Computers Math

Poems, essays and even books — is there anything the open AI platform ChatGPT can’t handle? These new AI developments have inspired researchers at TU Delft and the Swiss technical university EPFL to dig a little deeper: For instance, can ChatGPT also design a robot? And is this a good thing for the design process, or are there risks? The researchers published their findings in Nature Machine Intelligence.
What are the greatest future challenges for humanity? This was the first question that Cosimo Della Santina, assistant professor, and PhD student Francesco Stella, both from TU Delft, and Josie Hughes from EPFL, asked ChatGPT. “We wanted ChatGPT to design not just a robot, but one that is actually useful,” says Della Santina. In the end, they chose food supply as their challenge, and as they chatted with ChatGPT, they came up with the idea of creating a tomato-harvesting robot.
Helpful suggestions
The researchers followed all of ChatGPT’s design decisions. The input proved particularly valuable in the conceptual phase, according to Stella. “ChatGPT extends the designer’s knowledge to other areas of expertise. For example, the chat robot taught us which crop would be most economically valuable to automate.” But ChatGPT also came up with useful suggestions during the implementation phase: “Make the gripper out of silicone or rubber to avoid crushing tomatoes” and “a Dynamixel motor is the best way to drive the robot.” The result of this partnership between humans and AI is a robotic arm that can harvest tomatoes.
ChatGPT as a researcher
The researchers found the collaborative design process to be positive and enriching. “However, we did find that our role as engineers shifted towards performing more technical tasks,” says Stella. In Nature Machine Intelligence, the researchers explore the varying degrees of cooperation between humans and Large Language Models (LLM), of which ChatGPT is one. In the most extreme scenario, AI provides all the input to the robot design, and the human blindly follows it. In this case, the LLM acts as the researcher and engineer, while the human acts as the manager, in charge of specifying the design objectives.
Risk of misinformation
Such an extreme scenario is not yet possible with today’s LLMs. And the question is whether it is desirable. “In fact, LLM output can be misleading if it is not verified or validated. AI bots are designed to generate the ‘most probable’ answer to a question, so there is a risk of misinformation and bias in the robotic field,” Della Santina says. Working with LLMs also raises other important issues, such as plagiarism, traceability and intellectual property.
Della Santina, Stella and Hughes will continue to use the tomato-harvesting robot in their research on robotics. They are also continuing their study of LLMs to design new robots. Specifically, they are looking at the autonomy of AIs in designing their own bodies. “Ultimately an open question for the future of our field is how LLMs can be used to assist robot developers without limiting the creativity and innovation needed for robotics to rise to the challenges of the 21st century,” Stella concludes. More

Significantly improved electric vehicle (EV) batteries could be a step closer thanks to a new study led by University of Oxford researchers, published today in Nature. Using advanced imaging techniques, this revealed mechanisms which cause lithium metal solid-state batteries (Li-SSBs) to fail. If these can be overcome, solid-state batteries using lithium metal anodes could deliver a step-change improvement in EV battery range, safety and performance, and help advance electrically powered aviation.
One of the co-lead authors of the study Dominic Melvin, a PhD student in the University of Oxford’s Department of Materials, said: ‘Progressing solid-state batteries with lithium metal anodes is one of the most important challenges facing the advancement of battery technologies. While lithium-ion batteries of today will continue to improve, research into solid-state batteries has the potential to be high-reward and a gamechanger technology.’
Li-SSBs are distinct from other batteries because they replace the flammable liquid electrolyte in conventional batteries with a solid electrolyte and use lithium metal as the anode (negative electrode). The use of the solid electrolyte improves the safety, and the use of lithium metal means more energy can be stored. A critical challenge with Li-SSBs, however, is that they are prone to short circuit when charging due to the growth of ‘dendrites’: filaments of lithium metal that crack through the ceramic electrolyte. As part of the Faraday Institution’s SOLBAT project, researchers from the University of Oxford’s Departments of Materials, Chemistry and Engineering Science, have led a series of in-depth investigations to understand more about how this short-circuiting happens.
