IELTS Academic Reading Practice Test 135

In the autumn of 1969, a graduate student at the University of California, Los Angeles, sat at a terminal and began to type a message to another machine several hundred kilometres away. The link ran over an ordinary telephone line, and the message was to be the word LOGIN. The system crashed after two letters, so the first communication between the two computers consisted of the syllables LO, but the connection had been made. The technology that would eventually become the internet had begun to work.

The project was paid for by the Advanced Research Projects Agency, an office of the United States Department of Defense created in 1958 in response to the launch of the first Soviet satellite. ARPA funded ambitious research in universities rather than in industry, and in 1962 it appointed J. C. R. Licklider, a psychologist who had recently circulated memoranda imagining computers linked into a single network so that researchers could share data and programs. Licklider stayed only two years, but the office he established supported the work that made the network possible.

Two researchers reached the key technical idea independently. At the RAND Corporation in California, Paul Baran was asked to design a communications system that could survive a nuclear attack. In reports published between 1960 and 1964 he proposed a network with no central hub, in which messages were broken into small blocks, each addressed separately, so that if some routes were destroyed the rest could still carry the traffic. In Britain, Donald Davies of the National Physical Laboratory reached a similar conclusion while designing a system for scientists to share computing time, and it was Davies who coined the word packet for these blocks of data.

The practical design of the network was the work of Lawrence Roberts, who moved to ARPA in 1966. His plan, presented at a conference in 1967, was for a set of small computers, called Interface Message Processors, to act as switches. Each would be attached to a telephone line and to one large computer at a university site, receiving packets and forwarding them towards their destination. The contract to build the machines went to Bolt Beranek and Newman, a small firm in Cambridge, Massachusetts. The first was delivered to Los Angeles in September 1969, and the second followed a month later.

By the end of 1969 four sites were connected: Los Angeles, the Stanford Research Institute, the University of California at Santa Barbara and the University of Utah. The network grew steadily from there, and by 1971 it linked more than a dozen sites. The system was proving useful for far more than the sharing of expensive computers, which had been the original justification.

Electronic mail appeared almost by accident. In 1971 a programmer at the Cambridge firm adapted a messaging program that worked on a single computer so that notes could pass between machines on the network. Within two years email had become the network's dominant activity, carrying far more traffic than the remote logins and file transfers for which it had been designed. In 1973 the network reached outside the United States for the first time, with connections to University College London and to a research institute in Norway.

The original network, however, could not easily be joined to others. In 1974 Vint Cerf and Bob Kahn published a design for a common protocol that would allow separate networks to be linked into what they called an internetwork, and in doing so they set out rules that still govern internet traffic. Their proposal gave a single protocol two jobs: forwarding packets to the right address, and checking that data had arrived intact and putting it back in order. A few years later the pair divided those jobs between two protocols, the Internet Protocol and the Transmission Control Protocol, and the combination took the name TCP/IP.

Adopting the new standard was slow and contentious, but on 1 January 1983 the operators switched over in a single coordinated change, an event remembered in the industry as flag day. From then on, any network that spoke TCP/IP could join. The original ARPANET was closed in 1990, by which time it had been overtaken by the National Science Foundation's network, which linked universities across the United States and, for a while, barred commercial traffic.

Meanwhile, at CERN, the European laboratory for particle physics near Geneva, Tim Berners-Lee had proposed a way of organising documents so that they could be linked to one another and read by anyone with a computer and a connection. The first web page went online in 1991. It was the Mosaic browser, released in 1993, that brought the web to a mass audience, because it displayed images alongside text and could be installed by non-specialists.

Many features of the modern internet can be traced to the assumptions of its designers. The network had no central switchboard that could be seized or switched off, and it assumed that links would fail and packets would be lost. Control was placed at the edges, in the computers that used the network, rather than in the network itself: a router forwards data without knowing what it contains. That principle, sometimes described as keeping the network simple and the endpoints clever, allowed anyone to invent a new application without asking permission.

