices.

Wearing a device was still unsatisfactory, and the race to implant one was won in Sweden. In October 1958 the surgeon Åke Senning operated at the Karolinska Institute in Stockholm on Arne Larsson, whose heart was failing. The device he used had been designed by Rune Elmqvist, an engineer, and was sealed in a case of epoxy resin. It worked for only a few hours before failing, and a second implant lasted a little longer, but Larsson survived. He received more than twenty further pacemakers and lived into his eighties, outliving both men.

In 1960 the surgeon William Chardack and the engineer Wilson Greatbatch implanted a pacemaker in a patient in Buffalo, New York, and their device proved durable enough for regular use. Its zinc-mercury cells lasted two years at most, however, so replacement operations were frequent. The answer was a new kind of battery. Greatbatch developed a cell using lithium and iodine which produced a very small current but kept producing it for years; by the mid-1970s such cells were standard. They gave pacemakers a working life of ten years or more.

Later decades brought refinements rather than revolutions. Devices that speed up during exercise, using sensors that detect movement or breathing, appeared in the 1980s. Since the 1990s a related technique, biventricular pacing, has treated some forms of heart failure by stimulating both ventricles at once. More recently, engineers have made devices compatible with magnetic resonance imaging, which patients with implants were long forbidden to undergo. They have also shrunk the system into a capsule a few centimetres long that is fixed directly into the heart muscle. This removes the wires, or leads, that were for decades its most fragile part.

Pacemakers are now among the most common implanted devices, and hundreds of thousands are fitted each year. Modern versions record the heart's activity and can be checked remotely, so a clinician can spot a failing battery without the patient leaving home. Yet the principle has not changed since Hyman's clockwork motor: watch for a missed beat, and supply one.

Computer vision is the branch of computing concerned with extracting information from images. To a machine, a photograph is nothing more than a grid of numbers, each recording the brightness of a pixel in red, green and blue. The task of recognising a face or a pedestrian in that grid is remarkably hard. The same object can appear in a different pose, at a different distance, in shadow or in bright sunlight, and a slight change of viewpoint alters every number in the array. Humans manage this effortlessly, which made the problem look easier than it is. Early researchers, in the 1960s, confidently expected that a machine able to see would be built within a generation.

For two decades the work was dominated by hand-crafted rules. Programmers looked for edges, where brightness changes sharply, and for corners, and then tried to assemble these features into shapes. A more successful approach, developed in the late 1990s, described small patches of an image by the pattern of light and dark within them, in a way that did not change when the object was rotated or resized. The scale-invariant feature transform, as it was called, allowed one photograph of a building to be matched with another taken from a different angle. Such methods were reliable only within narrow limits, however, because the rules had to be written by someone who already knew what to look for.

The modern approach reverses this logic: instead of telling the computer what to look for, engineers let it work out the rules from examples. A convolutional neural network is built from layers of artificial neurons, each connected to a small patch of the layer below. A filter, a small grid of numbers, slides across the image and computes a weighted sum at every position, producing a map of where a particular pattern occurs. In the first layers these filters come to respond to edges and colour boundaries; deeper layers combine them into textures, curves and eventually whole objects. Between layers the resolution is reduced, so that later stages see a smaller, more abstract version of the picture. The network begins with random filter values and adjusts them, millions of times, until its answers match the labels attached to the training images.

The breakthrough depended on data as much as on mathematics. In 2009 a team led by the computer scientist Fei-Fei Li released ImageNet, a collection of millions of photographs that had been labelled by hand and sorted into thousands of everyday categories. An annual competition was established in which programs attempted to classify images they had never seen. In 2012 a network called AlexNet, built by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton, won by a wide margin, cutting the error rate almost in half compared with the best of the older methods. Its design was not wholly new, but it exploited two developments that earlier researchers had lacked: very large labelled datasets and the parallel processing power of graphics cards.

Within a decade computer vision had moved from the laboratory into daily life. When a photograph is uploaded to a social network, a chain of filters decides which faces it contains so that they can be tagged. Postal sorting machines read handwritten postcodes, and factories use cameras to inspect parts for cracks too small for a human eye to catch at speed. In medicine, systems trained on thousands of retinal scans can screen for the eye damage caused by diabetes, and in some countries such software has been approved for use without a specialist reviewing every image. Cars use vision to read road signs and to detect pedestrians, while agricultural machines identify weeds so that herbicide is sprayed only where it is needed.

The limitations are equally instructive. A network learns whatever regularities are present in its training data, including the prejudices of whoever chose the images. A widely cited study published in 2018 found that commercial facial analysis systems made far more mistakes on darker-skinned women than on lighter-skinned men. The datasets used to build them had contained relatively few such faces. Performance also deteriorates when the world looks different from the training set: a system trained on photographs taken in daylight may fail in rain, fog or at night. Objects that are partly hidden or seen from an unusual angle still cause errors. Researchers have also shown that changing the colour of a handful of pixels, in a way no person would notice, can persuade a network to attach a completely wrong label to an image.

