IELTS Academic Reading Practice Test 129

Until the second half of the twentieth century, making a copy of a document was a slow and often messy business. Offices relied on carbon paper, which produced a single faint duplicate at the same time as the original was typed, or on machines that used wet chemicals and specially coated paper. For most people, the alternative was simply to write the text out again. The change came with a process that used no liquid at all. Its inventor was not an engineer at a large corporation but a patent lawyer working alone in a small room in New York.

Chester Carlson had grown up in poverty and had worked his way through college. By the mid-1930s he was working in the patent department of an electronics company in New York, and his days were spent copying out long specifications by hand or paying a typist to do it. Carlson had studied physics and chemistry as an undergraduate, and he became convinced that there must be a better way. His starting point was a phenomenon that was already well known: certain materials, including sulfur and selenium, conduct electricity far more readily when light falls on them than when they are in darkness.

This property, known as photoconductivity, suggested a way of forming an image without a camera. If a plate coated with such a material were given an electric charge in the dark, the charge would remain until light reached it. When light reflected from a page fell on the plate, the charge would survive only where the page was dark. A fine powder carrying the opposite charge, dusted over the plate, would cling to those areas and reproduce the text, and the powder could then be transferred to paper. Carlson tested the idea at home, in the evenings, in a rented room, using materials bought from a local shop.

On 22 October 1938, in a rented room in the Astoria district of Queens, New York, Carlson and his assistant Otto Kornei succeeded for the first time. Kornei, a physicist employed on a modest wage, rubbed a zinc plate with a sulfur coating and charged it by friction. Carlson wrote the words '10-22-38 ASTORIA' in ink on a glass slide, placed it over the plate and shone a lamp through it. The plate was then dusted with powder and the image pressed onto waxed paper. The date and the place thus became the first photocopy.

Persuading anyone else to take the idea seriously proved far harder. Between 1939 and 1944 Carlson approached more than twenty companies, including some of the largest in the United States, and was turned down by all of them. Several replied that there was no commercial future in the process. His patent, covering what he called electrophotography, was granted in 1942, but without money to develop it the patent was of little use. Relief came in 1944, when the Battelle Memorial Institute, a research organisation in Ohio, agreed to fund further work in return for a share of future royalties.

At Battelle the process was improved and demonstrated to interested firms, and in 1947 a small company in Rochester, New York, took out a licence. The Haloid Company made photographic paper and was a fraction of the size of the corporations that had rejected Carlson, but its management was willing to take a risk. In 1948 Haloid coined the word xerography, from the Greek words for dry and writing, to distinguish the process from photography. A year later it announced the Model A, the first commercial copier to use the new principle. It was a heavy, hand-operated machine: the operator had to place each sheet by hand and carry out the stages of the process one after another.

The machine that made the photocopier a feature of every office appeared in 1959. The Xerox 914 was the first automatic copier to produce images on ordinary, untreated paper, and its name came from the largest sheet it could handle, nine inches by fourteen. It could turn out seven copies a minute, far quicker than the equipment it replaced, but it was expensive to build. Because the company leased the machines rather than selling them outright, it carried the cost for years before the income arrived, and it came close to running out of money.

Inside the 914, the key component was a rotating drum coated with selenium. In the dark, a wire held at a high voltage sprays the drum with electric charge. A lamp then illuminates the original document, and the light reflected from its white areas falls on the drum. Because selenium conducts electricity when illuminated, the charge drains away there, leaving a pattern that matches the text. Fine black powder, called toner, is brought close to the drum and is attracted only to the charged regions. A sheet of paper is pressed against the drum and charged from behind, so that the toner is pulled onto it. The paper then passes between heated rollers, where the powder melts into the fibres. A blade scrapes the drum clean and the cycle begins again.

Haloid changed its name to Xerox in 1961 and grew rapidly during the 1960s. Carlson, who gave away much of his wealth before his death in 1968, became one of the most generous philanthropists in the United States. The photocopier went on to change the daily routine of offices around the world.