In this latest study, the group used an advanced imaging technique called X-ray computed tomography at Diamond Light Source to visualise dendrite failure in unprecedented detail during the charging process. The new imaging study revealed that the initiation and propagation of the dendrite cracks are separate processes, driven by distinct underlying mechanisms. Dendrite cracks initiate when lithium accumulates in sub-surface pores. When the pores become full, further charging of the battery increases the pressure, leading to cracking. In contrast, propagation occurs with lithium only partially filling the crack, through a wedge-opening mechanism which drives the crack open from the rear.
This new understanding points the way forward to overcoming the technological challenges of Li-SSBs. Dominic Melvin said: ‘For instance, while pressure at the lithium anode can be good to avoid gaps developing at the interface with the solid electrolyte on discharge, our results demonstrate that too much pressure can be detrimental, making dendrite propagation and short-circuit on charging more likely.’
Sir Peter Bruce, Wolfson Chair, Professor of Materials at the University of Oxford, Chief Scientist of the Faraday Institution, and corresponding author of the study, said: ‘The process by which a soft metal such as lithium can penetrate a highly dense hard ceramic electrolyte has proved challenging to understand with many important contributions by excellent scientists around the world. We hope the additional insights we have gained will help the progress of solid-state battery research towards a practical device.’
According to a recent report by the Faraday Institution, SSBs may satisfy 50% of global demand for batteries in consumer electronics, 30% in transportation, and over 10% in aircraft by 2040.
Professor Pam Thomas, CEO, Faraday Institution, said: ‘SOLBAT researchers continue to develop a mechanistic understanding of solid-state battery failure — one hurdle that needs to be overcome before high-power batteries with commercially relevant performance could be realised for automotive applications. The project is informing strategies that cell manufacturers might use to avoid cell failure for this technology. This application-inspired research is a prime example of the type of scientific advances that the Faraday Institution was set up to drive.’ More

The debut of artificial intelligence chatbot ChatGPT has set the world abuzz with its ability to churn out human-like text and conversations. Still, many telltale signs can help us distinguish AI chatbots from humans, according to a study published on June 7 in the journal Cell Reports Physical Science. Based on the signs, the researchers developed a tool to identify AI-generated academic science writing with over 99% accuracy.
“We tried hard to create an accessible method so that with little guidance, even high school students could build an AI detector for different types of writing,” says first author Heather Desaire, a professor at the University of Kansas. “There is a need to address AI writing, and people don’t need a computer science degree to contribute to this field.”
“Right now, there are some pretty glaring problems with AI writing,” says Desaire. “One of the biggest problems is that it assembles text from many sources and there isn’t any kind of accuracy check — it’s kind of like the game Two Truths and a Lie.”
Although many AI text detectors are available online and perform fairly well, they weren’t built specifically for academic writing. To fill the gap, the team aimed to build a tool with better performance precisely for this purpose. They focused on a type of article called perspectives, which provide an overview of specific research topics written by scientists. The team selected 64 perspectives and created 128 ChatGPT-generated articles on the same research topics to train the model. When they compared the articles, they found an indicator of AI writing — predictability.
Contrary to AI, humans have more complex paragraph structures, varying in the number of sentences and total words per paragraph, as well as fluctuating sentence length. Preferences in punctuation marks and vocabulary are also a giveaway. For example, scientists gravitate towards words like “however,” “but” and “although,” while ChatGPT often uses “others” and “researchers” in writing. The team tallied 20 characteristics for the model to look out for.
When tested, the model aced a 100% accuracy rate at weeding out AI-generated full perspective articles from those written by humans. For identifying individual paragraphs within the article, the model had an accuracy rate of 92%. The research team’s model also outperformed an available AI text detector on the market by a wide margin on similar tests.
Next, the team plans to determine the scope of the model’s applicability. They want to test it on more extensive datasets and across different types of academic science writing. As AI chatbots advance and become more sophisticated, the researchers also want to know if their model will stand.
