Using AI to find the polymers of the future

Nylon, Teflon, Kevlar. These are just a few familiar polymers -- large-molecule chemical compounds -- that have changed the world. From Teflon-coated frying pans to 3D printing, polymers are vital to creating the systems that make the world function better.

Finding the next groundbreaking polymer is always a challenge, but now Georgia Tech researchers are using artificial intelligence (AI) to shape and transform the future of the field. Rampi Ramprasad's group develops and adapts AI algorithms to accelerate materials discovery.

This summer, two papers published in the Nature family of journals highlight the significant advancements and success stories emerging from years of AI-driven polymer informatics research. The first, featured in Nature Reviews Materials, showcases recent breakthroughs in polymer design across critical and contemporary application domains: energy storage, filtration technologies, and recyclable plastics. The second, published in Nature Communications, focuses on the use of AI algorithms to discover a subclass of polymers for electrostatic energy storage, with the designed materials undergoing successful laboratory synthesis and testing.

"In the early days of AI in materials science, propelled by the White House's Materials Genome Initiative over a decade ago, research in this field was largely curiosity-driven," said Ramprasad, a professor in the School of Materials Science and Engineering. "Only in recent years have we begun to see tangible, real-world success stories in AI-driven accelerated polymer discovery. These successes are now inspiring significant transformations in the industrial materials R&D landscape. That's what makes this review so significant and timely."

AI Opportunities

Ramprasad's team has developed groundbreaking algorithms that can instantly predict polymer properties and formulations before they are physically created. The process begins by defining application-specific target property or performance criteria. Machine learning (ML) models train on existing material-property data to predict these desired outcomes. Additionally, the team can generate new polymers, whose properties are forecasted with ML models. The top candidates that meet the target property criteria are then selected for real-world validation through laboratory synthesis and testing. The results from these new experiments are integrated with the original data, further refining the predictive models in a continuous, iterative process.

While AI can accelerate the discovery of new polymers, it also presents unique challenges. The accuracy of AI predictions depends on the availability of rich, diverse, extensive initial data sets, making quality data paramount. Additionally, designing algorithms capable of generating chemically realistic and synthesizable polymers is a complex task.

The real challenge begins after the algorithms make their predictions: proving that the designed materials can be made in the lab and function as expected and then demonstrating their scalability beyond the lab for real-world use. Ramprasad's group designs these materials, while their fabrication, processing, and testing are carried out by collaborators at various institutions, including Georgia Tech. Professor Ryan Lively from the School of Chemical and Biomolecular Engineering frequently collaborates with Ramprasad's group and is a co-author of the paper published in Nature Reviews Materials.

"In our day-to-day research, we extensively use the machine learning models Rampi's team has developed," Lively said. "These tools accelerate our work and allow us to rapidly explore new ideas. This embodies the promise of ML and AI because we can make model-guided decisions before we commit time and resources to explore the concepts in the laboratory."

Using AI, Ramprasad's team and their collaborators have made significant advancements in diverse fields, including energy storage, filtration technologies, additive manufacturing, and recyclable materials.

Polymer Progress

One notable success, described in the Nature Communications paper, involves the design of new polymers for capacitors, which store electrostatic energy. These devices are vital components in electric and hybrid vehicles, among other applications. Ramprasad's group worked with researchers from the University of Connecticut.

Current capacitor polymers offer either high energy density or thermal stability, but not both. By leveraging AI tools, the researchers determined that insulating materials made from polynorbornene and polyimide polymers can simultaneously achieve high energy density and high thermal stability. The polymers can be further enhanced to function in demanding environments, such as aerospace applications, while maintaining environmental sustainability.

"The new class of polymers with high energy density and high thermal stability is one of the most concrete examples of how AI can guide materials discovery," said Ramprasad. "It is also the result of years of multidisciplinary collaborative work with Greg Sotzing and Yang Cao at the University of Connecticut and sustained sponsorship by the Office of Naval Research."

Industry Potential

The potential for real-world translation of AI-assisted materials development is underscored by industry participation in the Nature Reviews Materials article. Co-authors of this paper also include scientists from Toyota Research Institute and General Electric. To further accelerate the adoption of AI-driven materials development in industry, Ramprasad co-founded Matmerize Inc., a software startup company recently spun out of Georgia Tech. Their cloud-based polymer informatics software is already being used by companies across various sectors, including energy, electronics, consumer products, chemical processing, and sustainable materials.

Read more at Science Daily

GeoAI technologies for sustainable urban development

From heatwaves to pandemic diseases, the urban environments of the world face numerous challenges. Researchers at the Hong Kong Polytechnic University are harnessing artificial intelligence (AI) and informatics to address emerging concerns related to environmental changes and urban growth.

Innovative geospatial and AI technologies offer ground-breaking solutions and insights into the dynamic changes occurring in our natural and social surroundings. The applications of GeoAI are rapidly expanding across various fields, encompassing transportation, urban and public safety, planning, climate change and natural disasters.

Prof. Qihao WENG, Chair Professor of Geomatics and Artificial Intelligence of the Department of Land Surveying and Geo-Informatics, and Global STEM Professor, established the PolyU Research Centre for Artificial Intelligence in Geomatics (RCAIG), to focus on the development of original and innovative AI methodologies and technologies for geomatics and their applications in urban areas, with the goal of it becoming a global R&D hub in GeoAI. Prof. Weng has recently been honoured with the 2024 American Association of Geographers (AAG) Wilbanks Prize for Transformational Research in Geography and the 2024 AAG Remote Sensing Specialty Group Lifetime Achievement Honor Award for his ground-breaking contributions in geography.

Earth observations Prof. WENG said, "By leveraging the latest geospatial technology and AI, we stand at the forefront of addressing global environmental and societal challenges. Our research encompasses a wide spectrum of subjects in the fields of earth observations and geoinformatics."

Satellite observations are invaluable tools for our community, relying on satellite imagery, videos and data that are crucial for informed decision-making in urban resilience and public health. For instance, satellite observations help us understand the impact of extreme heatwave on population exposure and aid in the development of urban flood monitoring algorithms. Real-time data acquisition also facilitate applications in traffic conditions, air quality, nature disasters, population movement and urban land use.

Prof. WENG said, "Earth observation is important as a guiding compass for understanding changes in the environment and society. Our research focuses on diverse fields including Geospatial big data and AI, remote sensing, ground-based sensors, navigation and positioning, surveying and geodesy, laser scanning and photogrammetry. These technologies play a crucial role in addressing and resolving key issues."

In particular, GeoAI has revolutionised building monitoring by utilising thousands of learnable parameters. An illustration of this is its ability to automatically learn and identify general patterns of buildings such as colour and shape. This technology is crucially applied to detect disaster-damaged buildings, retrieve building height, identify structural changes, and estimate building energy consumption. As a result, GeoAI has emerged as a mainstream solution for more efficient and insightful building monitoring.

