Computers Math

In a world first, a ‘biological camera’ bypasses the constraints of current DNA storage methods, harnessing living cells and their inherent biological mechanisms to encode and store data. This represents a significant breakthrough in encoding and storing images directly within DNA, creating a new model for information storage reminiscent of a digital camera.
Led by Principal Investigator Associate Professor Chueh Loo Poh from the College of Design and Engineering at the National University of Singapore, and the NUS Synthetic Biology for Clinical and Technological Innovation (SynCTI), the team’s findings, which could potentially shake up the data-storage industry, were published in Nature Communications on 3 July 2023.
A new paradigm to address global data overload
As the world continues to generate data at an unprecedented rate, data has come to be seen as the ‘currency’ of the 21st century. Estimated to be 33 ZB in 2018, it has been forecasted that the Global Datasphere will reach 175 ZB by 2025. That has sparked a quest for a storage alternative that can transcend the confines of conventional data storage and address the environmental impact of resource-intensive data centres.
It is only recently that the idea of using DNA to store other types of information, such as images and videos, has garnered attention. This is due to DNA’s exceptional storage capacity, stability, and long-standing relevance as a medium for information storage.
“We are facing an impending data overload. DNA, the key biomaterial of every living thing on Earth, stores genetic information that encodes for an array of proteins responsible for various life functions. To put it into perspective, a single gram of DNA can hold over 215,000 terabytes of data — equivalent to storing 45 million DVDs combined,” said Assoc Prof Poh.

“DNA is also easy to manipulate with current molecular biology tools, can be stored in various forms at room temperature, and is so durable it can last centuries,” says Cheng Kai Lim, a graduate student working with Assoc Prof Poh.
Despite its immense potential, current research in DNA storage focuses on synthesising DNA strands outside the cells. This process is expensive and relies on complex instruments, which are also prone to errors.
To overcome this bottleneck, Assoc Prof Poh and his team turned to live cells, which contain an abundance of DNA that can act as a ‘data bank’, circumventing the need to synthesise the genetic material externally.
Through sheer ingenuity and clever engineering, the team developed ‘BacCam’ — a novel system that merges various biological and digital techniques to emulate a digital camera’s functions using biological components.
“Imagine the DNA within a cell as an undeveloped photographic film,” explained Assoc Prof Poh. “Using optogenetics — a technique that controls the activity of cells with light akin to the shutter mechanism of a camera, we managed to capture ‘images’ by imprinting light signals onto the DNA ‘film’.”
Next, using barcoding techniques akin to photo labelling, the researchers marked the captured images for unique identification. Machine-learning algorithms were employed to organise, sort, and reconstruct the stored images. These constitute the ‘biological camera’, mirroring a digital camera’s data capture, storage, and retrieval processes.
The study showcased the camera’s ability to capture and store multiple images simultaneously using different light colours. More crucially, compared to earlier methods of DNA data storage, the team’s innovative system is easily reproducible and scalable.
“As we push the boundaries of DNA data storage, there is an increasing interest in bridging the interface between biological and digital systems,” said Assoc Prof Poh.
“Our method represents a major milestone in integrating biological systems with digital devices. By harnessing the power of DNA and optogenetic circuits, we have created the first ‘living digital camera,’ which offers a cost-effective and efficient approach to DNA data storage. Our work not only explores further applications of DNA data storage but also re-engineers existing data-capture technologies into a biological framework. We hope this will lay the groundwork for continued innovation in recording and storing information.” More

