IELTS Academic Reading Practice Test 6

A For most of the twentieth century, scientific understanding of memory was dominated by a simple metaphor: the brain as a recording device, with memories stored intact in a particular place and retrieved later more or less as they were encoded. The work of the American neuroscientist Karl Lashley, who spent decades in the 1920s through 1940s removing portions of rat brains in search of the "engram" — the physical trace of a memory — produced the unexpected finding that no single location appeared to be essential for memory. Lashley concluded that memory must be distributed across many regions, a view that has been substantially confirmed by modern research while being refined in important ways.

B The most influential modern framework, due principally to the Canadian psychologist Donald Hebb, holds that memories are stored as patterns of connection strength between neurons. Hebb's 1949 principle, often paraphrased as "neurons that fire together, wire together," proposed that the simultaneous firing of two neurons strengthens the synaptic connection between them, making future co-firing more likely. The discovery of long-term potentiation in the rabbit hippocampus by Norwegian researchers Terje Lømo and Tim Bliss in 1973 is widely said to have provided the first direct biological evidence for this proposed mechanism, and decades of subsequent research have elaborated its molecular basis.

C Different types of memory generally depend on different brain systems. Declarative memory — for facts and events — relies critically on the hippocampus, a seahorse-shaped structure in the medial temporal lobe. Damage to the hippocampus, whether through stroke, surgery, or disease, produces an inability to form new declarative memories while leaving older memories and most other cognitive functions intact. The famous patient known as H.M., whose hippocampi were surgically removed in 1953 to treat severe epilepsy, lived for the next 55 years unable to form new lasting memories of his daily experiences while retaining his pre-surgical memories and the ability to learn new motor skills.

D Procedural memory, by contrast, involves the basal ganglia and the cerebellum. This is the kind of memory that allows a violinist to play a passage without consciously remembering each note, or that allows an experienced driver to navigate a car without thinking about the steps. The case of H.M. and many subsequent studies generally show that procedural memory can persist normally even when declarative memory is severely impaired — and the reverse pattern, intact declarative memory with impaired procedural memory, occurs in certain neurological conditions such as Parkinson's disease.

E Memory consolidation — described as the process by which a fragile new memory becomes stable and resistant to disruption — has been a major focus of recent research. Sleep appears to play a central role: studies using both rodents and humans have shown that the brain replays the patterns of activity associated with daytime experiences during slow-wave sleep, broadly effectively rehearsing them in the absence of sensory input. The transfer of memories from the hippocampus, where they are initially encoded, to the neocortex, where they appear to be stored long-term, takes place largely during sleep across weeks or months.

F A more recent and controversial development is the discovery that consolidated memories appear to become temporarily unstable when they are recalled — a phenomenon called reconsolidation. Studies in which animals are given protein-synthesis inhibitors at the moment a memory is retrieved have demonstrated that the memory can be permanently disrupted, suggesting that recalling a memory is not simply reading it but rewriting it. The implications for treatment of post-traumatic stress disorder and other memory-related conditions are being actively investigated, although early clinical trials have, on the whole, produced mixed results.

G Perhaps the most surprising finding of recent memory research, in many psychologists' view, is how often human memories are simply wrong. Cognitive psychologist Elizabeth Loftus has demonstrated in dozens of experiments that misleading questions, suggestion, and post-event information can create confident, vivid memories of events that never happened. These findings reportedly have profound implications for the legal system, where eyewitness testimony has historically been treated as among the most compelling forms of evidence. Loftus's work has contributed to substantial reforms in police interview procedures and to the abandonment of certain therapeutic techniques that inadvertently encouraged the creation of false memories.

A Hydrogen is the most abundant element in the universe but exists on Earth almost entirely in compound form, principally as water and as the hydrogen atoms locked in hydrocarbons. To use hydrogen as a fuel, energy must be invested to separate it from these compounds — energy that can later be recovered when the hydrogen recombines with oxygen, releasing only water as a by-product. This in principle makes hydrogen an attractive carrier for renewable electricity, allowing energy generated when the sun shines or the wind blows to be stored and transported in chemical form. The vision of a hydrogen economy has been periodically promoted since the 1970s, although until recently the technologies and prices required have remained out of reach.

B The cleanliness of hydrogen as a fuel depends entirely on how it is produced. The cheapest current method is steam methane reforming, in which natural gas reacts with steam at high temperature to yield hydrogen and carbon dioxide. This produces what industry calls grey hydrogen — chemically clean at point of use, but carbon-intensive in production. Blue hydrogen, in which the carbon dioxide is captured and stored underground, can reduce these emissions, although critics argue that capture rates in real installations have rarely matched the theoretical claims. Green hydrogen, produced by passing renewable electricity through water in an electrolyser, is the cleanest option and the one most often referred to when commentators speak of a hydrogen economy.

C Electrolysers themselves have advanced dramatically. Alkaline electrolysers, the oldest commercial type, have been used for decades in industrial chemistry. Proton exchange membrane electrolysers, refined since the 1970s, are smaller, more responsive, and better suited to pairing with intermittent renewable sources, although they generally require expensive platinum-group catalysts. Solid oxide electrolysers, the newest type, reportedly operate at very high temperatures and can achieve efficiencies above 80 percent but are not yet widely commercially deployed. The cost of green hydrogen production has fallen substantially as electrolyser manufacturing has scaled and renewable electricity prices have collapsed.