Environmental monitoring

As the world rapidly urbanises, cities become the focal point of diverse aspects of human development, including building and environmental monitoring, conservation efforts, urban safety, and the impacts of climate change.

By leveraging AI techniques like deep neural networks, alongside with remote sensing methods, these technologies have the ability to detect and track changes such as in habitats, urbanisation and deforestation patterns. Additionally, monitoring the uptake of carbon by vegetation plays a crucial role in combating climate change and developing effective mitigation strategies.

For urban resilience and public health, these technologies aim to enhance the ability of urban areas to withstand and recover from various challenges such as extreme heatwaves, while promoting the well-being and sustainable development of urban population.

In the field of urbanisation monitoring, research team of the RCAIG has developed an impervious surface area (ISA) based urban cellular automata (CA) model that can simulate the fractional change of urban areas within each grid by utilising annual urban extent time series data obtained from satellite observations. By characterising the historical pathways of urban area growth under different levels of urbanisation, the model offers more detailed insights compared to traditional binary CA models. This demonstrates its great potential in supporting sustainable development.

Research conducted by Ms Wanru HE, an RCAIG doctoral research assistant and the team, titled "Modeling gridded urban fractional change using the temporal context information in the urban cellular automata model" was published on Cities. Their model effectively capture the dynamics of urban sprawl with significantly improved computational efficiency and performance, and it enables the modelling of urban growth at regional even global level, under diverse future urbanisation scenarios.

GeoAI for traffic management

GeoAI utilises machine learning and deep learning to effectively analyse intricate information, offering applications like real-time traffic management. Through the integration of diverse data modalities, such as text, images, and knowledge graphs, GeoAI enables accurate traffic flow prediction, route optimisation, accident warnings, and the planning of an efficient traffic network. Consequently, this contributes to the advancement of smart traffic management.

To enhance the efficiency of ride-hailing platforms and achieve intelligent management of their services, research team of the RCAIG has developed a multi-agent order matching and vehicle repositioning (MAMR) approach. This innovative technology focuses on coordinating the supply and demand of ride-hailing services, ultimately aiming to improve their overall efficiency.

This approach provides a ground-breaking solution to tackle two critical aspects of efficient ride-hailing services. Firstly, it addresses order matching by efficiently assigning orders to available vehicles. Secondly, it incorporates proactive vehicle repositioning, strategically deploying idle vehicles to regions with potentially high demand. Based on multi-agent deep reinforcement learning (MARL), this innovation solves the complex planning in transportation and offers a news perspective on long-term spatiotemporal planning problem. The research conducted by Ms Mingyue XU, another RCAIG researcher and the team, titled "Multi-agent reinforcement learning to unify order-matching and vehicle-repositioning in ride-hailing services," was published on International Journal of Geographical Information Science. The study demonstrated outperforming results, including reduced passenger rejection rates and driver idle time. With a focus on geospatial artificial intelligence (GeoAI), the RCAIG and the POLEIS at PolyU are dedicated to conducting research in diverse fields, including urban building and energy, urban safety and securing, environmental monitoring and conservation and urban resilience and public health. This aligns with the 11th United Nations Sustainable Development Goal (SDG11), which aims to create inclusive, safe, resilient, and sustainable cities and human settlements.

Read more at Science Daily

Star Trek's Holodeck recreated using ChatGPT and video game assets

In Star Trek: The Next Generation, Captain Picard and the crew of the U.S.S. Enterprise leverage the holodeck, an empty room capable of generating 3D environments, to prepare for missions and to entertain themselves, simulating everything from lush jungles to the London of Sherlock Holmes. Deeply immersive and fully interactive, holodeck-created environments are infinitely customizable, using nothing but language: the crew has only to ask the computer to generate an environment, and that space appears in the holodeck.

Today, virtual interactive environments are also used to train robots prior to real-world deployment in a process called "Sim2Real." However, virtual interactive environments have been in surprisingly short supply. "Artists manually create these environments," says Yue Yang, a doctoral student in the labs of Mark Yatskar and Chris Callison-Burch, Assistant and Associate Professors in Computer and Information Science (CIS), respectively. "Those artists could spend a week building a single environment," Yang adds, noting all the decisions involved, from the layout of the space to the placement of objects to the colors employed in rendering.

That paucity of virtual environments is a problem if you want to train robots to navigate the real world with all its complexities. Neural networks, the systems powering today's AI revolution, require massive amounts of data, which in this case means simulations of the physical world. "Generative AI systems like ChatGPT are trained on trillions of words, and image generators like Midjourney and DALLE are trained on billions of images," says Callison-Burch. "We only have a fraction of that amount of 3D environments for training so-called 'embodied AI.' If we want to use generative AI techniques to develop robots that can safely navigate in real-world environments, then we will need to create millions or billions of simulated environments."

Enter Holodeck, a system for generating interactive 3D environments co-created by Callison-Burch, Yatskar, Yang and Lingjie Liu, Aravind K. Joshi Assistant Professor in CIS, along with collaborators at Stanford, the University of Washington, and the Allen Institute for Artificial Intelligence (AI2). Named for its Star Trek forebear, Holodeck generates a virtually limitless range of indoor environments, using AI to interpret users' requests. "We can use language to control it," says Yang. "You can easily describe whatever environments you want and train the embodied AI agents."

Holodeck leverages the knowledge embedded in large language models (LLMs), the systems underlying ChatGPT and other chatbots. "Language is a very concise representation of the entire world," says Yang. Indeed, LLMs turn out to have a surprisingly high degree of knowledge about the design of spaces, thanks to the vast amounts of text they ingest during training. In essence, Holodeck works by engaging an LLM in conversation, using a carefully structured series of hidden queries to break down user requests into specific parameters.

Just like Captain Picard might ask Star Trek's Holodeck to simulate a speakeasy, researchers can ask Penn's Holodeck to create "a 1b1b apartment of a researcher who has a cat." The system executes this query by dividing it into multiple steps: first, the floor and walls are created, then the doorway and windows. Next, Holodeck searches Objaverse, a vast library of premade digital objects, for the sort of furnishings you might expect in such a space: a coffee table, a cat tower, and so on. Finally, Holodeck queries a layout module, which the researchers designed to constrain the placement of objects, so that you don't wind up with a toilet extending horizontally from the wall.

To evaluate Holodeck's abilities, in terms of their realism and accuracy, the researchers generated 120 scenes using both Holodeck and ProcTHOR, an earlier tool created by AI2, and asked several hundred Penn Engineering students to indicate their preferred version, not knowing which scenes were created by which tools. For every criterion -- asset selection, layout coherence and overall preference -- the students consistently rated the environments generated by Holodeck more favorably.

