Aurelia Butler

Quantum computers of the future hold promise in solving all sorts of problems. For example, they could lead to more sustainable materials, new medicines, and even crack the hardest problems in fundamental physics. But compared to classical computers in use today, rudimentary quantum computers are more prone to errors. Wouldn’t it be nice if researchers could just take out a special quantum eraser and get rid of the mistakes?
Reporting in the journal Nature, a group of researchers led by Caltech is among the first to demonstrate a type of quantum eraser. The physicists show that they can pinpoint and correct for mistakes in quantum computing systems known as “erasure” errors.
“It’s normally very hard to detect errors in quantum computers, because just the act of looking for errors causes more to occur,” says Adam Shaw, co-lead author of the new study and a graduate student in the laboratory of Manuel Endres, a professor of physics at Caltech. “But we show that with some careful control, we can precisely locate and erase certain errors without consequence, which is where the name erasure comes from.”
Quantum computers are based on the laws of physics that govern the subatomic realm, such as entanglement, a phenomenon in which particles remain connected to and mimic each other without being in direct contact. In the new study, the researchers focused on a type of quantum-computing platform that uses arrays of neutral atoms, or atoms without a charge. Specifically, they manipulated individual alkaline-earth neutral atoms confined inside “tweezers” made of laser light. The atoms were excited to high-energy states — or “Rydberg” states — in which neighboring atoms start interacting.
“The atoms in our quantum system talk to each other and generate entanglement,” explains Pascal Scholl, the other co-lead author of the study and a former postdoctoral scholar at Caltech now working at a quantum computing company in France called PASQAL.
Entanglement is what allows quantum computers to outperform classical computers. “However, nature doesn’t like to remain in these quantum entangled states,” Scholl explains. “Eventually, an error happens, which breaks the entire quantum state. These entangled states can be thought of as baskets full of apples, where the atoms are the apples. With time, some apples will start to rot, and if these apples are not removed from the basket and replaced by fresh ones, all the apples will rapidly become rotten. It is not clear how to fully prevent these errors from happening, so the only viable option nowadays is to detect and correct them.”
The new error-catching system is designed in such a way that erroneous atoms fluoresce, or light up, when hit with a laser. “We have images of the glowing atoms that tell us where the errors are, so we can either leave them out of the final statistics or apply additional laser pulses to actively correct them,” Scholl says.
The theory for implementing erasure detection in neutral atom systems was first developed by Jeff Thompson, a professor of electrical and computer engineering at Princeton University, and his colleagues. That team also recently reported demonstrating the technique in the journal Nature.
By removing and locating errors in their Rydberg atom system, the Caltech team says that they can improve the overall rate of entanglement, or fidelity. In the new study, the team reports that only one in 1,000 pairs of atoms failed to become entangled. That’s a factor-of-10 improvement over what was achieved previously and is the highest-ever observed entanglement rate in this type of system.
Ultimately, these results bode well for quantum computing platforms that use Rydberg neutral atom arrays. “Neutral atoms are the most scalable type of quantum computer, but they didn’t have high-entanglement fidelities until now,” says Shaw. More