Current research attacks these weaknesses from several directions. One is to reduce the dependence on hand-labelled data by using self-supervised learning, in which a model is trained to predict part of an image from the rest and so teaches itself about the structure of the visual world. Another is the vision transformer, a design borrowed from language processing that weighs the relationships between all parts of an image at once rather than examining it piece by piece. Engineers are also combining cameras with depth sensors and radar, so that a vehicle knows how far away an object is rather than merely that something is there. Smaller, more efficient networks now run on a phone, which means that images can be analysed without being sent to a distant server.

None of this amounts to seeing in the human sense. A network that identifies a chair has no idea what a chair is for, and it cannot reason about what it has not been trained on. The most useful description of these systems is that they are powerful pattern-matching tools, extraordinarily good at some narrowly defined tasks and unreliable outside them. That is a considerable achievement, but it is not the general-purpose vision that the pioneers of the 1960s promised.

Geoengineering is an awkward word for an awkward idea: deliberate, large-scale intervention in the Earth's climate to counteract the effects of greenhouse gases. It covers two very different families of technology. The first, carbon dioxide removal, takes the gas out of the atmosphere and stores it somewhere else, in forests, in rocks or deep underground. The second, solar radiation management, leaves the gas where it is and instead reflects a little more sunlight back into space. The distinction matters, because one addresses the cause of climate change, however slowly, while the other only masks the symptom.

Solar radiation management attracts attention because it could work quickly and cheaply. The most discussed version copies what large volcanoes do. When Mount Pinatubo erupted in the Philippines in 1991, it injected sulphur dioxide into the stratosphere, where it formed a haze of tiny droplets that reflected sunlight; global average temperatures fell by roughly half a degree for a year or two. Delivering a similar haze deliberately, using aircraft or high-altitude balloons, would in principle be far cheaper than rebuilding the world's energy systems, and its cooling effect would be felt within months rather than decades.

Other proposals in the same family include brightening low clouds over the ocean by spraying them with fine seawater droplets, which would make them reflect more sunlight, and thinning the high cirrus clouds that trap heat beneath them. Computer models suggest that each could lower temperatures, but the models are coarse, and the regional consequences, for rainfall in the Sahel, for the Asian monsoon, for the recovery of the ozone layer, remain uncertain. No one can yet say which countries would gain and which would lose.

Carbon dioxide removal is less dramatic and far slower. Trees absorb carbon as they grow, but a forest can be lost to fire or disease, and the land it occupies might be needed for food. Bioenergy with carbon capture, in which crops are burned for power and the resulting gas is buried, stores carbon but competes for the same land and water. Enhanced weathering spreads crushed silicate rock on fields so that it absorbs carbon as it breaks down. Each of these approaches faces questions of scale, cost and verification: how do you prove that a tonne of carbon really has been stored for a century?

Direct air capture, the most engineered of the options, draws air through a chemical absorbent and then releases the carbon dioxide for storage. Its advantage is that it needs little land and can be sited where the geology suits; its disadvantage is energy and price. A plant opened in Iceland in 2021 was designed to capture about 4,000 tonnes a year, while the world emits roughly forty billion tonnes. Current costs are usually quoted in the hundreds of dollars per tonne, and they must fall sharply before the technology matters at all.

The most serious objection to solar radiation management is what would happen if it stopped. A cooled planet is not a repaired one: the greenhouse gases remain, and the heat they trap is being held back only by the reflective layer. If a programme were interrupted by war, economic crisis or political disagreement, temperatures would rise very rapidly towards the level they would have reached anyway. Species and societies would then have to adapt at a speed they never evolved for. The effect has been called termination shock.

A second objection is moral hazard: that the mere promise of a technological fix will weaken the resolve to cut emissions. This is an argument about politics rather than physics, and the evidence is thinner than it is usually assumed to be. Some studies of public opinion find that people who learn about geoengineering become slightly less inclined to reduce their own energy use; others find no such effect. That the risk is unproven does not make it imaginary, and the pattern would be hard to detect before it had done its damage.

Meanwhile there is almost no governance. No treaty governs the stratosphere, and no institution has the authority to decide whether a country or a wealthy individual may alter the climate for everyone. A small Harvard experiment that planned to release a few kilograms of material from a balloon was called off in 2021 after objections from indigenous groups, without answering any of the questions it was designed to address. Reports from the Royal Society in 2009, the US National Academies in 2021 and the United Nations Environment Programme in 2023 all reached a similar conclusion: research is needed, and rules are needed before research turns into deployment.

Where does that leave the argument? I accept that cutting emissions must remain the priority, and that solar radiation management is no substitute for it. But the choice is not between researching geoengineering and living in a stable climate; it is between understanding the options before a crisis forces someone to use them, and discovering their consequences in the middle of one. Carbon removal deserves public money where it can be shown to work and to be measured, and solar radiation management deserves cautious, transparent, publicly funded research, not deployment, and not silence.