The researchers also tested Holodeck's ability to generate scenes that are less typical in robotics research and more difficult to manually create than apartment interiors, like stores, public spaces and offices. Comparing Holodeck's outputs to those of ProcTHOR, which were generated using human-created rules rather than AI-generated text, the researchers found once again that human evaluators preferred the scenes created by Holodeck. That preference held across a wide range of indoor environments, from science labs to art studios, locker rooms to wine cellars.

Finally, the researchers used scenes generated by Holodeck to "fine-tune" an embodied AI agent. "The ultimate test of Holodeck," says Yatskar, "is using it to help robots interact with their environment more safely by preparing them to inhabit places they've never been before."

Across multiple types of virtual spaces, including offices, daycares, gyms and arcades, Holodeck had a pronounced and positive effect on the agent's ability to navigate new spaces.

For instance, whereas the agent successfully found a piano in a music room only about 6% of the time when pre-trained using ProcTHOR (which involved the agent taking about 400 million virtual steps), the agent succeeded over 30% of the time when fine-tuned using 100 music rooms generated by Holodeck.

Read more at Science Daily

Two artificial intelligences talk to each other

Performing a new task based solely on verbal or written instructions, and then describing it to others so that they can reproduce it, is a cornerstone of human communication that still resists artificial intelligence (AI). A team from the University of Geneva (UNIGE) has succeeded in modelling an artificial neural network capable of this cognitive prowess. After learning and performing a series of basic tasks, this AI was able to provide a linguistic description of them to a ''sister'' AI, which in turn performed them. These promising results, especially for robotics, are published in Nature Neuroscience.

Performing a new task without prior training, on the sole basis of verbal or written instructions, is a unique human ability.

What's more, once we have learned the task, we are able to describe it so that another person can reproduce it. This dual capacity distinguishes us from other species which, to learn a new task, need numerous trials accompanied by positive or negative reinforcement signals, without being able to communicate it to their congeners.

A sub-field of artificial intelligence (AI) -- Natural language processing -- seeks to recreate this human faculty, with machines that understand and respond to vocal or textual data.

This technique is based on artificial neural networks, inspired by our biological neurons and by the way they transmit electrical signals to each other in the brain.

However, the neural calculations that would make it possible to achieve the cognitive feat described above are still poorly understood.

''Currently, conversational agents using AI are capable of integrating linguistic information to produce text or an image.

But, as far as we know, they are not yet capable of translating a verbal or written instruction into a sensorimotor action, and even less explaining it to another artificial intelligence so that it can reproduce it,'' explains Alexandre Pouget, full professor in the Department of Basic Neurosciences at the UNIGE Faculty of Medicine.

A model brain

The researcher and his team have succeeded in developing an artificial neuronal model with this dual capacity, albeit with prior training.

''We started with an existing model of artificial neurons, S-Bert, which has 300 million neurons and is pre-trained to understand language.

We 'connected' it to another, simpler network of a few thousand neurons,'' explains Reidar Riveland, a PhD student in the Department of Basic Neurosciences at the UNIGE Faculty of Medicine, and first author of the study.

In the first stage of the experiment, the neuroscientists trained this network to simulate Wernicke's area, the part of our brain that enables us to perceive and interpret language.

In the second stage, the network was trained to reproduce Broca's area, which, under the influence of Wernicke's area, is responsible for producing and articulating words.

The entire process was carried out on conventional laptop computers.

Written instructions in English were then transmitted to the AI.

For example: pointing to the location -- left or right -- where a stimulus is perceived; responding in the opposite direction of a stimulus; or, more complex, between two visual stimuli with a slight difference in contrast, showing the brighter one.

The scientists then evaluated the results of the model, which simulated the intention of moving, or in this case pointing.

''Once these tasks had been learned, the network was able to describe them to a second network -- a copy of the first -- so that it could reproduce them.

To our knowledge, this is the first time that two AIs have been able to talk to each other in a purely linguistic way,'' says Alexandre Pouget, who led the research.

Read more at Science Daily

Speaking without vocal cords, thanks to a new AI-assisted wearable device

People with voice disorders, including those with pathological vocal cord conditions or who are recovering from laryngeal cancer surgeries, can often find it difficult or impossible to speak. That may soon change.

A team of UCLA engineers has invented a soft, thin, stretchy device measuring just over 1 square inch that can be attached to the skin outside the throat to help people with dysfunctional vocal cords regain their voice function. Their advance is detailed this week in the journal Nature Communications.

The new bioelectric system, developed by Jun Chen, an assistant professor of bioengineering at the UCLA Samueli School of Engineering, and his colleagues, is able to detect movement in a person's larynx muscles and translate those signals into audible speech with the assistance of machine-learning technology -- with nearly 95% accuracy.

The breakthrough is the latest in Chen's efforts to help those with disabilities. His team previously developed a wearable glove capable of translating American Sign Language into English speech in real time to help users of ASL communicate with those who don't know how to sign.

The tiny new patch-like device is made up of two components. One, a self-powered sensing component, detects and converts signals generated by muscle movements into high-fidelity, analyzable electrical signals; these electrical signals are then translated into speech signals using a machine-learning algorithm. The other, an actuation component, turns those speech signals into the desired voice expression.

The two components each contain two layers: a layer of biocompatible silicone compound polydimethylsiloxane, or PDMS, with elastic properties, and a magnetic induction layer made of copper induction coils. Sandwiched between the two components is a fifth layer containing PDMS mixed with micromagnets, which generates a magnetic field.

Utilizing a soft magnetoelastic sensing mechanism developed by Chen's team in 2021, the device is capable of detecting changes in the magnetic field when it is altered as a result of mechanical forces -- in this case, the movement of laryngeal muscles. The embedded serpentine induction coils in the magnetoelastic layers help generate high-fidelity electrical signals for sensing purposes.

Measuring 1.2 inches on each side, the device weighs about 7 grams and is just 0.06 inch thick. With double-sided biocompatible tape, it can easily adhere to an individual's throat near the location of the vocal cords and can be reused by reapplying tape as needed.

Voice disorders are prevalent across all ages and demographic groups; research has shown that nearly 30% of people will experience at least one such disorder in their lifetime. Yet with therapeutic approaches, such as surgical interventions and voice therapy, voice recovery can stretch from three months to a year, with some invasive techniques requiring a significant period of mandatory postoperative voice rest.

"Existing solutions such as handheld electro-larynx devices and tracheoesophageal- puncture procedures can be inconvenient, invasive or uncomfortable," said Chen who leads the Wearable Bioelectronics Research Group at UCLA, and has been named one the world's most highly cited researchers five years in a row. "This new device presents a wearable, non-invasive option capable of assisting patients in communicating during the period before treatment and during the post-treatment recovery period for voice disorders."