“The first thing people want to know when they hear about the research is ‘Can I use this to tell if my students actually wrote their paper?'” said Desaire. While the model is highly skilled at distinguishing between AI and scientists, Desaire says it was not designed to catch AI-generated student essays for educators. However, she notes that people can easily replicate their methods to build models for their own purposes. More

An international research team led by the Hong Kong University of Science and Technology (HKUST) has developed an artificial intelligence (AI)-based model that uses genetic information to predict an individual’s risk of developing Alzheimer’s disease (AD) well before symptoms occur. This groundbreaking study paves the way for using deep learning methods to predict the risks of diseases and uncover their molecular mechanisms; this could revolutionize the diagnosis of, interventions for, and clinical research on AD and other common diseases such as cardiovascular diseases.
Researchers led by HKUST’s President, Prof. Nancy IP, in collaboration with the Chair Professor and Director of HKUST’s Big Data Institute, Prof. CHEN Lei, investigated whether AI — specifically deep learning models — can model AD risk using genetic information. The team established one of the first deep learning models for estimating AD polygenic risks in both European-descent and Chinese populations. Compared to other models, these deep learning models more accurately classify patients with AD and stratify individuals into distinct groups based on disease risks associated with alterations of various biological processes.
In current daily practice, AD is diagnosed clinically, using various means including cognitive tests and brain imaging, but often when patients are showing symptoms, it is already well past the optimal intervention window. Therefore, early forecasting of AD risk can greatly aid diagnosis and the development of intervention strategies. By combining the new deep learning model with genetic testing, an individual’s lifetime risk of developing AD can be estimated with more than 70% accuracy.
AD is a heritable disorder that can be attributed to genomic variants. As these variants are present from birth and remain constant throughout life, examining an individual’s DNA information can help predict their relative risk of developing AD, thereby enabling early intervention and timely management. While FDA-approved genetic testing for the APOE-?4 genetic variant can estimate AD risk, it may be insufficient to identify high-risk individuals, because multiple genetic risks contribute to the disease. Therefore, it is essential to develop tests that integrate information from multiple AD risk genes to accurately determine an individual’s relative risk of developing AD over their lifetime.
“Our study demonstrates the efficacy of deep learning methods for genetic research and risk prediction for Alzheimer’s disease. This breakthrough will greatly accelerate population-scale screening and staging of Alzheimer’s disease risk. Besides risk prediction, this approach supports the grouping of individuals according to their disease risk and provides insights into the mechanisms that contribute to the onset and progression of the disease,” said Prof. Nancy Ip.
Meanwhile, Prof. Chen Lei remarked that, “this study exemplifies how the application of AI to the biological sciences can significantly benefit biomedical and disease-related studies. By utilizing a neural network, we effectively captured nonlinearity in high-dimensional genomic data, which improved the accuracy of Alzheimer’s disease risk prediction. In addition, through AI-based data analysis without human supervision, we categorized at-risk individuals into subgroups, which revealed insights into the underlying disease mechanisms. Our research also highlights how AI can elegantly, efficiently, and effectively address interdisciplinary challenges. I firmly believe that AI will play a vital role in various healthcare fields in the near future.”
The study was conducted in collaboration with researchers at the Shenzhen Institute of Advanced Technology and University College London as well as clinicians at local Hong Kong hospitals including Prince of Wales Hospital and Queen Elizabeth Hospital. The findings were recently published in Communications Medicine. The research team is now refining the model and aims to ultimately incorporate it into standard screening workflows.
AD, which affects over 50 million people worldwide, is a fatal disease that involves cognitive dysfunction and the loss of brain cells. Its symptoms include progressive memory loss as well as impaired movement, reasoning, and judgment. More

Researchers from University of St. Gallen and Columbia Business School published a new Journal of Marketing article that examines how the perceived meaning of manual labor can help predict the adoption of autonomous products.