Proteins are the working molecules of every living cell; they digest food, carry oxygen, copy DNA and fight infection. Each protein is a chain of amino acids, of which twenty kinds occur in nature, strung together in an order specified by a gene. But a chain is useless on its own. To do its job it must fold into a precise three-dimensional shape, and that shape determines what the protein can do: an enzyme's active site must hold its target molecule in exactly the right orientation, and an antibody must present a surface that fits an invader. Knowing the shape is therefore the first step towards understanding the protein, and towards designing drugs that alter its behaviour.

For decades the relationship between sequence and shape was studied experimentally. In work that earned him a Nobel Prize in 1972, the American biochemist Christian Anfinsen showed that a small enzyme, ribonuclease, would refold into its working shape after being unfolded, provided conditions were right. This suggested that everything needed for folding is contained in the sequence itself. Yet how the chain finds that shape remained mysterious. In 1969 Cyrus Levinthal pointed out that a chain of a hundred amino acids could, in principle, adopt so many conformations that finding the correct one by trial and error would take longer than the age of the universe. Proteins evidently do not search at random, but the rules they follow proved very hard to discover.

Experimental methods have provided shapes for many thousands of proteins. X-ray crystallography, the oldest, requires a protein to be persuaded to form a crystal, which many will not do. Nuclear magnetic resonance can study proteins in solution but only small ones. Cryo-electron microscopy freezes samples and images them with electrons, and has become very powerful. All these techniques are demanding, however, and each can take months. The results are stored in the Protein Data Bank, established in 1971, which now holds more than 200,000 structures — a small fraction of the proteins known to exist.

Since 1994 the field has measured its own progress through an unusual exercise called CASP, the Critical Assessment of protein Structure Prediction. Every two years, laboratories are given the sequences of proteins whose structures have been solved but not yet published, and are asked to submit predictions. The structures are then released so that the predictions can be scored. Because participants cannot have seen the answers, CASP offers an unusually honest test of a method.

The most dramatic moment in CASP's history came in 2020. A team at DeepMind, a London-based artificial intelligence company owned by Google's parent, entered a system called AlphaFold 2. Its predictions were close enough to the experimental structures that, on one widely used measure, its median score was about 92 out of 100, a level the organisers described as competitive with experimental methods. AlphaFold draws on two sources of information. The first is the evolutionary record: databases hold the sequences of millions of proteins from different species, and a program can align these to reveal which positions change together. Because such patterns usually mean that two positions touch in the folded protein, they act as constraints on the shape. The second is deep learning: a neural network, trained on the structures already in the Protein Data Bank, learns to convert those constraints into three-dimensional coordinates.

The consequences were rapid. In 2021 DeepMind and the European Molecular Biology Laboratory released a public database containing predicted structures for about 350,000 proteins, including nearly the entire set encoded by the human genome. The following year the database was expanded to more than 200 million entries, covering the great majority of proteins known to science. A structure that once required years of laboratory work had become available, for many proteins, in seconds, at no cost.

Yet a predicted structure is not a solved one. AlphaFold returns a single model of one possible conformation, whereas real proteins move, and their flexibility is often central to how they work. The system also supplies a confidence value for each amino acid, and these values reveal its weaknesses: long stretches of many proteins are genuinely disordered and never settle into a fixed shape. Binding is another limitation. A protein rarely acts alone — it may carry a drug molecule, a metal ion or another protein — and the original AlphaFold did not model these partners. A later version, AlphaFold 3, announced in 2024, extended the approach to complexes of proteins with other molecules, using a different kind of neural network.

The applications are already substantial. Drug designers use predicted structures to find pockets on a protein's surface where a molecule might attach, and to check whether a compound fits before synthesising it. In 2022, a team working on an enzyme that breaks down PET plastic used machine-learning predictions to help redesign it so that it remained active at higher temperatures. Predictions have also been applied to parasites that cause neglected diseases and to the proteins of antibiotic-resistant bacteria.

Where does this leave the laboratory? Few scientists expect crystallography or cryo-electron microscopy to disappear, because predictions must be checked and because fine chemical detail still requires experiment. The larger change is one of emphasis. For years, structural biology was limited by how many shapes could be measured; now it is limited by how many interesting questions can be asked. The field has moved, in other words, from solving structures to using them, and the bottleneck has shifted from measurement to imagination.