Researchers from Queen Mary University of London have made groundbreaking advancements in bionics with the development of a new electric variable-stiffness artificial muscle. Published in Advanced Intelligent Systems, this innovative technology possesses self-sensing capabilities and has the potential to revolutionize soft robotics and medical applications. The artificial muscle seamlessly transitions between soft and hard states, while also sensing forces and deformations. With flexibility and stretchability similar to natural muscle, it can be integrated into intricate soft robotic systems and adapt to various shapes. By adjusting voltages, the muscle rapidly changes its stiffness and can monitor its own deformation through resistance changes. The fabrication process is simple and reliable, making it ideal for a range of applications, including aiding individuals with disabilities or patients in rehabilitation training.
In a study published recently in Advanced Intelligent Systems, researchers from Queen Mary University of London have made significant advancements in the field of bionics with the development of a new type of electric variable-stiffness artificial muscle that possesses self-sensing capabilities. This innovative technology has the potential to revolutionize soft robotics and medical applications.
Muscle contraction hardening is not only essential for enhancing strength but also enables rapid reactions in living organisms. Taking inspiration from nature, the team of researchers at QMUL’s School of Engineering and Materials Science has successfully created an artificial muscle that seamlessly transitions between soft and hard states while also possessing the remarkable ability to sense forces and deformations.
Dr. Ketao Zhang, a Lecturer at Queen Mary and the lead researcher, explains the importance of variable stiffness technology in artificial muscle-like actuators. “Empowering robots, especially those made from flexible materials, with self-sensing capabilities is a pivotal step towards true bionic intelligence,” says Dr. Zhang.
The cutting-edge artificial muscle developed by the researchers exhibits flexibility and stretchability similar to natural muscle, making it ideal for integration into intricate soft robotic systems and adapting to various geometric shapes. With the ability to withstand over 200% stretch along the length direction, this flexible actuator with a striped structure demonstrates exceptional durability.
By applying different voltages, the artificial muscle can rapidly adjust its stiffness, achieving continuous modulation with a stiffness change exceeding 30 times. Its voltage-driven nature provides a significant advantage in terms of response speed over other types of artificial muscles. Additionally, this novel technology can monitor its deformation through resistance changes, eliminating the need for additional sensor arrangements and simplifying control mechanisms while reducing costs.
The fabrication process for this self-sensing artificial muscle is simple and reliable. Carbon nanotubes are mixed with liquid silicone using ultrasonic dispersion technology and coated uniformly using a film applicator to create the thin layered cathode, which also serves as the sensing part of the artificial muscle. The anode is made directly using a soft metal mesh cut, and the actuation layer is sandwiched between the cathode and the anode. After the liquid materials cure, a complete self-sensing variable-stiffness artificial muscle is formed.
The potential applications of this flexible variable stiffness technology are vast, ranging from soft robotics to medical applications. The seamless integration with the human body opens up possibilities for aiding individuals with disabilities or patients in performing essential daily tasks. By integrating the self-sensing artificial muscle, wearable robotic devices can monitor a patient’s activities and provide resistance by adjusting stiffness levels, facilitating muscle function restoration during rehabilitation training.
“While there are still challenges to be addressed before these medical robots can be deployed in clinical settings, this research represents a crucial stride towards human-machine integration,” highlights Dr. Zhang. “It provides a blueprint for the future development of soft and wearable robots.”
The groundbreaking study conducted by researchers at Queen Mary University of London marks a significant milestone in the field of bionics. With their development of self-sensing electric artificial muscles, they have paved the way for advancements in soft robotics and medical applications. More

