IELTS Academic Reading Practice Test 150

Of all the technologies that fill a modern car, the humblest is also the oldest. The lead-acid battery, invented in 1859 by the French physicist Gaston Planté, still starts the engine of almost every petrol and diesel vehicle on the road. It has outlived the steam engine and the cathode-ray tube, and it survives in an era when lithium-ion cells power phones, laptops and a growing share of new cars. Its endurance owes less to elegance than to three stubborn advantages: it is cheap, it is safe and it can be recycled almost completely.

The principle is simple. Two plates of lead sit in dilute sulphuric acid, the electrolyte through which charged particles travel. One plate, the positive electrode, carries a layer of lead dioxide; the other is made of spongy metallic lead. When the plates are connected to a circuit, a chemical reaction releases electrons at the negative plate and absorbs them at the positive one. Each cell produces roughly two volts, so six cells joined in series give the twelve volts a car's electrical system expects. Passing current back through the cell reverses the reaction.

Planté did not set out to build a car battery. In the 1850s he was studying the behaviour of metals in acid, and in 1859 he made a cell from two sheets of lead separated by rubber strips and rolled into a spiral. He demonstrated a larger version, of nine such cells, to the French Academy of Sciences in 1860. The weakness lay in the plates: the lead dioxide had to be built up by passing current through the cell repeatedly, a slow process known as forming that consumed a great deal of energy. Batteries made this way were reliable but expensive, and at first only laboratories and a few telegraph companies bought them.

The decisive improvement came in 1881, when Camille Alphonse Faure, a French chemist, patented a way of coating the plates with a paste of lead oxides. The paste was pressed onto grids of lead alloy, so that forming took a fraction of the time and each plate stored far more charge. Faure turned the battery into a manufactured product, and factories in France, Britain and Germany were soon making them by the thousand. His design, rather than Planté's, made the first electric vehicles possible.

For a brief period in the 1890s and 1900s, battery power looked like the future of transport. Electric cabs carried passengers through London and Paris, and inventors fitted cells to tricycles. Submarines adopted them too, because a boat running on batteries could travel underwater without the air a combustion engine demanded. The competition ended in 1912, when Cadillac introduced an electric starter designed by Charles Kettering. A small lead-acid battery could now turn a petrol engine over at the push of a button, and the hand crank, like the electric car, rapidly disappeared.

The next seventy years brought a long series of refinements rather than a single breakthrough. Lead-antimony alloys strengthened the grids that hold the active material, and better separators allowed the plates to be made thinner. In the 1970s manufacturers began replacing antimony with calcium, which reduced the loss of water during charging and allowed batteries to be sealed for life. Sealed designs, including gel cells and absorbed glass mat batteries, can be mounted in almost any position.

Today the battery's largest market is still the one it captured in 1912: the starter battery, a heavy, cheap block that must deliver an enormous current for a few seconds and then be recharged by the engine. Beyond that it provides backup power for telephone exchanges and data centres, and stores electricity alongside wind and solar farms. Its recycling record is remarkable: in Europe and North America well over 90 per cent of the lead in used batteries is recovered and melted down again.

The technology's failings are equally well known. A lead-acid battery stores only about forty watt-hours per kilogram, a fraction of what a lithium-ion cell holds, so a battery capable of moving a car would weigh several hundred kilograms. It dislikes being left flat: if the lead sulphate that forms on the plates is allowed to crystallise, some of it cannot be converted back and the capacity falls permanently. Deep discharges shorten its life, cold reduces the current it can deliver, and heat makes it fail early.

None of this has ended research. The lead-carbon battery mixes carbon into the negative plate, which slows the build-up of sulphate and allows the battery to work for long periods in a partly charged state — a pattern common in hybrid vehicles and in grids supplied by variable wind and solar power. Other groups are developing bipolar designs, in which one conductive plate serves as the positive electrode of one cell and the negative electrode of the next. Because lead is cheap, such batteries appeal to countries that cannot yet afford lithium-ion storage.