There are few pieces of street furniture less popular than the speed camera. Drivers who have just been photographed describe it as a trap, a tax collector in a yellow box, and motoring organisations demand that the money be spent on better roads instead. Yet the case against cameras is weaker than it first appears, and the case for them is more conditional. Cameras do reduce injuries where they are well placed, but they are a tool with sharp limits, and the way they are funded and presented matters as much as the technology itself.

Automated speed enforcement is older than most drivers realise. The first large programmes appeared in the late 1980s and early 1990s, in the Australian state of Victoria and in Britain, and the basic design has barely changed. A fixed camera at the roadside is linked to sensors buried in the road surface or to a radar unit that measures the speed of each approaching vehicle. If a vehicle exceeds the threshold set for that road, the unit photographs its rear, capturing the registration plate and usually a second image a fraction of a second later. The plate is then read by software, checked against the vehicle register, and a penalty notice is posted to the registered keeper, who is required by law to identify the driver.

Does this reduce harm? The most careful summary is a review published by the Cochrane Collaboration in 2010, which pooled the results of dozens of studies from several countries. It concluded that speed cameras cut the number of collisions, injuries and deaths in the area around them, and that the effect was consistent across the studies that met the authors' standards. The same review was notably more cautious about the wider picture: there was not enough reliable evidence to say how much of the benefit spread to roads beyond the camera sites.

Part of the difficulty is statistical. Cameras are deliberately installed at places where crashes have been unusually frequent, and crash numbers at such sites tend to fall afterwards whether or not anything is done, simply because a run of bad luck is unlikely to repeat itself. This regression to the mean can make a camera look more effective than it is. Careful evaluations try to allow for it, often by comparing the camera site with similar roads elsewhere, but the comparison is never perfect.

One design has performed better than the rest. Average-speed cameras, which read a number plate at two or more points along a stretch of road and calculate the time taken to travel between them, do not allow the driver to slow down briefly and accelerate again. Because the whole route is monitored, speeds fall along its entire length, and compliance is usually higher than at single fixed sites. The systems are expensive to install and work best on busy roads, but they have become the preferred choice at roadworks and on stretches where a history of serious collisions suggests that speed is the main problem.

Other problems are harder to engineer away. A driver who spots a camera late brakes sharply, which is a hazard in itself, and then accelerates once past it; the average speed may fall while the risk of a rear-end collision rises. A minority of drivers copy, alter or clone number plates to avoid detection, and devices that warn of cameras or display their locations in advance are legal in many countries. Vandalism is not unknown: cameras in several countries have been attacked by people who regard them as illegitimate.

Fairness is the crux. The objection that most damages public trust is the suspicion that cameras exist to raise money rather than to save lives. The suspicion is often baseless — in many schemes the fines barely cover the cost of running them — but it is easily nourished when the location of cameras is secret, when contracts pay a private operator according to the number of tickets issued, or when revenue disappears into general taxation. Where cameras are painted a conspicuous colour, signposted in advance and the proceeds visibly spent on road improvements, surveys generally find more support than opposition.

Cameras, moreover, cannot be the whole answer, and treating them as such is a mistake that both their critics and their defenders make. Speed is only one factor in road collisions; the design of the road, the presence of pedestrians and cyclists, the standards of vehicle maintenance and the behaviour of drivers all matter. Sweden's Vision Zero policy, adopted in 1997, starts from the principle that no death on the road is acceptable and combines lower speed limits, redesigned streets, safer vehicles and enforcement. In such a system the camera is one instrument among many, and its value depends on the restrictions and the road design around it.

My own view is that the camera has earned its place. Reducing speed reduces the severity of the injuries that follow when something goes wrong: a pedestrian struck at thirty kilometres an hour is far more likely to survive than one struck at fifty. Automated enforcement achieves that reduction continuously. But the technology should be judged, and funded, as a safety measure with a duty to show that it works, not as a convenient substitute for investment or as a machine for punishment. A state that installs cameras without explaining them, and without fixing the roads they stand beside, should not be surprised when drivers regard them with contempt.