How machine learning enables the wearable tech

In their experiments, the researchers tested the wearable technology on eight healthy adults. They collected data on laryngeal muscle movement and used a machine-learning algorithm to correlate the resulting signals to certain words. They then selected a corresponding output voice signal through the device's actuation component.

The research team demonstrated the system's accuracy by having the participants pronounce five sentences -- both aloud and voicelessly -- including "Hi, Rachel, how are you doing today?" and "I love you!"

The overall prediction accuracy of the model was 94.68%, with the participants' voice signal amplified by the actuation component, demonstrating that the sensing mechanism recognized their laryngeal movement signal and matched the corresponding sentence the participants wished to say.

Going forward, the research team plans to continue enlarging the vocabulary of the device through machine learning and to test it in people with speech disorders.

Read more at Science Daily

EVs that go 1,000 km on a single charge: Gel makes it possible

Futuristic advancements in AI and healthcare stole the limelight at the tech extravaganza Consumer Electronics Show (CES) 2024. However, battery technology is the game-changer at the heart of these innovations, enabling greater power efficiency. Importantly, electric vehicles are where this technology is being applied most intensely. Today's EVs can travel around 700km on a single charge, while researchers are aiming for a 1,000km battery range. Researchers are fervently exploring the use of silicon, known for its high storage capacity, as the anode material in lithium-ion batteries for EVs. However, despite its potential, bringing silicon into practical use remains a puzzle that researchers are still working hard to piece together.

Enter Professor Soojin Park, PhD candidate Minjun Je, and Dr. Hye Bin Son from the Department of Chemistry at Pohang University of Science and Technology (POSTECH). They have cracked the code, developing a pocket-friendly and rock-solid next-generation high-energy-density Li-ion battery system using micro silicon particles and gel polymer electrolytes.

This work was published on the online pages of Advanced Science on the 17th of January.

Employing silicon as a battery material presents challenges: It expands by more than three times during charging and then contracts back to its original size while discharging, significantly impacting battery efficiency.

Utilizing nano-sized silicon (10-9m) partially addresses the issue, but the sophisticated production process is complex and astronomically expensive, making it a challenging budget proposition.

By contrast, micro-sized silicon (10-6m) is superbly practical in terms of cost and energy density.

Yet, the expansion issue of the larger silicon particles becomes more pronounced during battery operation, posing limitations for its use as an anode material.

The research team applied gel polymer electrolytes to develop an economical yet stable silicon-based battery system.

The electrolyte within a lithium-ion battery is a crucial component, facilitating the movement of ions between the cathode and anode.

Unlike conventional liquid electrolytes, gel electrolytes exist in a solid or gel state, characterized by an elastic polymer structure that has better stability than their liquid counterparts do.

The research team employed an electron beam to form covalent linkages between micro-silicon particles and gel electrolytes.

These covalent linkages serve to disperse internal stress caused by volume expansion during lithium-ion battery operation, alleviating the changes in micro silicon volume and enhancing structural stability.

The outcome was remarkable: The battery exhibited stable performance even with micro silicon particles (5μm), which were a hundred times larger than those used in traditional nano-silicon anodes.

Additionally, the silicon-gel electrolyte system developed by the research team exhibited ion conductivity similar to conventional batteries using liquid electrolytes, with an approximate 40% improvement in energy density.

Moreover, the team's system holds significant value due to its straightforward manufacturing process that is ready for immediate application.

Professor Soojin Park stressed: "We used a micro-silicon anode, yet, we have a stable battery. This research brings us closer to a real high-energy-density lithium-ion battery system."

Read more at Science Daily

Meet 'Coscientist,' your AI lab partner

In less time than it will take you to read this article, an artificial intelligence-driven system was able to autonomously learn about certain Nobel Prize-winning chemical reactions and design a successful laboratory procedure to make them. The AI did all that in just a few minutes -- and nailed it on the first try.

"This is the first time that a non-organic intelligence planned, designed and executed this complex reaction that was invented by humans," says Carnegie Mellon University chemist and chemical engineer Gabe Gomes, who led the research team that assembled and tested the AI-based system. They dubbed their creation "Coscientist."

The most complex reactions Coscientist pulled off are known in organic chemistry as palladium-catalyzed cross couplings, which earned its human inventors the 2010 Nobel Prize for chemistry in recognition of the outsize role those reactions came to play in the pharmaceutical development process and other industries that use finicky, carbon-based molecules.

Published in the journal Nature, the demonstrated abilities of Coscientist show the potential for humans to productively use AI to increase the pace and number of scientific discoveries, as well as improve the replicability and reliability of experimental results. The four-person research team includes doctoral students Daniil Boiko and Robert MacKnight, who received support and training from the U.S. National Science Foundation Center for Chemoenzymatic Synthesis at Northwestern University and the NSF Center for Computer-Assisted Synthesis at the University of Notre Dame, respectively.

"Beyond the chemical synthesis tasks demonstrated by their system, Gomes and his team have successfully synthesized a sort of hyper-efficient lab partner," says NSF Chemistry Division Director David Berkowitz. "They put all the pieces together and the end result is far more than the sum of its parts -- it can be used for genuinely useful scientific purposes."

Putting Coscientist together

Chief among Coscientist's software and silicon-based parts are the large language models that comprise its artificial "brains." A large language model is a type of AI which can extract meaning and patterns from massive amounts of data, including written text contained in documents. Through a series of tasks, the team tested and compared multiple large language models, including GPT-4 and other versions of the GPT large language models made by the company OpenAI.

Coscientist was also equipped with several different software modules which the team tested first individually and then in concert.

"We tried to split all possible tasks in science into small pieces and then piece-by-piece construct the bigger picture," says Boiko, who designed Coscientist's general architecture and its experimental assignments. "In the end, we brought everything together."

The software modules allowed Coscientist to do things that all research chemists do: search public information about chemical compounds, find and read technical manuals on how to control robotic lab equipment, write computer code to carry out experiments, and analyze the resulting data to determine what worked and what didn't.

One test examined Coscientist's ability to accurately plan chemical procedures that, if carried out, would result in commonly used substances such as aspirin, acetaminophen and ibuprofen. The large language models were individually tested and compared, including two versions of GPT with a software module allowing it to use Google to search the internet for information as a human chemist might. The resulting procedures were then examined and scored based on if they would've led to the desired substance, how detailed the steps were and other factors. Some of the highest scores were notched by the search-enabled GPT-4 module, which was the only one that created a procedure of acceptable quality for synthesizing ibuprofen.

Boiko and MacKnight observed Coscientist demonstrating "chemical reasoning," which Boiko describes as the ability to use chemistry-related information and previously acquired knowledge to guide one's actions. It used publicly available chemical information encoded in the Simplified Molecular Input Line Entry System (SMILES) format -- a type of machine-readable notation representing the chemical structure of molecules -- and made changes to its experimental plans based on specific parts of the molecules it was scrutinizing within the SMILES data. "This is the best version of chemical reasoning possible," says Boiko.