Artificial intelligence (AI) comes with promises of helping coders code faster, drivers drive safer, and making daily tasks less time-consuming. But in a commentary published October 10 in the journal Joule, the founder of Digiconomist demonstrates that the tool, when adopted widely, could have a large energy footprint, which in the future may exceed the power demands of some countries.
“Looking at the growing demand for AI service, it’s very likely that energy consumption related to AI will significantly increase in the coming years,” says author Alex de Vries, a Ph.D. candidate at Vrije Universiteit Amsterdam.
Since 2022, generative AI, which can produce text, images, or other data, has undergone rapid growth, including OpenAI’s ChatGPT. Training these AI tools requires feeding the models a large amount of data, a process that is energy intensive. Hugging Face, an AI-developing company based in New York, reported that its multilingual text-generating AI tool consumed about 433 megawatt-hours (MWH) during training, enough to power 40 average American homes for a year.
And AI’s energy footprint does not end with training. De Vries’s analysis shows that when the tool is put to work — generating data based on prompts — every time the tool generates a text or image, it also uses a significant amount of computing power and thus energy. For example, ChatGPT could cost 564 MWh of electricity a day to run.
While companies around the world are working on improving the efficiencies of AI hardware and software to make the tool less energy intensive, de Vries says that an increase in machines’ efficiency often increases demand. In the end, technological advancements will lead to a net increase in resource use, a phenomenon known as Jevons’ Paradox.
“The result of making these tools more efficient and accessible can be that we just allow more applications of it and more people to use it,” de Vries says.
Google, for example, has been incorporating generative AI in the company’s email service and is testing out powering its search engine with AI. The company processes up to 9 billion searches a day currently. Based on the data, de Vries estimates that if every Google search uses AI, it would need about 29.2 TWh of power a year, which is equivalent to the annual electricity consumption of Ireland.
This extreme scenario is unlikely to happen in the short term because of the high costs associated with additional AI servers and bottlenecks in the AI server supply chain, de Vries says. But the production of AI servers is projected to grow rapidly in the near future. By 2027, worldwide AI-related electricity consumption could increase by 85 to 134 TWh annually based on the projection of AI server production.
The amount is comparable to the annual electricity consumption of countries such as the Netherlands, Argentina, and Sweden. Moreover, improvements in AI efficiency could also enable developers to repurpose some computer processing chips for AI use, which could further increase AI-related electricity consumption.
“The potential growth highlights that we need to be very mindful about what we use AI for. It’s energy intensive, so we don’t want to put it in all kinds of things where we don’t actually need it,” de Vries says. More

In the last 20 years, disasters have claimed more than a million lives and caused nearly $3 trillion in economic losses worldwide, according to the United Nations.
Disaster relief organizations (DROs) mobilize critical resources to help impacted communities, and they use social media to distribute information rapidly and broadly. Many DROs post content via multiple accounts within a single platform to represent both national and local levels.
Specifically examining wildfires in collaboration with the Canadian Red Cross (CRC), new research from the University of Notre Dame contradicts existing crisis communication theory that recommends DROs speak with one voice during the entirety of wildfire response operations.
“Speak with One Voice? Examining Content Coordination and Social Media Engagement During Disasters” is forthcoming in Information Systems Research from Alfonso Pedraza-Martinez, the Greg and Patty Fox Collegiate Professor of IT, Analytics and Operations at the University of Notre Dame’s Mendoza College of Business.
Social media informs victims about wildfires, but it also connects volunteers, donors and other supporters. Accounts can send coordinated messages targeting the same audience (match) or different audiences (mismatch).
According to crisis communication theory, a disaster relief organization’s communication channels should speak with one voice through multiple accounts targeting the same audience, but the team’s study recommends a more nuanced approach.
“We find the national and local levels should match audiences during the early wildfire response when uncertainty is very high, but they should mismatch audiences during recovery while the situation is still critical but uncertainty has decreased,” said Pedraza-Martinez, who specializes in humanitarian operations and disaster management. “We find that user engagement increases when the national headquarters lead the production of content and the local accounts follow either by tweeting to a matching or mismatching audience, depending on timing in the operation.”
The study reveals that engagement improves by 4.3 percent from a match only during the uncertain and urgent response phase, while a divergence of content creation decisions, or mismatch, yields 29.6 percent more engagement when uncertainty subsides during the recovery phase. More

Models built on machine learning in health care can be victims of their own success, according to researchers at the Icahn School of Medicine and the University of Michigan. Their study assessed the impact of implementing predictive models on the subsequent performance of those and other models. Their findings — that using the models to adjust how care is delivered can alter the baseline assumptions that the models were “trained” on, often for worse — were detailed in the October 9 online issue of Annals of Internal Medicine.
“We wanted to explore what happens when a machine learning model is deployed in a hospital and allowed to influence physician decisions for the overall benefit of patients,” says first and corresponding author Akhil Vaid, M.D., Clinical Instructor of Data-Driven and Digital Medicine (D3M), part of the Department of Medicine at Icahn Mount Sinai. “For example, we sought to understand the broader consequences when a patient is spared from adverse outcomes like kidney damage or mortality. AI models possess the capacity to learn and establish correlations between incoming patient data and corresponding outcomes, but use of these models, by definition, can alter these relationships. Problems arise when these altered relationships are captured back into medical records.”
The study simulated critical care scenarios at two major health care institutions, the Mount Sinai Health System in New York and Beth Israel Deaconess Medical Center in Boston, analyzing 130,000 critical care admissions. The researchers investigated three key scenarios:
1. Model retraining after initial use
Current practice suggests retraining models to address performance degradation over time. Retraining can improve performance initially by adapting to changing conditions, but the Mount Sinai study shows it can paradoxically lead to further degradation by disrupting the learned relationships between presentation and outcome.
2. Creating a new model after one has already been in use
Following a model’s predictions can save patients from adverse outcomes such as sepsis. However, death may follow sepsis, and the model effectively works to prevent both. Any new models developed in the future for prediction of death will now also be subject to upset relationships as before. Since we do not know the exact relationships between all possible outcomes, any data from patients with machine-learning influenced care may be inappropriate to use in training further models. More