The lead-acid battery will win no prizes for elegance. It is heavy, its energy density is poor by modern standards and its chemistry has changed remarkably little in more than a century and a half. Yet the qualities that made it indispensable in the twentieth century — low cost, tolerance of abuse and a supply chain that returns almost all of its lead to the factory — are precisely those that matter in a world beginning to count the full life-cycle cost of the devices it depends on.

A blockchain is a record of transactions that is copied across many computers and kept in step by a set of rules rather than by an administrator. The idea was described in a paper published in 2008 under the pseudonym Satoshi Nakamoto, and the following year it became the basis of Bitcoin, the first widely used cryptocurrency. A blockchain can record the transfer of anything that can be written down — money, contracts, shipping documents, land titles — but it is best understood as a way of agreeing on the order in which events happened when no one is in charge.

The record is divided into blocks, each containing a batch of transactions, a timestamp and a cryptographic fingerprint, or hash, of the block before it. Because each hash is derived from the contents of the earlier block, altering a single character in an old entry changes its hash, which no longer matches the one recorded in the next block; that block would have to be recalculated too, and so would every block after it. The further back a record lies, the more computing effort would be needed to rewrite it, and the copies held by other computers would disagree with the altered one.

The fingerprinting relies on hash functions, mathematical operations that turn any input into an output of fixed length — 256 bits in the case of SHA-256, the function Bitcoin uses. A good hash function is one-way: it is easy to compute the output from the input but effectively impossible to work backwards. It is also sensitive: changing one letter of a document produces an output that looks entirely unrelated. To summarise thousands of transactions in a single value, blockchains use a structure called a Merkle tree, named after Ralph Merkle, who described the idea in 1979, in which pairs of hashes are repeatedly combined until only one remains.

Agreeing on which blocks to accept requires a rule that every participant can apply. Bitcoin uses proof of work. Computers, often called miners, compete to solve a deliberately difficult puzzle: they must find an input that makes the hash of a candidate block fall below a certain value. Because the output of a hash function cannot be predicted, the only way to find such an input is to try enormous numbers of them, and the winner collects a reward. The process makes the record costly to alter, which is the point, but it also consumes large amounts of electricity. In September 2022 Ethereum, the second-largest cryptocurrency network, replaced mining with a different rule, and its energy use fell by around 99.9 per cent. Under proof of stake, the right to propose a block depends on the size of a deposit rather than on computing power, and validators who behave dishonestly lose their funds.

Not every blockchain is open to all. Public networks such as Bitcoin's allow anyone to read the ledger and submit transactions, whereas permissioned systems restrict participation to approved organisations. Hyperledger, an open-source project hosted by the Linux Foundation, provides frameworks of this kind for banks, shipping companies and government agencies. Such systems are faster and easier to govern, but they reintroduce a degree of control by a known group, which is exactly what blockchains were designed to avoid. In practice many of these networks also rely on a legally appointed administrator, so the technology supplies a shared audit trail rather than true decentralisation.

Where a single organisation can be trusted to keep the records, an ordinary database is simpler, cheaper and faster; a blockchain adds value only where no such organisation exists. A blockchain guarantees that a record has not been altered, not that it was true when written. Much of the practical activity is therefore in areas where several parties must agree. Banks have tested the technology for cross-border payments and trade finance, where documents pass between importers, exporters, shipping firms and customs. Retailers have used it to trace food and clothing through long supply chains, and some governments have experimented with land registries, where secure records of ownership are valuable and paper archives are fragile. In each case the benefit comes from a shared record that no single participant can quietly amend.

The limitations are considerable. Bitcoin processes only a handful of transactions a second, against the tens of thousands that a major payment network can handle, and a new block appears roughly every ten minutes. Increasing the size of blocks or the speed of mining only goes so far before the network becomes harder to run and more centralised, a tension often called the scalability trilemma. Storage also grows without limit, since every full participant must keep a complete copy of the record. And a group that controls more than half of the mining power could, in principle, rewrite recent history.