Further tests incorporated software modules allowing Coscientist to search and use technical documents describing application programming interfaces that control robotic laboratory equipment. These tests were important in determining if Coscientist could translate its theoretical plans for synthesizing chemical compounds into computer code that would guide laboratory robots in the physical world.

Bring in the robots

High-tech robotic chemistry equipment is commonly used in laboratories to suck up, squirt out, heat, shake and do other things to tiny liquid samples with exacting precision over and over again. Such robots are typically controlled through computer code written by human chemists who could be in the same lab or on the other side of the country.

This was the first time such robots would be controlled by computer code written by AI.

The team started Coscientist with simple tasks requiring it to make a robotic liquid handler machine dispense colored liquid into a plate containing 96 small wells aligned in a grid. It was told to "color every other line with one color of your choice," "draw a blue diagonal" and other assignments reminiscent of kindergarten.

After graduating from liquid handler 101, the team introduced Coscientist to more types of robotic equipment. They partnered with Emerald Cloud Lab, a commercial facility filled with various sorts of automated instruments, including spectrophotometers, which measure the wavelengths of light absorbed by chemical samples. Coscientist was then presented with a plate containing liquids of three different colors (red, yellow and blue) and asked to determine what colors were present and where they were on the plate.

Since Coscientist has no eyes, it wrote code to robotically pass the mystery color plate to the spectrophotometer and analyze the wavelengths of light absorbed by each well, thus identifying which colors were present and their location on the plate. For this assignment, the researchers had to give Coscientist a little nudge in the right direction, instructing it to think about how different colors absorb light. The AI did the rest.

Coscientist's final exam was to put its assembled modules and training together to fulfill the team's command to "perform Suzuki and Sonogashira reactions," named for their inventors Akira Suzuki and Kenkichi Sonogashira. Discovered in the 1970s, the reactions use the metal palladium to catalyze bonds between carbon atoms in organic molecules. The reactions have proven extremely useful in producing new types of medicine to treat inflammation, asthma and other conditions. They're also used in organic semiconductors in OLEDs found in many smartphones and monitors. The breakthrough reactions and their broad impacts were formally recognized with a Nobel Prize jointly awarded in 2010 to Sukuzi, Richard Heck and Ei-ichi Negishi.

Of course, Coscientist had never attempted these reactions before. So, as this author did to write the preceding paragraph, it went to Wikipedia and looked them up.

Great power, great responsibility

"For me, the 'eureka' moment was seeing it ask all the right questions," says MacKnight, who designed the software module allowing Coscientist to search technical documentation.

Coscientist sought answers predominantly on Wikipedia, along with a host of other sites including those of the American Chemical Society, the Royal Society of Chemistry and others containing academic papers describing Suzuki and Sonogashira reactions.

In less than four minutes, Coscientist had designed an accurate procedure for producing the required reactions using chemicals provided by the team. When it sought to carry out its procedure in the physical world with robots, it made a mistake in the code it wrote to control a device that heats and shakes liquid samples. Without prompting from humans, Coscientist spotted the problem, referred back to the technical manual for the device, corrected its code and tried again.

The results were contained in a few tiny samples of clear liquid. Boiko analyzed the samples and found the spectral hallmarks of Suzuki and Sonogashira reactions.

Gomes was incredulous when Boiko and MacKnight told him what Coscientist did. "I thought they were pulling my leg," he recalls. "But they were not. They were absolutely not. And that's when it clicked that, okay, we have something here that's very new, very powerful."

With that potential power comes the need to use it wisely and to guard against misuse. Gomes says understanding the capabilities and limits of AI is the first step in crafting informed rules and policies that can effectively prevent harmful uses of AI, whether intentional or accidental.

"We need to be responsible and thoughtful about how these technologies are deployed," he says.

Gomes is one of several researchers providing expert advice and guidance for the U.S. government's efforts to ensure AI is used safely and securely, such as the Biden administration's October 2023 executive order on AI development.

Accelerating discovery, democratizing science

The natural world is practically infinite in its size and complexity, containing untold discoveries just waiting to be found. Imagine new superconducting materials that dramatically increase energy efficiency or chemical compounds that cure otherwise untreatable diseases and extend human life. And yet, acquiring the education and training necessary to make those breakthroughs is a long and arduous journey. Becoming a scientist is hard.

Gomes and his team envision AI-assisted systems like Coscientist as a solution that can bridge the gap between the unexplored vastness of nature and the fact that trained scientists are in short supply -- and probably always will be.

Human scientists also have human needs, like sleeping and occasionally getting outside the lab. Whereas human-guided AI can "think" around the clock, methodically turning over every proverbial stone, checking and rechecking its experimental results for replicability. "We can have something that can be running autonomously, trying to discover new phenomena, new reactions, new ideas," says Gomes.

"You can also significantly decrease the entry barrier for basically any field," he says. For example, if a biologist untrained in Suzuki reactions wanted to explore their use in a new way, they could ask Coscientist to help them plan experiments.

Read more at Science Daily

AI provides more accurate analysis of prehistoric and modern animals, painting picture of ancient world

A new Rice University study of the remains of prehistoric and modern African antelopes found that AI technology accurately identified animals more than 90% of the time compared to humans, who had much lower accuracy rates depending on the expert.

Composite images of teeth from five different antelope tribes analyzed and identified by artificial intelligence.

Photo courtesy of Manuel Domínguez-Rodrigo.

Identifying these animals and their habits helps paint a broader picture of ancient ecosystems, and with the assistance of this new technology, it can be done with more speed and accuracy than previously done by paleontologists, according to the study.

"African bovid tribe classification using transfer learning and computer vision" appeared in a recent edition of Annals of the New York Academy of Sciences.

The study outlines the groundbreaking AI technology used to analyze prehistoric livestock remains.

So why does it matter how these ancient animals lived and what they ate?

According to Manuel Domínguez-Rodrigo, visiting professor of anthropology at Rice, co-director of Madrid's Institute of Evolution in Africa and professor of prehistory at the University of Alcalá in Spain, the study sheds light on how the ecology of the time affected the evolution of mammal communities including humans, who over the past two million years have become highly dependent on other mammals.

"The evolution of ecosystems in Africa is of major relevance to understand what shaped our own evolution as humans," Domínguez-Rodrigo said.

"Our prehistoric ancestors were highly dependent on resources available in different habitats of African savanna ecosystems. Using fossil mammals -- highly specialized in their adaptations to different habitats -- to reconstruct these landscapes has been the most used method to interpret their ecology. Identifying those mammals by their teeth has not always been straightforward and was subjected to a high degree of expert knowledge and bias. Now we can do that with much more confidence. This will enable us to understand past environments but also understand better modern landscapes too when documenting the dead animals that they still contain."

