IELTS Academic Reading Practice Test 180

For most of the twentieth century, financial planning was done on paper. An accountant would rule a large sheet into rows and columns, write figures in the spaces and add totals at the foot of each column. Because the arithmetic was done by hand, any change to a single figure meant that the totals, and every figure that depended on them, had to be worked out again. Large organisations employed clerks to keep such sheets up to date, and the work was slow and prone to error. Questions of the 'what if' kind, such as what would happen if sales fell by a tenth, were expensive to answer.

The idea for a better method occurred to Dan Bricklin in 1978, when he was a student at Harvard Business School. He later described watching a professor draw a table of figures on a blackboard, erase one value and recalculate the rest of the table by hand. Bricklin had worked as a programmer before going into business, and he imagined a screen on which the table itself would recalculate. With a colleague, Bob Frankston, he wrote a program they called VisiCalc, short for 'visible calculator', and the two men founded a company, Software Arts, to sell it.

VisiCalc went on sale in 1979 for the Apple II, one of the first personal computers to be widely owned. It presented a grid of cells, each identified by a letter and a number, such as B7. A cell could hold either a value or a formula, and a formula could refer to other cells, so that changing one figure updated every cell that depended on it. The memory of the small machines limited the first version to a few hundred rows and about sixty columns, which was nevertheless more than most users needed. The screen showed only a small window onto the grid.

Demand for the program was remarkable. Accountants, estate agents, engineers and small businesses bought computers in order to run it, and VisiCalc is often described as the first 'killer application', software so useful that customers bought a machine simply to use it. Sales ran into the hundreds of thousands, and the program is now regarded as one of the inventions that turned the personal computer from a hobbyist's toy into a tool of business.

Success attracted competitors, and the most significant appeared in 1983. Lotus 1-2-3 was written by Mitch Kapor's company, Lotus Development Corporation, for the IBM Personal Computer, a machine with more memory and a faster processor than the Apple II. It combined a spreadsheet with charting and a simple database in one program, and it offered menus so that users did not have to memorise commands. Lotus 1-2-3 quickly became the standard in offices. VisiCalc could not compete, and Lotus bought Software Arts in 1985 and discontinued VisiCalc.

The next stage belonged to Microsoft. Excel was first released for the Apple Macintosh in 1985 and for Windows two years later. Early versions were praised for their graphical interface, which let users adjust column widths and headings with a mouse rather than by typing instructions. The struggle for the market was not settled immediately, because many companies had spent years writing templates and training staff to use Lotus 1-2-3. But Excel was sold alongside word-processing and presentation software in a single package, and by the early 1990s it had become the leader.

Later changes have concerned capacity as much as features. For many years Excel was limited to 65,536 rows and 256 columns; the version released in 2007 raised these limits to 1,048,576 rows and 16,384 columns, more than a thousand times as many cells. The change reflected the falling cost of computer memory and the appetite of users, who now employ spreadsheets for tasks the inventors never imagined, from school timetables and mailing lists to the analysis of scientific data and the costing of building projects.

The spreadsheet has also produced legal argument. Rival companies that copied the sequence of commands in the Lotus 1-2-3 menus were sued, and in 1995 a federal appeals court in the United States decided that this menu structure was a method of operation and therefore could not be protected by copyright. The Supreme Court divided evenly on the question in 1996, which left the appeals court's decision in place.

Mistakes remain the technology's weakness. Because a formula is copied invisibly down a column or across a row, an error in one cell can spread through an entire model, and researchers who have audited spreadsheets used in business and government have repeatedly found errors in a substantial proportion of the files they examined. A misplaced decimal point, or a range of cells that stops one row short of the last entry, may be invisible on screen and yet produce a plausible but wrong total. Modern programs can trace which cells feed into a result, but no tool can supply the judgement that the person building the model must exercise.

Few pieces of software have changed office life so thoroughly. The electronic spreadsheet did not merely speed up arithmetic; it made possible a style of planning in which assumptions are altered and their consequences seen at once. Its descendant is now so familiar that a grid of cells seems an obvious way to arrange numbers, yet before 1979 no such thing existed.

When a person asks a voice assistant for the weather, the recording usually travels to a data centre hundreds of kilometres away, where a large model converts the speech to text, works out an answer and sends it back. This centralised arrangement, known as cloud computing, has dominated for two decades. Large facilities can buy hardware cheaply, share it among many customers and employ specialists to run it, and a customer pays only for what is used. Amazon Web Services, which opened in 2006, established the pattern, and today a handful of companies operate fleets of buildings filled with servers.

Distance, however, costs time. Light travelling through optical fibre moves at about 200,000 kilometres a second, roughly two-thirds of its speed in a vacuum, and every router on the route adds a further delay. A data centre 1,000 kilometres away therefore adds at least ten milliseconds to a round trip before any processing begins. That is imperceptible when a web page loads, but it is a large fraction of the time available to a machine that has to decide whether to brake. The delay is fixed by physics rather than by the speed of the computers at either end.

Bandwidth is the second constraint. A single security camera streaming high-definition video produces several megabits every second, and a factory with a thousand sensors or a city with ten thousand cameras would need enormous network capacity to send all of that raw data to a central server. Most of it, moreover, is uninteresting: an empty corridor, a vibration within the normal range, a temperature that has not changed. Edge computing is a response to both problems. Data are processed close to where they are produced, and only the conclusions are transmitted. A camera that reports 'no people present' sends a few bytes rather than a megabyte.

What counts as the edge varies from one application to another. It may be a gateway, a small computer in a factory that collects readings from machines; a base station in a mobile telephone network; or the device itself, such as a watch that counts steps. Content delivery networks were an early example of the same logic: since the late 1990s companies such as Akamai have kept copies of web pages and video in thousands of locations, so that a request is answered by a nearby server rather than one on another continent. Around 2012 the networking company Cisco proposed a related idea, which it called fog computing, for processing data inside the network rather than at its centre.