That experiences leave their trace in the connectivity of the brain has been known for a while, but a pioneering study by researchers at the German Center for Neurodegenerative Diseases (DZNE) and TUD Dresden University of Technology now shows how massive these effects really are. The findings in mice provide unprecedented insights into the complexity of large-scale neural networks and brain plasticity. Moreover, they could pave the way for new brain-inspired artificial intelligence methods. The results, based on an innovative “brain-on-chip” technology, are published in the scientific journal Biosensors and Bioelectronics.
The Dresden researchers explored the question of how an enriched experience affects the brain’s circuitry. For this, they deployed a so-called neurochip with more than 4,000 electrodes to detect the electrical activity of brain cells. This innovative platform enabled registering the “firing” of thousands of neurons simultaneously. The area examined — much smaller than the size of a human fingernail — covered an entire mouse hippocampus. This brain structure, shared by humans, plays a pivotal role in learning and memory, making it a prime target for the ravages of dementias like Alzheimer’s disease. For their study, the scientists compared brain tissue from mice, which were raised differently. While one group of rodents grew up in standard cages, which did not offer any special stimuli, the others were housed in an “enriched environment” that included rearrangeable toys and maze-like plastic tubes.
“The results by far exceeded our expectations,” said Dr. Hayder Amin, lead scientist of the study. Amin, a neuroelectronics and nomputational neuroscience expert, heads a research group at DZNE. With his team, he developed the technology and analysis tools used in this study. “Simplified, one can say that the neurons of mice from the enriched environment were much more interconnected than those raised in standard housing. No matter which parameter we looked at, a richer experience literally boosted connections in the neuronal networks. These findings suggest that leading an active and varied life shapes the brain on whole new grounds.”
Unprecedented Insight into Brain Networks
Prof. Gerd Kempermann, who co-leads the study and has been working on the question of how physical and cognitive activity helps the brain to form resilience towards aging and neurodegenerative disease, attests: “All we knew in this area so far has either been taken from studies with single electrodes or imaging techniques like magnetic resonance imaging. The spatial and temporal resolution of these techniques is much coarser than our approach. Here we can literally see the circuitry at work down to the scale of single cells. We applied advanced computational tools to extract a huge amount of details about network dynamics in space and time from our recordings.”
“We have uncovered a wealth of data that illustrates the benefits of a brain shaped by rich experience. This paves the way to understand the role of plasticity and reserve formation in combating neurodegenerative diseases, especially with respect to novel preventive strategies,” Prof. Kempermann said, who, in addition to being a DZNE researcher, is also affiliated with the Center for Regenerative Therapies Dresden (CRTD) at TU Dresden. “Also, this will help provide insights into disease processes associated with neurodegeneration, such as dysfunctions of brain networks.”
Potential Regarding Brain-inspired Artificial Intelligence
“By unraveling how experiences shape the brain’s connectome and dynamics, we are not only pushing the boundaries of brain research,” states Dr. Amin. “Artificial intelligence is inspired by how the brain computes information. Thus, our tools and the insights they allow to generate could open the way for novel machine learning algorithms.” More

As researchers make major advances in medical care, they are also discovering that the efficacy of these treatments can be enhanced by individualized approaches. Therefore, clinicians increasingly need methods that can both continuously monitor physiological signals and then personalize responsive delivery of therapeutics.
Need for safe, flexible bioelectronic devices
Implanted bioelectronic devices are playing a critical role in these treatments, but there are a number of challenges that have stalled their widespread adoption. These devices require specialized components for signal acquisition, processing, data transmission, and powering. Up to now, achieving these capabilities in an implanted device has entailed using numerous rigid and non-biocompatible components that can lead to tissue disruption and patient discomfort. Ideally, these devices need to be biocompatible, flexible, and stable in the long term in the body. They also must be fast and sensitive enough to record rapid, low-amplitude biosignals, while still being able to transmit data for external analysis.
Columbia researchers invent first stand-alone, flexible, fully organic bioelectronic device
Columbia Engineering researchers announced today that they have developed the first stand-alone, conformable, fully organic bioelectronic device that can not only acquire and transmit neurophysiologic brain signals, but can also provide power for device operation. This device, about 100 times smaller than a human hair, is based on an organic transistor architecture that incorporates a vertical channel and a miniaturized water conduit demonstrating long-term stability, high electrical performance, and low-voltage operation to prevent biological tissue damage. The findings are outlined in a new study, published today in Nature Materials.
Both researchers and clinicians knew there was a need for transistors that concurrently pose all of these features: low voltage of operation, biocompatibility, performance stability, conformability for in vivo operation; and high electrical performance, including fast temporal response, high transconductance, and crosstalk-free operation. Silicon-based transistors are the most established technologies, but they are not a perfect solution because they are hard, rigid, and unable to establish a very efficient ion interface with the body. ]
The team addressed these issues by introducing a scalable, self-contained, sub-micron IGT (internal-ion-gated organic electrochemical transistor) architecture, the vIGT. They incorporated a vertical channel arrangement that augments the intrinsic speed of the IGT architecture by optimizing channel geometry and permitting a high density arrangement of transistors next to each other — , 155,000of them per centimeter square.