The study, forthcoming in the Journal of Marketing, is titled “Meaning of Manual Labor Impedes Consumer Adoption of Autonomous Products” and is authored by Emanuel de Bellis, Gita Venkataramani Johar, and Nicola Poletti.
Whether it is cleaning homes or mowing lawns, consumers increasingly delegate manual tasks to autonomous products. These gadgets operate without human oversight and free consumers from mundane chores. However, anecdotal evidence suggests that people feel a sense of satisfaction when they complete household chores. Are autonomous products such as robot vacuums and cooking machines depriving consumers from meaningful experiences?
This new research shows that, despite unquestionable benefits such as gains in efficiency and convenience, autonomous products strip away a source of meaning in life. As a result, consumers are hesitant to buy these products.
The researchers argue that manual labor is an important source of meaning in life. This is in line with research showing that everyday tasks have value — chores such as cleaning may not make us happy, but they add meaning to our lives. As de Bellis explains, “Our studies show that ‘meaning of manual labor’ causes consumers to reject autonomous products. For example, these consumers have a more negative attitude toward autonomous products and are also more prone to believe in the disadvantages of autonomous products relative to their advantages.”
Highlight Saving Time for Other Meaningful Tasks
On one hand, autonomous products take over tasks from consumers, typically leading to a reduction in manual labor and hence in the ability to derive meaning from manual tasks. On the other hand, by taking over manual tasks, autonomous products provide consumers with the opportunity to spend time on other, potentially more meaningful, tasks and activities. “We suggest that companies highlight so-called alternative sources of meaning in life, which should reduce consumers’ need to derive meaning specifically from manual tasks. Highlighting other sources of meaning, such as through family or hobbies, at the time of the adoption decision should counteract the negative effect on autonomous product adoption,” says Johar.
In fact, a key value proposition for many of these technologies is that they free up time. iRobot claims that its robotic vacuum cleaner Roomba saves owners as much as 110 hours of cleaning a year. Some companies go even a step further by suggesting what consumers could do with their freed-up time. For example, German home appliance company Vorwerk promotes its cooking machine Thermomix with “more family time” and “Thermomix does the work so you can make time for what matters most.” Instead of promoting the quality of task completion (i.e., cooking a delicious meal), the company emphasizes that consumers can spend time on other, arguably more meaningful, activities.
This study demonstrates that the perceived meaning of manual labor (MML) — a novel concept introduced by the researchers — is key to predicting the adoption of autonomous products. Poletti says that “Consumers with a high MML tend to resist the delegation of manual tasks to autonomous products, irrespective of whether these tasks are central to one’s identity or not. Marketers can start by segmenting consumers into high and low MML consumers.” Unlike other personality variables that can only be reliably measured using complex psychometric scales, the extent of consumers’ MML might be assessed simply by observing their behavioral characteristics, such as whether consumers tend to do the dishes by hand, whether they prefer a manual car transmission, or what type of activities and hobbies they pursue. Activities like woodworking, cookery, painting, and fishing are likely predictors of high MML. Similarly, companies can measure likes on social media for specific activities and hobbies that involve manual labor. Finally, practitioners can ask consumers to rate the degree to which manual versus cognitive tasks are meaningful to them. Having segmented consumers according to their MML, marketers can better target and focus their messages and efforts.
In promotions, firms can highlight the meaningful time consumers gain with the use of autonomous products (e.g., “this product allows you to spend time on more meaningful tasks and pursuits than cleaning”). Such an intervention can prevent the detrimental effects of meaning of manual labor on autonomous product adoption. More

A simple sponge has improved how robots grasp, scientists from the University of Bristol have found.
This easy-to-make sponge-jamming device can help stiff robots handle delicate items carefully by mimicking the nuanced touch, or variable stiffness, of a human.
Robots can skip, jump and do somersaults, but they’re too rigid to hold an egg easily. Variable-stiffness devices are potential solutions for contact compliance on hard robots to reduce damage, or for improving the load capacity of soft robots.