Scientists at the UCL Institute for Neurology have developed new tools, based on AI language models, that can characterise subtle signatures in the speech of patients diagnosed with schizophrenia.
The research, published in PNAS, aims to understand how the automated analysis of language could help doctors and scientists diagnose and assess psychiatric conditions.
Currently, psychiatric diagnosis is based almost entirely on talking with patients and those close to them, with only a minimal role for tests such as blood tests and brain scans.
However, this lack of precision prevents a richer understanding of the causes of mental illness, and the monitoring of treatment.
The researchers asked 26 participants with schizophrenia and 26 control participants to complete two verbal fluency tasks, where they were asked to name as many words as they could either belonging to the category “animals” or starting with the letter “p,” in five minutes.
To analyse the answers given by participants, the team used an AI language model that had been trained on vast amounts of internet text to represent the meaning of words in a similar way to humans. They tested whether the words people spontaneously recalled could be predicted by the AI model, and whether this predictability was reduced in patients with schizophrenia.
They found that the answers given by control participants were indeed more predictable by the AI model than those generated by people with schizophrenia, and that this difference was largest in patients with more severe symptoms. More

A team of New York University computer scientists has created a neural network that can explain how it reaches its predictions. The work reveals what accounts for the functionality of neural networks — the engines that drive artificial intelligence and machine learning — thereby illuminating a process that has largely been concealed from users.
The breakthrough centers on a specific usage of neural networks that has become popular in recent years — tackling challenging biological questions. Among these are examinations of the intricacies of RNA splicing — the focal point of the study — which plays a role in transferring information from DNA to functional RNA and protein products.
“Many neural networks are black boxes — these algorithms cannot explain how they work, raising concerns about their trustworthiness and stifling progress into understanding the underlying biological processes of genome encoding,” says Oded Regev, a computer science professor at NYU’s Courant Institute of Mathematical Sciences and the senior author of the paper, which appears in the Proceedings of the National Academy of Sciences. “By harnessing a new approach that improves both the quantity and the quality of the data for machine-learning training, we designed an interpretable neural network that can accurately predict complex outcomes and explain how it arrives at its predictions.”
Regev and the paper’s other authors, Susan Liao, a faculty fellow at the Courant Institute, and Mukund Sudarshan, a Courant doctoral student at the time of the study, created a neural network based on what is already known about RNA splicing.
Specifically, they developed a model — the data-driven equivalent of a high-powered microscope — that allows scientists to trace and quantify the RNA splicing process, from input sequence to output splicing prediction.
“Using an ‘interpretable-by-design’ approach, we’ve developed a neural network model that provides insights into RNA splicing — a fundamental process in the transfer of genomic information,” notes Regev. “Our model revealed that a small, hairpin-like structure in RNA can decrease splicing.”
The researchers confirmed the insights their model provides through a series of experiments. These results showed a match with the model’s discovery: Whenever the RNA molecule folded into a hairpin configuration, splicing was halted, and the moment the researchers disrupted this hairpin structure, splicing was restored.
The research was supported by grants from the National Science Foundation (MCB-2226731), the Simons Foundation, the Life Sciences Research Foundation, an Additional Ventures Career Development Award, and a PhRMA Fellowship. More