Research has therefore turned to ways of keeping the security of a public chain while escaping its slowness. 'Layer two' systems called rollups process transactions elsewhere and post compressed proofs back to the main chain, and sharding would divide the ledger into pieces that different computers handle in parallel. Zero-knowledge proofs, which allow one party to convince another that a statement is true without revealing why, are already used to keep transactions private, and cryptographers are designing signatures that would resist attack by a future quantum computer. Whether these techniques will produce systems that ordinary institutions can use remains an open question.

Hospitals now generate more data in a day than their predecessors did in a decade: images, blood results, prescriptions, notes typed by doctors in a hurry. Artificial intelligence of the kind that has transformed speech recognition and translation depends on examples, and for medicine the examples are these records. No single hospital, however large, sees enough cases of a rare cancer, or enough images of a damaged retina, to train a reliable system on its own. Pooling data from many institutions is therefore the obvious solution, and it is also the one that makes patients and politicians uneasy.

The case for pooling is strong. Several widely cited studies have found that software can detect signs of disease in retinal photographs, skin lesions and chest X-rays with an accuracy comparable to that of experienced specialists. Such systems improve markedly as the number and variety of training images grow, and they fail most often when they have seen only one kind of patient. A model taught exclusively on scans from a hospital in one country may perform poorly on patients from another, whose age profile, diet or genetics differ. Variety in the training data is not a luxury; it is the difference between a tool that travels and one that does not.

Rare conditions make the point even more sharply. A neurologist may see two cases of a particular degenerative disease in a career. Even a large teaching hospital may hold only a few hundred examples of an uncommon tumour, far too few to teach a system to recognise it. Only by combining scans and genetic sequences from centres in many countries can researchers assemble a collection large enough for a model to learn from. For families affected by such conditions, the argument that their data should stay in separate silos is hard to defend.

There are warnings, too. A study published in the journal Science in 2019 examined a commercial algorithm used in American hospitals to decide which patients should be enrolled in extra care programmes. The algorithm predicted cost rather than illness, and because less money had historically been spent on Black patients with the same symptoms, it consistently underestimated how ill they were. The fault lay not in the data as such but in the question the model had been asked. The episode is a reminder that an algorithm is only as good, or as fair, as the target it has been told to optimise. Systems trained on unrepresentative populations reproduce, and can amplify, the assumptions of the institutions that collected the data.

Privacy is the second objection, and it is not imaginary. Removing names from a dataset does not make it anonymous. In the late 1990s a researcher showed that she could identify the medical records of a state governor from a hospital dataset that had had names removed, simply by combining it with a public voter register and matching a few details. Since then the quantity of information available for such matching has grown enormously, and a determined analyst can often re-identify individuals from records that were thought to be safe.

Technical fixes help, but none is complete. In federated learning, a model is trained inside each hospital and only the adjustments it makes are shared, so the records themselves never leave the building. Synthetic data, generated to resemble real records without belonging to anyone, can be used for some purposes. Yet federated models can leak information through those adjustments, and synthetic records can reproduce details of the real ones. Anonymisation, encryption and these newer methods all raise the cost of misuse without abolishing it.

Consent is the hardest question. Asking every patient to approve every study would make large-scale research impossible, and most people, when surveys ask them, say they are willing for their data to be used for research provided they are told what is happening and by whom. What they object to is surprise: the discovery that records given for their own care have been passed to a company that intends to sell a product trained on them. The reasonable answer is a system of broad permission rather than a fiction of individual bargaining. Patients should also be able to refuse, clearly and at any time, without having to justify the decision.

Governance therefore matters more than consent forms. Approval should rest with independent panels that include patients. A scheme to pool the records of family doctors' patients in England was suspended in 2014 and abandoned two years later. Critics had pointed out that the public had not been told clearly what would happen to their data or how to opt out of the scheme. The lesson was not that sharing is unacceptable but that it cannot be done quietly, and no single company should be given exclusive access to a national collection.

The debate is usually presented as a choice between privacy and medical progress. That framing is misleading. The alternative to well-run sharing is not privacy; it is the present situation, in which data are already bought, sold and combined, often without the knowledge of the people they describe. Pooled records, governed publicly and used for purposes the public can see, are a better defence of privacy than a fragmentation that protects no one. The question is not whether medical data will be shared, but on what terms.