Scalable vGITs are the fastest electrochemical transistors
The vIGTs are composed of biocompatible, commercially available materials that do not require encapsulation in biological environments and are not impaired by exposure to water or ions. The composite material of the channel can be reproducibly manufactured in large quantities and is solution-processible, making it more accessible to a broad range of fabrication processes. They are flexible and compatible with integration into a wide variety of conformable plastic substrates and have long-term stability, low inter-transistor crosstalk, and high-density integration capacity, allowing fabrication of efficient integrated circuits.
“Organic electronics are not known for their high performance and reliability,” said the study’s leader Dion Khodagholy, associate professor of electrical engineering. “But with our new vGIT architecture, we were able to incorporate a vertical channel that has its own supply of ions. This self-sufficiency of ions made the transistor to be particularly fast — in fact, they are currently the fastest electrochemical transistors.”
To push the speed of operation even further, the team used advanced nanofabrication techniques to miniaturize and densify these transistors at submicro-meter scales. Fabrication took place in the cleanroom of the Columbia Nano Initiative.
Collaborating with CUIMC clinicians
To develop the architecture, the researchers first needed to understand the challenges involved with diagnosis and treatment of patients with neurological disorders like epilepsy, as well as the methodologies currently used. They worked with colleagues at the Department of Neurology at Columbia University Irving Medical Center, in particular, with Jennifer Gelinas, assistant professor of neurology, electrical and biomedical engineering and director of the Epilepsy and Cognition Lab.

The combination of high-speed, flexibility. and low-voltage operation enables the transistors to not only be used for neural signal recording but also for data transmission as well as powering the device, leading to a fully conformable implant. The researchers used this feature to demonstrate fully soft and confirmable implants capable of recording and transmitting high resolution neural activity from both outside, on the surface of the brain, as well as inside, deep within the brain.
“This work will potentially open a wide range of translational opportunities and make medical implants accessible to a large patient demographic who are traditionally not qualified for implantable devices due to the complexity and high risks of such procedures,” said Gelinas.
“It’s amazing to think that our research and devices could help physicians with better diagnostics and could have a positive impact on patients’ quality of life,” added the study’s lead author Claudia Cea, who recently completed her PhD and will be a postdoctoral fellow at MIT this fall.
Next steps
The researchers plan next to join forces with neurosurgeons at CUIMC to validate the capabilities of vIGT-based implants in operating rooms. The team expects to develop soft and safe implants that can detect and identify various pathological brain waves caused by neurological disorders. More

Unused solar, wind, and hydroelectric power in the U.S. could support the exponential growth of transactions involving non-fungible tokens (NFTs), Cornell Engineering researchers have found.
Fengqi You, the Roxanne E. and Michael J. Zak Professor in Energy Systems Engineering in Cornell Engineering, is corresponding author of “Climate Concerns and the Future of Non-Fungible Tokens: Leveraging Environmental Benefits of the Ethereum Merge,” which published July 10 in Proceedings of the National Academy of Sciences. You’s co-author is Apoorv Lal, graduate student in chemical and biomolecular engineering and a member of the You Research Group.
Processing of NFT transactions, which has increased fourfold over the past five years, was once highly energy-intensive but has been made more sustainable with a recent switch to a more energy-efficient algorithm. But those savings, the researchers said, will be largely offset by the anticipated boom in yearly NFT activity.
Excess renewable energy, due to lack of storage capability, forces grid operators to curtail production. You’s idea would put that unused energy-production potential to good use.
“It’s the same idea as a car sitting in someone’s garage,” You said. “If it’s not being driven, they could lend it to someone for carsharing. In our case, wind, solar and hydro power sources that aren’t being utilized could be used to do something good.”
“Of course, this would be up to the industry and policymakers,” he said, “but technology-wise, we show it’s very feasible because these power sources are there already.”
Their key finding: The increased NFT processing activity could be powered, in part, from un- or underutilized existing power sources. Fifty megawatts of potential hydropower from existing U.S. dams that are not currently used to generate power, or a 15% utilization of wind and solar energy that can’t currently be used or stored from sources in Texas, could be used to power an exponential increase in NFT transactions.