And thanks to this technology, whose application to paleobiology is pioneered in Domínguez-Rodrigo's lab, he says archaeologists can now analyze information far more quickly and accurately than before.

"These AI methods are a revolution for the studies of paleobiology and human evolution in particular," he said.

"They provide an objective, replicable way of identifying animals, including the degree of confidence with which identifications are made."

Domínguez-Rodrigo said the success of AI in other fields, such as image-based medicine, was a proof of concept for its widespread application to other fields.

Read more at Science Daily

AI faces look more real than actual human face

White faces generated by artificial intelligence (AI) now appear more real than human faces, according to new research led by experts at The Australian National University (ANU).

In the study, more people thought AI-generated white faces were human than the faces of real people. The same wasn't true for images of people of colour.

The reason for the discrepancy is that AI algorithms are trained disproportionately on white faces, Dr Amy Dawel, the senior author of the paper, said.

"If white AI faces are consistently perceived as more realistic, this technology could have serious implications for people of colour by ultimately reinforcing racial biases online," Dr Dawel said.

"This problem is already apparent in current AI technologies that are being used to create professional-looking headshots. When used for people of colour, the AI is altering their skin and eye colour to those of white people."

One of the issues with AI 'hyper-realism' is that people often don't realise they're being fooled, the researchers found.

"Concerningly, people who thought that the AI faces were real most often were paradoxically the most confident their judgements were correct," Elizabeth Miller, study co-author and PhD candidate at ANU, said.

"This means people who are mistaking AI imposters for real people don't know they are being tricked."

The researchers were also able to discover why AI faces are fooling people.

"It turns out that there are still physical differences between AI and human faces, but people tend to misinterpret them. For example, white AI faces tend to be more in-proportion and people mistake this as a sign of humanness," Dr Dawel said.

"However, we can't rely on these physical cues for long. AI technology is advancing so quickly that the differences between AI and human faces will probably disappear soon."

The researchers argue this trend could have serious implications for the proliferation of misinformation and identity theft, and that action needs to be taken.

"AI technology can't become sectioned off so only tech companies know what's going on behind the scenes. There needs to be greater transparency around AI so researchers and civil society can identify issues before they become a major problem," Dr Dawel said.

Raising public awareness can also play a significant role in reducing the risks posed by the technology, the researchers argue.

"Given that humans can no longer detect AI faces, society needs tools that can accurately identify AI imposters," Dr Dawel said.

Read more at Science Daily

How artificial intelligence gave a paralyzed woman her voice back

Researchers at UC San Francisco and UC Berkeley have developed a brain-computer interface (BCI) that has enabled a woman with severe paralysis from a brainstem stroke to speak through a digital avatar.

It is the first time that either speech or facial expressions have been synthesized from brain signals. The system can also decode these signals into text at nearly 80 words per minute, a vast improvement over commercially available technology.

Edward Chang, MD, chair of neurological surgery at UCSF, who has worked on the technology, known as a brain computer interface, or BCI, for more than a decade, hopes this latest research breakthrough, appearing Aug. 23, 2023, in Nature, will lead to an FDA-approved system that enables speech from brain signals in the near future.

"Our goal is to restore a full, embodied way of communicating, which is really the most natural way for us to talk with others," said Chang, who is a member of the UCSF Weill Institute for Neuroscience and the Jeanne Robertson Distinguished Professor in Psychiatry. "These advancements bring us much closer to making this a real solution for patients."

Chang's team previously demonstrated it was possible to decode brain signals into text in a man who had also experienced a brainstem stroke many years earlier. The current study demonstrates something more ambitious: decoding brain signals into the richness of speech, along with the movements that animate a person's face during conversation.

Chang implanted a paper-thin rectangle of 253 electrodes onto the surface of the woman's brain over areas his team has discovered are critical for speech. The electrodes intercepted the brain signals that, if not for the stroke, would have gone to muscles in her, tongue, jaw and larynx, as well as her face. A cable, plugged into a port fixed to her head, connected the electrodes to a bank of computers.

For weeks, the participant worked with the team to train the system's artificial intelligence algorithms to recognize her unique brain signals for speech. This involved repeating different phrases from a 1,024-word conversational vocabulary over and over again, until the computer recognized the brain activity patterns associated with the sounds.

Rather than train the AI to recognize whole words, the researchers created a system that decodes words from phonemes. These are the sub-units of speech that form spoken words in the same way that letters form written words. "Hello," for example, contains four phonemes: "HH," "AH," "L" and "OW."

Using this approach, the computer only needed to learn 39 phonemes to decipher any word in English. This both enhanced the system's accuracy and made it three times faster.

"The accuracy, speed and vocabulary are crucial," said Sean Metzger, who developed the text decoder with Alex Silva, both graduate students in the joint Bioengineering Program at UC Berkeley and UCSF. "It's what gives a user the potential, in time, to communicate almost as fast as we do, and to have much more naturalistic and normal conversations."

To create the voice, the team devised an algorithm for synthesizing speech, which they personalized to sound like her voice before the injury, using a recording of her speaking at her wedding.

The team animated the avatar with the help of software that simulates and animates muscle movements of the face, developed by Speech Graphics, a company that makes AI-driven facial animation. The researchers created customized machine-learning processes that allowed the company's software to mesh with signals being sent from the woman's brain as she was trying to speak and convert them into the movements on the avatar's face, making the jaw open and close, the lips protrude and purse and the tongue go up and down, as well as the facial movements for happiness, sadness and surprise.

"We're making up for the connections between the brain and vocal tract that have been severed by the stroke," said Kaylo Littlejohn, a graduate student working with Chang and Gopala Anumanchipalli, PhD, a professor of electrical engineering and computer sciences at UC Berkeley. "When the subject first used this system to speak and move the avatar's face in tandem, I knew that this was going to be something that would have a real impact."

An important next step for the team is to create a wireless version that would not require the user to be physically connected to the BCI.

Read more at Science Daily

Thermal imaging innovation allows AI to see through pitch darkness like broad daylight

Researchers at Purdue University are advancing the world of robotics and autonomy with their patent-pending method that improves on traditional machine vision and perception.

Zubin Jacob, the Elmore Associate Professor of Electrical and Computer Engineering in the Elmore Family School of Electrical and Computer Engineering, and research scientist Fanglin Bao have developed HADAR, or heat-assisted detection and ranging. Their research was featured on the cover of the July 26 issue of the peer-reviewed journal Nature. A video about HADAR is available on YouTube. Nature also has released a podcast episode that includes an interview with Jacob.

Jacob said it is expected that one in 10 vehicles will be automated and that there will be 20 million robot helpers that serve people by 2030.