D End-use applications generally fall into three rough categories. The first is industrial: hydrogen is already a major industrial chemical, used in oil refining, ammonia production for fertilisers, and steelmaking experiments using direct reduction rather than coke. Replacing fossil-derived hydrogen with green hydrogen in these existing industrial uses would, in many analysts' view, offer some of the largest near-term emissions reductions. The second category is transport. Hydrogen fuel cells, which combine hydrogen and oxygen to produce electricity directly, are competitive with battery-electric drivetrains for heavy long-distance vehicles such as lorries, ships, and possibly aircraft, where the weight of batteries becomes a serious constraint. The third category is heating, although the case for replacing natural gas in domestic boilers with hydrogen is more contested.

E Hydrogen has, on the whole, significant practical disadvantages that constrain its use. It is the lightest gas, requiring either very high pressure or cryogenic temperatures to achieve useful energy density. It tends to leak through joints and seals that contain other gases, and embrittles certain metals on prolonged exposure. Distribution requires either dedicated pipelines or specialised tanks. These issues do not make hydrogen unworkable, but they generally explain why most analysts expect hydrogen to find its most successful applications in industrial settings, where infrastructure can be purpose-built, rather than in distributed consumer use.

F Government policy is widely said to have driven much of the recent expansion. The European Union's REPowerEU plan published in 2022 set a target of producing 10 million tonnes of renewable hydrogen domestically by 2030, with an additional 10 million tonnes imported. The United States Inflation Reduction Act of 2022 provided tax credits of up to three dollars per kilogram for clean hydrogen, generally the most generous incentive of its kind anywhere. Japan and South Korea, both heavily dependent on imported energy, have established national hydrogen strategies focused on importing green hydrogen from sunnier regions. Whether these policies will translate into commercial-scale infrastructure, in many writers' view, remains the central question of the next decade.

A The application of artificial intelligence to medicine has progressed in fits and starts over five decades, but the pace of advance is widely said to have accelerated dramatically since around 2015. Machine-learning systems, particularly the deep neural networks that have transformed many fields, now match or exceed expert human performance on several specific diagnostic tasks. The promise is, in many proponents' view, enormous: faster diagnosis, fewer missed cases, and access to specialist-level reasoning in regions that lack the trained workforce. Realising this promise, however, requires solving problems of data quality, fairness, regulation, and clinical integration that have proved substantially harder than the underlying mathematics.

B Medical imaging is widely regarded as the most successful early application. In 2017, a team at Stanford University trained a deep neural network on 130,000 dermatology images and showed it matched the diagnostic accuracy of board-certified dermatologists in identifying skin cancers. Subsequent work has demonstrated comparable performance for diabetic retinopathy on retinal photographs, breast cancer on mammograms, and tuberculosis on chest X-rays. The pattern of success is consistent: a clearly defined visual task, a large labelled dataset, and a binary or low-cardinality classification problem.

C The promise of AI in pathology is, in many writers' view, equally striking. Pathologists generally examine tissue samples to make critical diagnoses, but the work is laborious and varies in accuracy. A 2019 study published in The Lancet Digital Health found that a deep-learning system trained on prostate biopsies matched the performance of experienced pathologists at identifying cancerous tissue and produced consistent results generally free of inter-observer variability. The Food and Drug Administration in the United States has begun authorising clinical use of such systems on a case-by-case basis, although adoption has, on the whole, been slower than its proponents predicted.

D Beyond imaging, AI is being applied to electronic health record data to identify patients at elevated risk of clinical deterioration. Systems trained on millions of patient records can flag patients on hospital wards likely to develop sepsis hours before traditional clinical scoring systems do. However, real-world results have often been disappointing. A 2021 evaluation of a widely deployed sepsis-prediction algorithm found that it correctly identified only 7 percent of patients who actually developed sepsis and generated frequent false alarms, producing what nurses described as alert fatigue.

E Drug discovery represents another frontier. The conventional pharmaceutical pipeline takes more than a decade and costs over two billion dollars per approved drug, with most candidate molecules failing late in development. Machine-learning systems can in principle reportedly accelerate this process by predicting which molecular structures are most likely to bind to a chosen target and which are likely to have unacceptable toxicity. The 2020 demonstration by DeepMind's AlphaFold system, which predicts protein three-dimensional structures from amino acid sequences with near-experimental accuracy, is widely said to have opened entirely new approaches to designing drugs against previously inaccessible targets.

F Concerns about fairness and bias have, in many ethicists' view, grown alongside the technology. AI systems generally learn the patterns present in their training data, and historical medical data reflects historical inequities. A 2019 study published in Science showed that a widely used commercial algorithm for allocating high-risk care management resources systematically under-allocated to Black patients because it used spending on health care, rather than actual health, as a proxy for need — and Black patients had historically received less spending for equivalent illness. Similar biases have been demonstrated in algorithms used for skin cancer diagnosis on darker skin, where training images had been disproportionately drawn from light-skinned populations.

G The regulatory landscape is evolving rapidly. Traditional medical-device regulation assumes a fixed product whose performance can be characterised before approval and is expected to remain stable thereafter. Machine-learning systems, by contrast, can be designed to update continuously as they are exposed to new data. Regulators including the FDA and the European Medicines Agency have developed new pathways for adaptive AI systems, although critics argue the frameworks generally remain immature and that real clinical integration is being held back by the resulting uncertainty.

H Looking ahead, most analysts reportedly expect AI to take on an increasingly important role in healthcare without coming close to replacing clinicians. The pattern is, in many observers' view, likely to be augmentation rather than substitution: AI systems assisting with image triage, providing decision support for difficult diagnoses, and surfacing patterns in large patient datasets that no individual clinician could observe. The hardest challenges, as is often the case, are not technical but social and organisational: ensuring that the gains flow widely, that the technology is deployed safely, and that the human relationships at the centre of medicine are not displaced by the convenience of algorithmic prediction.