IELTS Academic Reading Practice Test 148

For most of the eighteenth and early nineteenth centuries, iron and steel were different materials with very different prices. Iron could be produced in large quantities in a blast furnace, but it contained several per cent of carbon, which made it hard and brittle. Steel, with only a small fraction of carbon, was tougher and could be hardened, yet it was made by slow methods: blister steel, produced by heating iron in charcoal for days, and crucible steel, developed in Sheffield in the 1740s by Benjamin Huntsman, in which small batches were melted in clay pots. Steel was made in small quantities and sold at a high price.

The man who changed this was Henry Bessemer, born in 1813 at Charlton in Hertfordshire, the son of an engineer. Bessemer was a prolific inventor with a talent for profitable ideas: he made a fortune from a bronze powder used to imitate gold paint, and from improvements to the manufacture of glass. He had no experience of ironmaking and no formal training in metallurgy. What he did have was a grasp of chemistry, and a particular interest in what happens when air is passed through molten metal.

The idea that made his name was simple to state. Molten iron from a blast furnace contains carbon and other impurities, which must be removed before it can become steel. Bessemer proposed to remove them by blowing air through the liquid metal. The oxygen in the air would combine with the carbon and burn it away, while the oxidation of silicon and manganese would release enough heat to keep the iron molten without any fuel being added. In August 1856 he described the process at a meeting of the British Association in Cheltenham. Within months, steelmakers who had bought licences found that the method often failed, producing a brittle metal that cracked under the hammer.

Part of the difficulty was mechanical. Bessemer's vessel, which became known as a converter, was a large shell of steel plate shaped rather like a pear. Its inside was lined with firebrick. The vessel was mounted so that it could be tilted to receive the molten iron and then turned upright for the blast. Air entered through a number of holes, or tuyères, in the bottom.

At first a shower of sparks and a short flame showed that silicon and manganese were burning; then came a brilliant and much longer flame as the carbon was consumed. Skilled operators learned to judge the end of the blow by the appearance of this flame. Leaving the air on too long removed more carbon than was wanted.

The second problem concerned the quality of the metal. Oxygen from the blast remained dissolved in the steel and made it liable to crack. The British metallurgist Robert Mushet found that adding a small quantity of an alloy of iron and manganese, known as spiegeleisen, removed the oxygen and restored the right proportion of carbon. Meanwhile, in the United States, William Kelly had been experimenting since 1847 with a similar notion of blowing air through molten iron, and in 1857 he was granted a patent for it. Bessemer's American rights were eventually bought, and Kelly received a payment as part of the settlement.

Far more serious was a third problem, which Bessemer could not solve. His process removed carbon, silicon and manganese efficiently, but it had almost no effect on phosphorus, which makes steel brittle. Because most iron ore mined in Britain contained phosphorus, only a handful of deposits — the haematite ores of Cumberland in England and of central Sweden — were suitable. Foreign manufacturers with access to such ore, notably Alfred Krupp in Germany, adopted the process enthusiastically, while British steelmakers found their own ores useless in a Bessemer converter.

The answer came from a Welsh clerk and his cousin. Sidney Gilchrist Thomas worked in a London police court and studied chemistry in the evenings, and he reasoned that if the lining of the converter were alkaline rather than acidic, phosphorus could be drawn out of the iron into the slag. He tested the idea with his cousin Percy Gilchrist, a chemist at a works in South Wales, and the two patented it in 1878, using a lining made from dolomite and adding lime to the charge. The first commercial trials took place at Bolckow Vaughan's works in Middlesbrough the following year. Thomas died of tuberculosis in 1885, at the age of 34, before the technique had transformed steelmaking across Europe.

The consequences were enormous. Steel rails cost several times as much as iron ones in the 1850s, but the price fell sharply as production grew, and within two decades steel had displaced iron for railway track. Cheap steel made possible longer bridges, larger ships, armoured warships and the tall framed buildings of American cities. World output rose from a few tens of thousands of tons a year in the 1860s to millions of tons by the 1890s.