A conventional computer keeps its memory in one place and does its processing in another. Every operation requires data to be moved between the two along a channel that engineers call the bus, and the movement costs energy. This arrangement, named after the mathematician John von Neumann, has served computing well for seventy years, but it has become a serious constraint. The training of a large artificial-intelligence model can consume many megawatt-hours of electricity, and much of that power is spent not on calculating but on shifting numbers from one part of the machine to another.

The brain solves the problem differently. It contains roughly 86 billion neurons, each of which may be connected to thousands of others through junctions called synapses. A neuron accumulates signals from its neighbours and, when the total crosses a threshold, fires a brief electrical pulse, or spike. Because most neurons are silent most of the time, energy is consumed mainly when something actually happens. The entire system runs on about twenty watts, roughly the power of a dim light bulb, yet it can recognise a face, hold a conversation and learn a new skill.

Neuromorphic engineering is the attempt to build computing hardware on similar principles. In a neuromorphic chip the basic unit of information is not a number held in a register but the timing of spikes; information is carried by which neurons fire, when, and how often. Crucially, memory and computation are not separated. The strength of a connection, which represents what has been learned, sits in the same place as the mechanism that uses it. Some designs also compute in an analogue fashion, so that a connection can hold a continuous range of values rather than the ones and zeros of ordinary digital logic.

The best-known digital example is TrueNorth, a chip designed by IBM and described in 2014. It contained 5.4 billion transistors and 4,096 processing cores, and mimicked a million neurons joined by 256 million synapses. Its designers reported that it consumed about 70 milliwatts while running a simple recognition task, a tiny fraction of the power needed by a general-purpose processor doing comparable work. TrueNorth was a research device rather than a product, and it had to be programmed in an unfamiliar way, but it showed that the architecture could be built at scale.

Other groups work in analogue electronics. The most promising component is the memristor, a device whose electrical resistance depends on the current that has flowed through it and is retained when the power is switched off. Because it keeps this value, a memristor can act as an artificial synapse, and it can be made extremely small. Millions of them can be arranged in crossbar arrays, grids of fine wires in which every crossing modifies a signal. Arrays of this kind perform the multiplications that neural networks depend on, using a fraction of the energy of digital circuits.

Several large research programmes have produced chips that universities can now obtain. Intel's Loihi, announced in 2017, holds about 130,000 neurons on a single piece of silicon; its successor, Loihi 2, released in 2021, holds a million. SpiNNaker, built at the University of Manchester, takes a different route, linking a million small conventional cores so that they can exchange spikes in imitation of a nervous system. The BrainScaleS system in Heidelberg uses analogue circuits that run roughly a thousand times faster than the biology they copy, which makes it possible to watch learning unfold in seconds.

Uses are beginning to appear outside the laboratory. Event cameras contain a chip whose pixels report only when the brightness at that point changes, rather than capturing a full frame many times a second; they respond in microseconds and produce far less data than a conventional sensor. Similar circuits have been built to imitate the cochlea of the inner ear, turning sound into spikes for speech recognition. Neuromorphic processors have been tested in drones, in robotic hands that must react quickly to touch, and in wearable devices that need to run for months on a single battery.

The obstacles are considerable. Training is the largest. The standard method for training a neural network, in which errors are propagated backwards through the network, assumes that each connection has a fixed and precise value, and it maps awkwardly onto spikes. Learning in the brain appears to depend instead on the relative timing of spikes on either side of a synapse, and implementing such rules efficiently remains an open problem. Analogue devices are also inconsistent: no two memristors behave in exactly the same way, so a network built from them must be trained to tolerate its own imperfections.

Enthusiasts sometimes suggest that neuromorphic chips will soon replace conventional processors. That seems unlikely. The advantages appear only when data are sparse and events are rare, which suits sensors and control systems but not the calculation of a spreadsheet or the rendering of a film. The realistic prospect is a hybrid: small neuromorphic circuits built into ordinary devices, quietly handling the tasks for which biology is better suited, while a digital processor does the rest. Investors, who remember earlier waves of enthusiasm for brain-inspired computing, have reason to be cautious.