Blockchain technologies, including NFT transactions, offer a high level of security in a variety of applications, but the energy required to process each transaction is problematic in a warming world.
“In the beginning, people only cared about the usefulness of these applications,” Lal said. “But then they started to realize the energy and climate impacts, because the crux of all these applications is the utilization of massive amounts of energy.”
Without any efforts to make NFT transaction processing more sustainable, the authors wrote, their annual emissions will reach an equivalent of 0.37 megatons of carbon dioxide — close to the CO2 emissions from 1 million single-trip flights for a passenger from New York to London.
In September of 2022, the Ethereum blockchain responded to the call for more sustainable trading by switching from an energy-intensive proof of work (PoW) algorithm to a proof of stake (PoS) consensus mechanism, which requires less computing power. Energy consumption decreased drastically following the switch, known as the Ethereum Merge.
Still, the authors wrote, an exponential rise in recorded NFT transactions would translate to more validators operating on the network. Toward the end of this decade, energy consumed by an exponential increase in NFT transactions could be equivalent to that of 100,000 U.S. households.

So even with significantly less energy consumption for individual NFT transactions, the cumulative effect of increased numbers of validators operating on fossil fuel-dominant grids will lead to a further rise in the associated carbon debt.
“By the end of this decade,” You said, “the carbon produced by NFT transactions may be roughly equivalent to that produced in one year by a 600-megawatt coal-fired power plant.”
The authors evaluated two hydroelectric energy carriers — green hydrogen and green ammonia (more energy-dense than hydrogen) — for their viability, noting that their cost savings are influenced by multiple factors, including transportation distances and the utilization levels of available renewable energy sources.
Retrofitting these existing power sources could be challenging, the authors said, but would still be good for energy carriers and the planet.
“NFT processing is very power-hungry,” You said, “so this turns out to be a good way to take advantage of these curtailments.”
You is a senior faculty fellow of the Cornell Atkinson Center for Sustainability and co-director of Cornell University AI for Science Institute.
This research was supported by a grant from the National Science Foundation. More

One of the main ways cells “talk” to each other to coordinate essential biological activities such as muscle contraction, hormone release, neuronal firing, digestion and immune activation is through calcium signaling.
Rice University scientists have used light-activated molecular machines to trigger intercellular calcium wave signals, revealing a powerful new strategy for controlling cellular activity, according to a new study published in Nature Nanotechnology. This technology could lead to improved treatments for people with heart problems, digestive issues and more.
“Most of the drugs developed up to this point use chemical binding forces to drive a specific signaling cascade in the body,” said Jacob Beckham, a chemistry graduate student and lead author on the study. “This is the first demonstration that, instead of chemical force, you can use mechanical force — induced, in this case, by single-molecule nanomachines — to do the same thing, which opens up a whole new chapter in drug design.”
Scientists used small-molecule-based actuators that rotate when stimulated by visible light to induce a calcium-signaling response in smooth muscle cells.
We lack conscious control over many of the critical muscles in our body: The heart is an involuntary muscle, and there is smooth muscle tissue lining our veins and arteries, controlling blood pressure and circulation; smooth muscle lines our lungs and intestines and is involved in digestion and breathing. The ability to intervene in these processes with a molecular-level mechanical stimulus could be game-changing.
“Beckham has shown that we can control, for example, cells’ signaling in a heart muscle, which is really interesting,” said James Tour, Rice’s T. T. and W. F. Chao Professor of Chemistry and a professor of materials science and nanoengineering.