Bessemer's process did not remain dominant. From the 1890s the open-hearth furnace, which was slower but easier to control and could use scrap as well as pig iron, took an increasing share of production. Basic oxygen steelmaking, introduced in Linz in Austria in 1952, uses pure oxygen rather than air and eventually replaced the Bessemer converter almost everywhere. The principle that Bessemer demonstrated — that the impurities in iron can be burned out by a gas blown through it — survives in that process today.

Every message sent across the internet, from a bank transfer to a photograph of a grandchild, is protected by mathematics. Encryption takes readable information, known as plaintext, and transforms it into a stream of apparently random characters, or ciphertext, which is useless to anyone who does not hold the key needed to reverse the transformation. Modern ciphers work on blocks of binary digits and combine them with a key in a sequence of mathematical operations repeated many times, so that changing a single bit of the key produces entirely different output. The most widely used is the Advanced Encryption Standard, or AES, chosen by the United States government in 2001 after a public competition.

AES belongs to the older of the two families of encryption, known as symmetric. The same key both locks and unlocks the message. The weakness lies not in the cipher but in the key. If two strangers wish to communicate, they must first agree on a secret key, and if one of them sends it across the same network an eavesdropper may be listening. For centuries this was the central problem of cryptography, and governments once solved it by delivering keys in person by trusted couriers.

The breakthrough came in 1976, when Whitfield Diffie and Martin Hellman published a scheme in which two parties could agree on a shared secret over a public channel. Their insight was that some mathematical operations are easy in one direction and practically impossible in reverse; such operations act as a trapdoor. The method that followed, RSA, relies on one of them: it is simple to multiply two large prime numbers, but no one knows how to factor the result back into them when the numbers are hundreds of digits long. RSA was developed at the Massachusetts Institute of Technology in 1977 by Ron Rivest, Adi Shamir and Leonard Adleman. In asymmetric cryptography, one key, the public key, is given to everyone; the other, the private key, is kept secret.

Because the mathematics behind these schemes requires far larger numbers, asymmetric encryption is slow — often thousands of times slower than AES. In practice the two are combined. When a browser connects to a bank, the public-key exchange is used for a few milliseconds only, to agree on a fresh symmetric key for that session; the pages themselves are then protected by AES. This hybrid arrangement, known as Transport Layer Security, or TLS, is the reason a padlock appears beside the address bar. A new key for each session limits the damage if one is later discovered: past traffic cannot be read with it.

Encryption is not the only tool. A cryptographic hash function takes any amount of data and produces a fixed-length string of digits, such as the 256-bit output of SHA-256, in a process that cannot be run backwards. Hashing protects the integrity of a file: if a single character changes, the resulting hash is completely different, so software updates can be checked against a published value. Websites also store passwords as hashes rather than as text, so that a stolen database does not reveal them directly. Because identical passwords would produce identical hashes, a random string known as a salt is added to each before hashing.

Hashes also underlie the digital signature, which does more than keep a message secret: it proves who sent it and that it has not been altered. The sender hashes the document and encrypts the hash with a private key; anyone can decrypt it with the public key and compare the result. To know that a public key really belongs to a bank, browsers rely on certificates issued by authorities whose own keys are built into the software. This chain of trust is a social arrangement as much as a mathematical one, and it has been attacked far more often than the ciphers themselves.

That is the recurring lesson of the field. Very few systems have been broken by cracking AES or RSA; the weaknesses are usually elsewhere. A program may be written so that it leaks information through the time it takes to respond, a website may still use an outdated version of TLS, or a key may be stolen by tricking an employee. Engineers speak of the human being as the weakest link: a single misconfigured server can expose the records of millions. Encryption protects data in transit and at rest, but it cannot protect it from those who are entitled to see it.

The next threat is more fundamental. In 1994 the mathematician Peter Shor showed that a quantum computer, if one could be built at sufficient scale, would factor large numbers and compute the discrete logarithms on which public-key systems depend, in a time that grows only slowly with the size of the key. Symmetric ciphers are less vulnerable: doubling the length of the key restores most of their security. Because encrypted traffic can be recorded today and decrypted later, agencies and companies have begun to migrate. In 2024 the National Institute of Standards and Technology published the first of a set of post-quantum standards, based on mathematical problems for which no efficient quantum algorithm is known.

Encryption is built into the chips of a phone, the protocols of a browser and the applications that carry messages. The debate that remains is not about whether the mathematics can keep a secret — it can — but about how far governments should be allowed to demand access, and whether any method of granting them access can be kept from everyone else.