The advantages fall into four groups. The first is latency: a vehicle that must recognise a child stepping into the road has a few tens of milliseconds in which to act. The second is reliability, since a machine that continues to work when its connection fails is safer than one that stops. The third is privacy, because video from a hospital ward or a family's living room need never leave the building if the analysis happens locally. The fourth is cost, since transmitting a short summary is cheaper than transmitting a continuous stream.

The difficulty is that edge devices are small. A microcontroller in a sensor may have a few hundred kilobytes of memory and be expected to run for years on a battery, which is a tiny fraction of the resources available in a data centre. Models must therefore be reduced before they will fit. Pruning removes connections in a neural network that contribute little to its output; quantisation represents numbers with fewer bits. Together they can shrink a model to a fraction of its original size at a modest cost in accuracy. The field, known as TinyML, already allows an inexpensive chip to recognise a spoken word or a face while drawing only milliwatts of power.

Managing thousands of devices is harder than managing one data centre. Every device must be updated, secured and monitored, and many sit in places that are difficult to reach. Physical security is a further worry, because equipment in a public place can be stolen or tampered with, and the data it holds may be valuable. Manufacturers are therefore developing chips that can prove which software they are running, and methods of installing updates in small, reversible steps so that a failed update does not disable a device permanently.

One promising line of research is federated learning, in which a model is trained across many devices without the raw data ever being collected in one place. Each phone or sensor computes an adjustment to the model, and only those adjustments are sent to a server, which combines them into an improved version. Because the data stay where they were generated, the approach offers obvious privacy advantages. Recent work has shown, however, that the adjustments can sometimes reveal information about the data behind them, so additional protection, such as adding a controlled amount of noise, is needed.

Few specialists expect the edge to replace the cloud. The arrangement most engineers anticipate is a hierarchy: a sensor filters, a gateway aggregates, a regional centre runs a large model, and a distant data centre trains models and stores archives. The question is not which layer will win, but which jobs belong at each.

In September 2017 a small company called Bird left a few hundred electric scooters on the pavements of Santa Monica, California, and invited people to ride them for a dollar plus a small charge for each minute of use. Within two years similar machines had appeared in more than a hundred cities, and investors responded with enthusiasm. The reaction was extreme in both directions. Enthusiasts announced the end of car ownership, while critics counted injuries, blocked pavements and scooters pulled out of rivers. Neither reaction now looks sensible.

The case for the scooter rests on three claims: that it is cheap, that it is clean and that it replaces car journeys. The first is largely true for the rider, who avoids taxi fares and parking charges. The second is more complicated. A scooter produces no exhaust emissions, but making one requires a lithium battery and an aluminium frame, and the shared versions of 2018 were collected each night in diesel vans so that their batteries could be recharged. Life-cycle analyses have found that whether a shared scooter is greener than the journey it replaces depends on how long the machine lasts and how it is maintained.

The third claim is the most important and the least certain. Several city surveys have found that a majority of shared scooter trips replace walking, cycling or public transport rather than a car journey, so the environmental gain is small and may be negative where the alternative was walking. Where a scooter is used instead of a short taxi ride, the benefit is real. The difficulty is that operators cannot choose which journeys their customers make; only the layout of the city and the price of alternatives can do that.

Safety is the strongest objection. Riders share the road with vehicles several times their weight, stand on a small deck at a height that makes falls painful, and in most cities are not required to wear a helmet. Studies of hospital records have reported high rates of fractures to the wrist, arm and head, and many injuries involve people using a scooter for the first time. Yet the design of the street, not the scooter, is what puts riders in danger.

Parking is the complaint that citizens raise most often. Because a rental scooter is unlocked by an application and abandoned wherever the rider stops, the machines accumulate in doorways, at bus stops and along the edges of pavements, where they obstruct people in wheelchairs and parents pushing buggies. Cities have responded with marked parking bays, with requirements to send a photograph of the parked machine at the end of each ride, and with caps on the size of fleets. Paris went further and banned rented scooters from 1 September 2023, after a referendum in which 89 per cent of those who voted supported the ban, on a turnout below eight per cent.

Paris illustrates a general problem. A city that treats the scooter as a nuisance to be removed loses whatever benefits it might have brought, while a city that leaves everything to the market gets overcrowded pavements. The most promising model is neither. Several European cities now award permits to two or three operators, require them to spread their machines evenly across districts, set a maximum speed, and oblige them to remove badly parked scooters within a stated time. Enforcement matters as much as the rules themselves, since a permit condition that is never checked changes nothing.

Britain has taken a cautious route. For several years privately owned e-scooters have been illegal on public roads, while rented machines in government-approved trial areas have been permitted, with speed limited and use confined mainly to roads and cycle lanes. The arrangement is confusing: a rider may legally hire a scooter but commit an offence by buying an identical one. It is hard to see the logic, although the trials have at least produced evidence about injuries and about how the machines are used. Several American cities, by contrast, allowed fleets to grow far beyond what their pavements could absorb before imposing limits.

My own view is that the scooter is a modest technology that has been asked to carry an unreasonable burden of expectation. It will not replace the car, and it does not need to. In a city where many journeys are shorter than three kilometres, a machine that costs little to run, occupies no parking space and can be combined with a train has an obvious place — provided that it is kept off pavements, that its speed is capped, and that it is charged by a vehicle that is not a diesel van.

The real question, then, is not whether cities should allow electric scooters but under what conditions. Judged by that test, the answers are unglamorous: parking bays, protected lanes, limits on fleet size and honest reporting of injuries. The scooter is neither a revolution nor a plague; it is a small device whose value depends entirely on the streets we give it.