“If you stimulate just one cell in the heart, it will propagate the signal to the neighboring cells, which means you could have targeted, adjustable molecular control over heart function and possibly alleviate arrhythmias,” Tour said.
Activated by quarter-second-long light pulses, the molecular machines allowed scientists to control calcium signaling in a cardiac myocyte cell culture, causing the inactive cells to fire.
“The molecules essentially served as nano-defibrillators, getting these heart muscle cells to start beating,” Beckham said.
The ability to control cell-to-cell communication in muscle tissue could be useful for the treatment of a wide range of diseases characterized by calcium-signaling dysfunction.
“A lot of people who are paralyzed have huge digestive problems,” Tour said. “It would be a big deal if you could alleviate these issues by causing those relevant muscles to fire without any kind of chemical intervention.”
The molecule-sized devices activated the same calcium-based cellular signaling mechanism in a live organism, causing whole-body contraction in a fresh-water polyp, or Hydra vulgaris.

“This is the first example of taking a molecular machine and using it to control an entire functioning organism,” Tour said.
Cellular response varied based on the type and intensity of the mechanical stimulation: Fast, unidirectionally rotating molecular machines elicited intercellular calcium wave signals, while slower speeds and multidirectional rotation did not.
Moreover, adjusting the intensity of the light allowed scientists to control the strength of the cellular response.
“This is mechanical action at the molecular scale,” Tour said. “These molecules spin at 3 million rotations per second, and because we can adjust the duration and intensity of the light stimulus, we have precise spatiotemporal control over this very prevalent cellular mechanism.”
The Tour lab has shown in previous research that light-activated molecular machines can be deployed against antibiotic-resistant infectious bacteria, cancer cells and pathogenic fungi.
“This work expands the capabilities of these molecular machines in a different direction,” Beckham said. “What I love about our lab is that we are fearless when it comes to being creative and pursuing projects in ambitious new directions.”
“We’re currently working towards developing machines activated by light with a better depth of penetration to really actualize the potential of this research. We are also looking to get a better understanding of molecular-scale actuation of biological processes.”
The research was supported by the Discovery Institute, the Robert A. Welch Foundation (C-2017-20190330), the National Science Foundation Graduate Research Fellowship Program, the DEVCOM Army Research Laboratory (Cooperative Agreement W911NF-18-2-0234) and the European Union’s Horizon 2020 (Marie Sklodowska-Curie grant agreement 843116). More