A video circulates in which a well-known politician appears to say something foolish. The face is recognisably theirs, the voice convincing, and the words manufactured. Such clips are known as deepfakes, a term that now covers any image, recording or video generated or altered by machine learning so that it shows a person doing or saying something they never did. The technology is not entirely new: cinema has used computer-generated faces for decades. What is new is the combination of realism, speed and low cost, and the fact that the result can be produced by someone with no training in film production.

Many deepfakes rely on a type of machine learning called a generative adversarial network, or GAN, introduced by the researcher Ian Goodfellow in 2014. Two networks are trained against each other: one, the generator, produces synthetic images, while the other, the discriminator, tries to tell them from real ones. Each improves by reacting to the other until the discriminator can no longer reliably tell fake from real. A common technique for swapping faces trains one network to compress a face into a compact mathematical representation and a second to rebuild it; a shared decoder then renders either face using the expressions and head movements of the original video. Voice cloning works on a similar principle, and can now reproduce a recognizable voice from a few seconds of recorded speech.

The word itself was coined on an internet forum in 2017, when a user began posting explicit videos in which the faces of well-known actresses had been replaced. Within two years the software had moved from research laboratories to consumer applications, and smartphone apps now offer face replacement in a few taps. The expertise required has fallen steadily, while the quality has risen.

The harms are not speculative. Surveys of deepfake videos found online have repeatedly concluded that the largest share is non-consensual sexual imagery, directed mainly at women. In several countries men have been prosecuted for producing such material, and legislators have begun to write laws aimed specifically at it. Money is at risk too: a convincing imitation of a director's voice on the telephone has been enough to authorise a fraudulent payment. Politics is the third arena, where a fabricated clip released shortly before an election can spread faster than any correction.

Yet the most corrosive effect may be on belief itself. Once people know that convincing forgeries exist, they can dismiss genuine recordings as fakes. Researchers call this the liar's dividend: the benefit that accrues to a politician caught behaving badly and now has a plausible excuse. The damage done by deepfakes, on this view, is not only that people believe things that are false, but that they become slower to believe things that are true.

Given these harms, it is easy to see the appeal of prohibition. Some categories of deepfake have no defensible use: an explicit image made without consent is an act of abuse, whatever technology produced it. The law has been slow to respond, and by the time a case reaches a court the damage has been done. Platforms, meanwhile, profit from the attention that shocking material attracts, and have often removed it only after it has been viewed millions of times. On this account a general ban, backed by heavy penalties and a duty on companies to filter uploads, is the only proportionate response.

Against this, critics of prohibition make three arguments. The first is that a blanket ban is unenforceable. The software runs on an ordinary laptop, the files cross borders in seconds, and outlawing a technique is very different from outlawing an act. The second is that the technology has legitimate uses that a general ban would destroy: dubbing a film into another language while preserving the actor's mouth movements, producing training material for medical students, satirising public figures, or allowing a patient who is losing their voice to bank a synthetic version of it. The third argument is that the harm lies in the act, not the tool. Fraud, defamation and the making of sexual images without consent are already offences in most legal systems; what is missing is enforcement, not legislation.

Attention has therefore turned to technical measures. One is provenance: attaching a signed record to an image at the moment it is made, showing which device produced it and whether it has since been edited. Camera manufacturers, software companies and news organizations have cooperated on a common standard for this. The other is detection, which looks for the tell-tale signs that a synthetic image leaves behind — mismatched shadows, unnaturally smooth skin, or patterns in the frequency of the pixels that the eye cannot see. Detection is an arms race, and the advantage shifts with each new model.

Neither measure is sufficient alone. Provenance depends on adoption, and most of the images that do harm are made on devices that were never signed. Detection is probabilistic and always one step behind. My own view is that a general ban would be both unworkable and misdirected: it would criminalise satire and film-making while leaving the determined fraudster untouched. What is defensible is a set of narrow prohibitions, aimed at conduct that is harmful whatever the tool. Such rules would need serious enforcement. They would also justify a duty on platforms to label synthetic material, and a sustained effort in education so that viewers learn to ask where an image came from before they share it.