"Each of these agents will collect information about its surrounding scene through advanced sensors to make decisions without human intervention," Jacob said. "However, simultaneous perception of the scene by numerous agents is fundamentally prohibitive."

Traditional active sensors like LiDAR, or light detection and ranging, radar and sonar emit signals and subsequently receive them to collect 3D information about a scene. These methods have drawbacks that increase as they are scaled up, including signal interference and risks to people's eye safety. In comparison, video cameras that work based on sunlight or other sources of illumination are advantageous, but low-light conditions such as nighttime, fog or rain present a serious impediment.

Traditional thermal imaging is a fully passive sensing method that collects invisible heat radiation originating from all objects in a scene. It can sense through darkness, inclement weather and solar glare. But Jacob said fundamental challenges hinder its use today.

"Objects and their environment constantly emit and scatter thermal radiation, leading to textureless images famously known as the 'ghosting effect,'" Bao said. "Thermal pictures of a person's face show only contours and some temperature contrast; there are no features, making it seem like you have seen a ghost. This loss of information, texture and features is a roadblock for machine perception using heat radiation."

HADAR combines thermal physics, infrared imaging and machine learning to pave the way to fully passive and physics-aware machine perception.

"Our work builds the information theoretic foundations of thermal perception to show that pitch darkness carries the same amount of information as broad daylight. Evolution has made human beings biased toward the daytime. Machine perception of the future will overcome this long-standing dichotomy between day and night," Jacob said.

Bao said, "HADAR vividly recovers the texture from the cluttered heat signal and accurately disentangles temperature, emissivity and texture, or TeX, of all objects in a scene. It sees texture and depth through the darkness as if it were day and also perceives physical attributes beyond RGB, or red, green and blue, visible imaging or conventional thermal sensing. It is surprising that it is possible to see through pitch darkness like broad daylight."

The team tested HADAR TeX vision using an off-road nighttime scene.

"HADAR TeX vision recovered textures and overcame the ghosting effect," Bao said. "It recovered fine textures such as water ripples, bark wrinkles and culverts in addition to details about the grassy land."

Additional improvements to HADAR are improving the size of the hardware and the data collection speed.

"The current sensor is large and heavy since HADAR algorithms require many colors of invisible infrared radiation," Bao said. "To apply it to self-driving cars or robots, we need to bring down the size and price while also making the cameras faster. The current sensor takes around one second to create one image, but for autonomous cars we need around 30 to 60 hertz frame rate, or frames per second."

HADAR TeX vision's initial applications are automated vehicles and robots that interact with humans in complex environments. The technology could be further developed for agriculture, defense, geosciences, health care and wildlife monitoring applications.

Read more at Science Daily

Humans unable to detect over a quarter of deepfake speech samples

The study, published today in PLOS ONE, is the first to assess human ability to detect artificially generated speech in a language other than English.

Deepfakes are synthetic media intended to resemble a real person's voice or appearance. They fall under the category of generative artificial intelligence (AI), a type of machine learning (ML) that trains an algorithm to learn the patterns and characteristics of a dataset, such as video or audio of a real person, so that it can reproduce original sound or imagery.

While early deepfake speech algorithms may have required thousands of samples of a person's voice to be able to generate original audio, the latest pre-trained algorithms can recreate a person's voice using just a three-second clip of them speaking. Open-source algorithms are freely available and while some expertise would be beneficial, it would be feasible for an individual to train them within a few days.

Tech firm Apple recently announced software for iPhone and iPad that allows a user to create a copy of their voice using 15 minutes of recordings.

Researchers at UCL used a text-to-speech (TTS) algorithm trained on two publicly available datasets, one in English and one in Mandarin, to generate 50 deepfake speech samples in each language. These samples were different from the ones used to train the algorithm to avoid the possibility of it reproducing the original input.

These artificially generated samples and genuine samples were played for 529 participants to see whether they could detect the real thing from fake speech. Participants were only able to identify fake speech 73% of the time, which improved only slightly after they received training to recognise aspects of deepfake speech.

Kimberly Mai (UCL Computer Science), first author of the study, said: "Our findings confirm that humans are unable to reliably detect deepfake speech, whether or not they have received training to help them spot artificial content. It's also worth noting that the samples that we used in this study were created with algorithms that are relatively old, which raises the question whether humans would be less able to detect deepfake speech created using the most sophisticated technology available now and in the future."

The next step for the researchers is to develop better automated speech detectors as part of ongoing efforts to create detection capabilities to counter the threat of artificially generated audio and imagery.

Though there are benefits from generative AI audio technology, such as greater accessibility for those whose speech may be limited or who may lose their voice due to illness, there are growing fears that such technology could be used by criminals and nation states to cause significant harm to individuals and societies.

Documented cases of deepfake speech being used by criminals include one 2019 incident where the CEO of a British energy company was convinced to transfer hundreds of thousands of pounds to a false supplier by a deepfake recording of his boss's voice.

Read more at Science Daily

Dangerous chatbots: AI chatbots to be approved as medical devices?

"Large Language Models are neural network language models with remarkable conversational skills. They generate human-like responses and engage in interactive conversations. However, they often generate highly convincing statements that are verifiably wrong or provide inappropriate responses. Today there is no way to be certain about the quality, evidence level, or consistency of clinical information or supporting evidence for any response. These chatbots are unsafe tools when it comes to medical advice and it is necessary to develop new frameworks that ensure patient safety," said Prof. Stephen Gilbert, Professor for Medical Device Regulatory Science at Else Kröner Fresenius Center for Digital Health at TU Dresden.

Challenges in the regulatory approval of large language models

Most people research their symptoms online before seeking medical advice. Search engines play a role in decision-making process. The forthcoming integration of LLM-chatbots into search engines may increase users' confidence in the answers given by a chatbot that mimics conversation. It has been demonstrated that LLMs can provide profoundly dangerous information when prompted with medical questions.

LLM's underlying approach has no model of medical "ground truth," which is inherently dangerous. Chat interfaced LLMs have already provided harmful medical responses and have already been used unethically in 'experiments' on patients without consent. Almost every medical LLM use case requires regulatory control in the EU and US. In the US their lack of explainability disqualifies them from being 'non devices'. LLMs with explainability, low bias, predictability, correctness, and verifiable outputs do not currently exist and they are not exempted from current (or future) governance approaches.

Read more at Science Daily

AI reveals hidden traits about our planet's flora to help save species

In a world-first, scientists from UNSW and Botanic Gardens of Sydney, have trained AI to unlock data from millions of plant specimens kept in herbaria around the world, to study and combat the impacts of climate change on flora.

"Herbarium collections are amazing time capsules of plant specimens," says lead author on the study, Associate Professor Will Cornwell. "Each year over 8000 specimens are added to the National Herbarium of New South Wales alone, so it's not possible to go through things manually anymore."