A team from Ames National Laboratory conducted an in-depth investigation of the magnetism of TbMn6Sn6, a Kagome layered topological magnet. They were surprised to find that the magnetic spin reorientation in TbMn6Sn6 occurs by generating increasing numbers of magnetically isotropic ions as the temperature increases.
Rob McQueeney, a scientist at Ames Lab and project lead, explained that TbMn6Sn6has two different magnetic ions in the material, terbium and manganese. The direction of the manganese moments controls the topological state, “But it’s the terbium moment that determines the direction that the manganese points,” he said. “The idea is, you have these two magnetic species and it is the combination of their interactions which controls the direction of the moment.”
In this layered material, there is a magnetic phase transition that occurs as the temperature increases. During this phase transition, the magnetic moments switch from pointing perpendicular to the Kagome layer, or uniaxial, to pointing within the layer, or planar. This transition is called a spin reorientation.
McQueeney explained that in Kagome metals, the spin direction controls the properties of topological or Dirac electrons. Dirac electrons occur where the magnetic bands touch at one point. However, magnetic order causes gapping at the points where the bands are touching. This gapping stabilizes the topological Chern insulator state. “So you can go from a Dirac semimetal to a Chern insulator just by turning the direction of the moment,” he said.
As part of their TbMn6Sn6 investigation, the team performed inelastic neutron scattering experiments at the Spallation Neutron Source to understand how the magnetic interactions in the material drive the spin reorientation transition. McQueeney said that the terbium wants to be uniaxial at low temperatures, while the manganese is planar, so they are at odds.
According to McQueeney, the behavior at very low or very high temperatures is as expected. At low temperatures, the terbium is uniaxial (with electronic orbitals shaped like an ellipsoid). At high temperatures, the terbium is magnetically isotropic (with a spherical orbital shape), which allows the planar Mn to determine the overall moment direction. The team assumed that each terbium orbital would gradually deform from ellipsoidal to spherical. Instead, they found both types of terbium exist at intermediate temperatures, however the population of spherical terbium increases as the temperature increases.
“So, what we did was we determined how the magnetic excitations evolve from this uniaxial state into this easy plane state as a function of temperature. And the long-standing assumption of how it happens is correct,” said McQueeney. “But the nuance is that you can’t treat every terbium as being exactly the same on some timescale. Every terbium site can exist in two quantum states, uniaxial or isotropic, and if I look at a site, it’s either in one state or the other at some instant time. The probability that it’s uniaxial or isotropic depends on temperature.” We call this an orbital binary quantum alloy. More

In a peer-reviewed opinion paper publishing July 10 in the journal Patterns, researchers show that computer programs commonly used to determine if a text was written by artificial intelligence tend to falsely label articles written by non-native language speakers as AI-generated. The researchers caution against the use of such AI text detectors for their unreliability, which could have negative impacts on individuals including students and those applying for jobs.
“Our current recommendation is that we should be extremely careful about and maybe try to avoid using these detectors as much as possible,” says senior author James Zou, of Stanford University. “It can have significant consequences if these detectors are used to review things like job applications, college entrance essays or high school assignments.”
AI tools like OpenAI’s ChatGPT chatbot can compose essays, solve science and math problems, and produce computer code. Educators across the U.S. are increasingly concerned about the use of AI in students’ work and many of them have started using GPT detectors to screen students’ assignments. These detectors are platforms that claim to be able to identify if the text is generated by AI, but their reliability and effectiveness remain untested.
Zou and his team put seven popular GPT detectors to the test. They ran 91 English essays written by non-native English speakers for a widely recognized English proficiency test, called Test of English as a Foreign Language, or TOEFL, through the detectors. These platforms incorrectly labeled more than half of the essays as AI-generated, with one detector flagging nearly 98% of these essays as written by AI. In comparison, the detectors were able to correctly classify more than 90% of essays written by eighth-grade students from the U.S. as human-generated.
Zou explains that the algorithms of these detectors work by evaluating text perplexity, which is how surprising the word choice is in an essay. “If you use common English words, the detectors will give a low perplexity score, meaning my essay is likely to be flagged as AI-generated. If you use complex and fancier words, then it’s more likely to be classified as human written by the algorithms,” he says. This is because large language models like ChatGPT are trained to generate text with low perplexity to better simulate how an average human talks, Zou adds.
As a result, simpler word choices adopted by non-native English writers would make them more vulnerable to being tagged as using AI.
The team then put the human-written TOEFL essays into ChatGPT and prompted it to edit the text using more sophisticated language, including substituting simple words with complex vocabulary. The GPT detectors tagged these AI-edited essays as human-written.
“We should be very cautious about using any of these detectors in classroom settings, because there’s still a lot of biases, and they’re easy to fool with just the minimum amount of prompt design,” Zou says. Using GPT detectors could also have implications beyond the education sector. For example, search engines like Google devalue AI-generated content, which may inadvertently silence non-native English writers.
While AI tools can have positive impacts on student learning, GPT detectors should be further enhanced and evaluated before putting into use. Zou says that training these algorithms with more diverse types of writing could be one way to improve these detectors. More