A single-atom-thick sheet of carbon known as graphene has remarkable properties on its own, but things can get even more interesting when you stack up multiple sheets. When two or more overlying sheets of graphene are sightly misaligned — twisted at certain angles relative to each other — they take on a plethora of exotic identities.Depending on the twist angle, these materials, known as moiré quantum matter, can suddenly generate their own magnetic fields, become superconductors with zero electrical resistance, or conversely, turn into perfect insulators.
Joseph A. Stroscio and his colleagues at the National Institute of Standards and Technology (NIST), along with an international team of collaborators, have developed a “quantum ruler” to measure and explore the strange properties of these twisted materials. The work may also lead to a new, miniaturized standard for electrical resistance that could calibrate electronic devices directly on the factory floor, eliminating the need to send them to an off-site standards laboratory.
Collaborator Fereshte Ghahari, a physicist from George Mason University in Fairfax, Virginia, took two layers of graphene (known as bilayer graphene) of about 20 micrometers across and twisted them relative to another two layers to create a moiré quantum matter device. Ghahari made the device using the nanofabrication facility at NIST’s Center for Nanoscale Science and Technology. NIST researchers Marlou Slot and Yulia Maximenko then chilled this twisted material device to one-hundredth of a degree above absolute zero, reducing random motions of atoms and electrons and heightening the ability for electrons in the material to interact. After reaching ultralow temperatures, they examined how the energy levels of electrons in the layers of graphene changed when they varied the strength of a strong external magnetic field. Measuring and manipulating the energy levels of electrons is critical for designing and manufacturing semiconductor devices.
To measure the energy levels, the team used a versatile scanning tunneling microscope that Stroscio designed and built at NIST. When the researchers applied a voltage to the graphene bilayers in the magnetic field, the microscope recorded the tiny current from the electrons that “tunneled” out from the material to the microscope probe tip.
In a magnetic field, electrons move in circular paths. Ordinarily, the circular orbits of the electrons in solid materials have a special relationship with an applied magnetic field: The area enclosed by each circular orbit, multiplied by the applied field, can only take on a set of fixed, discrete values, due to the quantum nature of electrons. In order to maintain that fixed product, if the magnetic field is halved, then the area enclosed by an orbiting electron must double. The difference in energy between successive energy levels that follow this pattern can be used like tick marks on a ruler to measure the material’s electronic and magnetic properties. Any subtle deviation from this pattern would represent a new quantum ruler that can reflect the orbital magnetic properties of the particular quantum moiré material researchers are studying.
In fact, when the NIST researchers varied the magnetic field applied to the moiré graphene bilayers, they found evidence of a new quantum ruler at play. The area enclosed by the circular orbit of electrons multiplied by the applied magnetic field no longer equaled a fixed value. Instead, the product of those two numbers had shifted by an amount dependent on the magnetization of the bilayers.
This deviation translated into a set of different tick marks for the energy levels of the electrons. The findings promise to shed new light on how electrons confined to twisted sheets of graphene give rise to new magnetic properties. More

A new attempt to predict earthquakes with the aid of artificial intelligence has raised hopes that the technology could one day be used to limit earthquakes’ impact on lives and economies. Developed by researchers at The University of Texas at Austin, the AI algorithm correctly predicted 70% of earthquakes a week before they happened during a seven-month trial in China.
The AI was trained to detect statistical bumps in real-time seismic data that researchers had paired with previous earthquakes. The outcome was a weekly forecast in which the AI successfully predicted 14 earthquakes within about 200 miles of where it estimated they would happen and at almost exactly the calculated strength. It missed one earthquake and gave eight false warnings.
It’s not yet known if the same approach will work at other locations, but the effort is a milestone in research for AI-driven earthquake forecasting.
“Predicting earthquakes is the holy grail,” said Sergey Fomel, a professor in UT’s Bureau of Economic Geology and a member of the research team. “We’re not yet close to making predictions for anywhere in the world, but what we achieved tells us that what we thought was an impossible problem is solvable in principle.”
The trial was part of an international competition held in China in which the UT-developed AI came first out of 600 other designs. UT’s entry was led by bureau seismologist and the AI’s lead developer, Yangkang Chen. Findings from the trial are published in the journal Bulletin of the Seismological Society of America.
“You don’t see earthquakes coming,” said Alexandros Savvaidis, a senior research scientist who leads the bureau’s Texas Seismological Network Program (TexNet) — the state’s seismic network. “It’s a matter of milliseconds, and the only thing you can control is how prepared you are. Even with 70%, that’s a huge result and could help minimize economic and human losses and has the potential to dramatically improve earthquake preparedness worldwide.”
The researchers said that their method had succeeded by following a relatively simple machine learning approach. The AI was given a set of statistical features based on the team’s knowledge of earthquake physics, then told to train itself on a five-year database of seismic recordings. More