Anyone who has tried to have a washing machine repaired, or a smartphone screen replaced, will recognise the frustration that lies behind the campaign known as the right to repair. In the past, a broken appliance was taken to a workshop, opened up and mended. Today the same appliance may be sealed with adhesive rather than screws, may contain components that are sold only as part of a complete assembly, and may refuse to function until a replacement part has been authorised by the manufacturer's software. Owners are left with little choice but to pay for an official repair or to throw the product away and buy another.

The case for change rests first on cost and waste. Repair is often cheaper than replacement. According to the Global E-waste Monitor, the world generated around 62 million tonnes of electronic waste in 2022, and less than a quarter of it was formally collected and recycled; the rest was dumped, burned or handled informally. Much of this waste consists of devices that failed for a trivial reason. Extending the life of a product by a few years would reduce demand for new materials, cut the emissions associated with manufacturing and keep hazardous substances out of the ground.

There is also an argument about ownership. If you have bought an object, the reasoning goes, you should be free to open it, alter it or pay someone else to fix it. Independent repair shops make this possible in places where official service centres are distant or expensive, and in many rural areas they are the only option. Restricting access to parts, tools and technical information shuts such businesses out and concentrates the market in the hands of the original manufacturer.

Manufacturers answer that restrictions exist for reasons. Lithium batteries, mains voltage and high-pressure systems can injure a careless repairer, and a device that has been opened may fail in ways that harm its user. Connected products raise a further concern: a car or an insulin pump whose software can be modified by anyone is easier to attack. Companies also point to design. Sealed cases allow devices to be thinner, lighter and more water-resistant, and a manufacturer that must supply every part for a decade may be less willing to experiment with new materials.

Some of this is reasonable. Few people would argue that the software controlling a pacemaker should be open to modification by its owner, and aircraft maintenance is properly regulated. But the argument is often stretched further than the evidence supports. Sealed cases and difficult repairs are not always the price of good design; they are also an effective way of keeping customers inside a manufacturer's service network, where parts can be sold at a substantial margin. Security, similarly, depends on how systems are built and updated, not simply on who is allowed to open them.

Legislators have begun to respond. Since 2021, European rules on the design of household appliances have required manufacturers to supply spare parts for washing machines, dishwashers and refrigerators for between seven and ten years. In 2024 the European Union adopted a directive on promoting repair, which member states must write into national law by 2026; among other things, it gives consumers the right to ask for a repair rather than a replacement when that would cost less. France already requires a repairability index to be displayed on several categories of product, and several American states, beginning with New York in 2022, have passed laws of their own covering electronic goods.

Legislation of this kind is welcome, but it will not transform the economics of repair on its own. The cost of a repair is dominated by labour, and a technician's hourly rate in a rich country can easily exceed the value of a modest appliance. A rule that spare parts must be available for a decade does nothing for a machine whose motor costs two-thirds of the price of a new one. Nor does it create the skilled people to fit the parts, whose numbers have declined as products became more disposable.

What matters more, therefore, is design. A product that can be opened with a screwdriver, whose battery can be replaced without removing the screen, and whose software continues to receive security updates for a stated number of years is easier to keep alive than one that merely scores well on the availability of parts. Repair cafés, where volunteers help owners to mend their own possessions, have spread across Europe; they show how much can be achieved with advice and a few standard tools, and they also reveal how many failures are caused by a single cheap component.

My own view is that the right to repair is a justified response to a genuine shift in the balance of power between manufacturers and owners. It is not, however, a complete solution. Access to parts and manuals makes repair possible, design makes it practical, and the price of labour determines whether it happens at all. A policy that concentrates on the first of these while ignoring the other two will produce announcements and very little else. Ownership ought to mean the ability to keep something working, and that requires all three.