This study, published at the IEEE International Conference on Robotics and Automation (ICRA) 2023, shows that variable stiffness can be achieved by a silicone sponge.
Lead author Tianqi Yue from Bristol’s Department of Engineering Mathematics explained: “Stiffness, also known as softness, is important in contact scenarios.
“Robotic arms are too rigid so they cannot make such a soft human-like grasp on delicate objects, for example, an egg.

“What makes humans different from robotic arms is that we have soft tissues enclosing rigid bones, which act as a natural mitigating mechanism.
“In this paper, we managed to develop a soft device with variable stiffness, to be mounted on the end robotic arm for making the robot-object contact safe.”
Silicone sponge is a cheap and easy-to-fabricate material. It is a porous elastomer just like the cleaning sponge used in everyday tasks.
By squeezing the sponge, the sponge stiffens which is why it can be transformed into a variable-stiffness device.
This device could be used in industrial robots in scenarios including gripping jellies, eggs and other fragile substances. It can also be used in service robots to make human-robot interaction safer.
Mr Yue added: “We managed to use a sponge to make a cheap and nimble but effective device that can help robots achieve soft contact with objects. The great potential comes from its low cost and light weight.
“We believe this silicone-sponge based variable-stiffness device will provide a novel solution in industry and healthcare, for example, tunable-stiffness requirement on robotic polishing and ultrasound imaging.”
The team will now look at making the device achieve variable stiffness in multiple directions, including rotation. More

A University of Minnesota Twin Cities-led team has developed a new superconducting diode, a key component in electronic devices, that could help scale up quantum computers for industry use and improve the performance of artificial intelligence systems. Compared to other superconducting diodes, the researchers’ device is more energy efficient; can process multiple electrical signals at a time; and contains a series of gates to control the flow of energy, a feature that has never before been integrated into a superconducting diode.
The paper is published in Nature Communications, a peer-reviewed scientific journal that covers the natural sciences and engineering.
A diode allows current to flow one way but not the other in an electrical circuit. It’s essentially half of a transistor, the main element in computer chips. Diodes are typically made with semiconductors, but researchers are interested in making them with superconductors, which have the ability to transfer energy without losing any power along the way.
“We want to make computers more powerful, but there are some hard limits we are going to hit soon with our current materials and fabrication methods,” said Vlad Pribiag, senior author of the paper and an associate professor in the University of Minnesota School of Physics and Astronomy. “We need new ways to develop computers, and one of the biggest challenges for increasing computing power right now is that they dissipate so much energy. So, we’re thinking of ways that superconducting technologies might help with that.”
The University of Minnesota researchers created the device using three Josephson junctions, which are made by sandwiching pieces of non-superconducting material between superconductors. In this case, the researchers connected the superconductors with layers of semiconductors. The device’s unique design allows the researchers to use voltage to control the behavior of the device.
Their device also has the ability to process multiple signal inputs, whereas typical diodes can only handle one input and one output. This feature could have applications in neuromorphic computing, a method of engineering electrical circuits to mimic the way neurons function in the brain to enhance the performance of artificial intelligence systems.
“The device we’ve made has close to the highest energy efficiency that has ever been shown, and for the first time, we’ve shown that you can add gates and apply electric fields to tune this effect,” explained Mohit Gupta, first author of the paper and a Ph.D. student in the University of Minnesota School of Physics and Astronomy. “Other researchers have made superconducting devices before, but the materials they’ve used have been very difficult to fabricate. Our design uses materials that are more industry-friendly and deliver new functionalities.”
The method the researchers used can, in principle, be used with any type of superconductor, making it more versatile and easier to use than other techniques in the field. Because of these qualities, their device is more compatible for industry applications and could help scale up the development of quantum computers for wider use.
“Right now, all the quantum computing machines out there are very basic relative to the needs of real-world applications,” Pribiag said. “Scaling up is necessary in order to have a computer that’s powerful enough to tackle useful, complex problems. A lot of people are researching algorithms and usage cases for computers or AI machines that could potentially outperform classical computers. Here, we’re developing the hardware that could enable quantum computers to implement these algorithms. This shows the power of universities seeding these ideas that eventually make their way to industry and are integrated into practical machines.”
This research was funded primarily by the United States Department of Energy with partial support from Microsoft Research and the National Science Foundation.
In addition to Pribiag and Gupta, the research team included University of Minnesota School of Physics and Astronomy graduate student Gino Graziano and University of California, Santa Barbara researchers Mihir Pendharkar, Jason Dong, Connor Dempsey, and Chris Palmstrøm. More

Researchers have developed a new artificial intelligence (AI) framework that is better than previous technologies at analyzing and categorizing dialogue between individuals, with the goal of improving team training technologies. The framework will enable training technologies to better understand how well individuals are coordinating with one another and working as part of a team.
“There is a great deal of interest in developing AI-powered training technologies that can understand teamwork dynamics and modify their training to foster improved collaboration among team members,” says Wookhee Min, co-author of a paper on the work and a research scientist at North Carolina State University. “However, previous AI architectures have struggled to accurately assess the content of what team members are sharing with each other when they communicate.”
“We’ve developed a new framework that significantly improves the ability of AI to analyze communication between team members,” says Jay Pande, first author of the paper and a Ph.D. student at NC State. “This is a significant step forward for the development of adaptive training technologies that aim to facilitate effective team communication and collaboration.”
The new AI framework builds on a powerful deep learning model that was trained on a large, text-based language dataset. This model, called the Text-to-Text Transfer Transformer (T5), was then customized using data collected during squad-level training exercises conducted by the U.S. Army.
“We modified the T5 model to use contextual features of the team — such as the speaker’s role — to more accurately analyze team communication,” Min says. “That context can be important. For example, something a team leader says may need to be viewed differently than something another team member says.”
To test the performance of the new framework, the researchers compared it to two previous AI technologies. Specifically, the researchers tested the ability of all three AI technologies to understand the dialogue within a squad of six soldiers during a training exercise.

The AI framework was tasked with two things: classify what sort of dialogue was taking place, and follow the flow of information within the squad. Classifying the dialogue refers to determining the purpose of what was being said. For example, was someone requesting information, providing information, or issuing a command? Following the flow of information refers to how information was being shared within the team. For example, was information being passed up or down the chain of command?
“We found that the new framework performed substantially better than the previous AI technologies,” Pande says.
“One of the things that was particularly promising was that we trained our framework using data from one training mission, but tested the model’s performance using data from a different training mission,” Min says. “And the boost in performance over the previous AI models was notable — even though we were testing the model in a new set of circumstances.”
The researchers also note that they were able to achieve these results using a relatively small version of the T5 model. That’s important, because it means that they can get analysis in fractions of a second without a supercomputer.
“One next step for this work includes exploring the extent to which the new framework can be applied to a variety of other training scenarios,” Pande says.
“We tested the new framework with training data that was transcribed from audio files into text by humans,” Min says. “Another next step will involve integrating the framework with an AI model that transcribes audio data into text, so that we can assess the ability of this technology to analyze team communication data in real time. This will likely involve improving the framework’s ability to deal with noises and errors as the AI transcribes audio data.”
The paper, “Robust Team Communication Analytics with Transformer-Based Dialogue Modeling,” will be presented at the 24th International Conference on Artificial Intelligence in Education (AIED 2023), which will be held July 3-7 in Tokyo, Japan. The paper was co-authored by Jason Saville, a former graduate student at NC State; James Lester, the Goodnight Distinguished University Professor in Artificial Intelligence and Machine Learning at NC State; and Randall Spain of the U.S. Army Combat Capabilities Development Command (DEVCOM). Soldier Center.
This research was sponsored by the U.S. Army DEVCOM, Soldier Center under cooperative agreement W912CG-19-2-0001. More