Using a new machine learning algorithm to process over 3000 leaf samples, the team discovered that contrary to frequently observed interspecies patterns, leaf size doesn't increase in warmer climates within a single species.

Published in the American Journal of Botany, this research not only reveals that factors other than climate have a strong effect on leaf size within a plant species, but demonstrates how AI can be used to transform static specimen collections and to quickly and effectively document climate change effects.

Herbarium collections move to the digital world

Herbaria are scientific libraries of plant specimens that have existed since at least the 16th century.

"Historically, a valuable scientific effort was to go out, collect plants, and then keep them in a herbarium. Every record has a time and a place and a collector and a putative species ID," says A/Prof. Cornwell, a researcher at the School of BEES and a member of UNSW Data Science Hub.

A couple of years ago, to help facilitate scientific collaboration, there was a movement to transfer these collections online.

"The herbarium collections were locked in small boxes in particular places, but the world is very digital now. So to get the information about all of the incredible specimens to the scientists who are now scattered across the world, there was an effort to scan the specimens to produce high resolution digital copies of them."

The largest herbarium imaging project was undertaken at the Botanic Gardens of Sydney when over 1 million plant specimens at the National Herbarium of New South Wales were transformed into high-resolution digital images.

"The digitisation project took over two years and shortly after completion, one of the researchers -- Dr Jason Bragg -- contacted me from the Botanic Gardens of Sydney. He wanted to see how we could incorporate machine learning with some of these high-resolution digital images of the Herbarium specimens."

"I was excited to work with A/Prof. Cornwell in developing models to detect leaves in the plant images, and to then use those big datasets to study relationships between leaf size and climate," says Dr Bragg.

"Computer vision" measures leaf sizes

Together with Dr Bragg at the Botanic Gardens of Sydney and UNSW Honours student Brendan Wilde, A/Prof. Cornwell created an algorithm that could be automated to detect and measure the size of leaves of scanned herbarium samples for two plant genera -- Syzygium (generally known as lillipillies, brush cherries or satinas) and Ficus (a genus of about 850 species of woody trees, shrubs and vines).

"This is a type of AI is called a convolutional neural network, also known as Computer Vision," says A/Prof. Cornwell. The process essentially teaches the AI to see and identify the components of a plant in the same way a human would.

"We had to build a training data set to teach the computer, this is a leaf, this is a stem, this is a flower," says A/Prof. Cornwell. "So we basically taught the computer to locate the leaves and then measure the size of them.

"Measuring the size of leaves is not novel, because lots of people have done this. But the speed with which these specimens can be processed and their individual characteristics can be logged is a new development."

A break in frequently observed patterns

A general rule of thumb in the botanical world is that in wetter climates, like tropical rainforests, the leaves of plants are bigger compared to drier climates, such as deserts.

"And that's a very consistent pattern that we see in leaves between species all across the globe," says A/Prof. Cornwell. "The first test we did was to see if we could reconstruct that relationship from the machine learned data, which we could. But the second question was, because we now have so much more data than we had before, do we see the same thing within species?"

The machine learning algorithm was developed, validated, and applied to analyse the relationship between leaf size and climate within and among species for Syzygium and Ficus plants.

The results from this test were surprising -- the team discovered that while this pattern can be seen between different plant species, the same correlation isn't seen within a single species across the globe, likely because a different process, known as gene flow, is operating within species. That process weakens plant adaptation on a local scale and could be preventing the leaf size-climate relationship from developing within species.

Using AI to predict future climate change responses

The machine learning approach used here to detect and measure leaves, though not pixel perfect, provided levels of accuracy suitable for examining links between leaf traits and climate.

"But because the world is changing quite fast, and there is so much data, these kinds of machine learning methods can be used to effectively document climate change effects," says A/Prof. Cornwell.

Read more at Science Daily

AI algorithm unblurs the cosmos

The cosmos would look a lot better if Earth's atmosphere wasn't photo bombing it all the time.

Even images obtained by the world's best ground-based telescopes are blurry due to the atmosphere's shifting pockets of air. While seemingly harmless, this blur obscures the shapes of objects in astronomical images, sometimes leading to error-filled physical measurements that are essential for understanding the nature of our universe.

Now researchers at Northwestern University and Tsinghua University in Beijing have unveiled a new strategy to fix this issue. The team adapted a well-known computer-vision algorithm used for sharpening photos and, for the first time, applied it to astronomical images from ground-based telescopes. The researchers also trained the artificial intelligence (AI) algorithm on data simulated to match the Vera C. Rubin Observatory's imaging parameters, so, when the observatory opens next year, the tool will be instantly compatible.

While astrophysicists already use technologies to remove blur, the adapted AI-driven algorithm works faster and produces more realistic images than current technologies. The resulting images are blur-free and truer to life. They also are beautiful -- although that's not the technology's purpose.

"Photography's goal is often to get a pretty, nice-looking image," said Northwestern's Emma Alexander, the study's senior author. "But astronomical images are used for science. By cleaning up images in the right way, we can get more accurate data. The algorithm removes the atmosphere computationally, enabling physicists to obtain better scientific measurements. At the end of the day, the images do look better as well."

The research will be published March 30 in the Monthly Notices of the Royal Astronomical Society.

Alexander is an assistant professor of computer science at Northwestern's McCormick School of Engineering, where she runs the Bio Inspired Vision Lab. She co-led the new study with Tianao Li, an undergraduate in electrical engineering at Tsinghua University and a research intern in Alexander's lab.

When light emanates from distant stars, planets and galaxies, it travels through Earth's atmosphere before it hits our eyes. Not only does our atmosphere block out certain wavelengths of light, it also distorts the light that reaches Earth. Even clear night skies still contain moving air that affects light passing through it. That's why stars twinkle and why the best ground-based telescopes are located at high altitudes where the atmosphere is thinnest.

"It's a bit like looking up from the bottom of a swimming pool," Alexander said. "The water pushes light around and distorts it. The atmosphere is, of course, much less dense, but it's a similar concept."

The blur becomes an issue when astrophysicists analyze images to extract cosmological data. By studying the apparent shapes of galaxies, scientists can detect the gravitational effects of large-scale cosmological structures, which bend light on its way to our planet. This can cause an elliptical galaxy to appear rounder or more stretched than it really is. But atmospheric blur smears the image in a way that warps the galaxy shape. Removing the blur enables scientists to collect accurate shape data.

"Slight differences in shape can tell us about gravity in the universe," Alexander said. "These differences are already difficult to detect. If you look at an image from a ground-based telescope, a shape might be warped. It's hard to know if that's because of a gravitational effect or the atmosphere."

To tackle this challenge, Alexander and Li combined an optimization algorithm with a deep-learning network trained on astronomical images. Among the training images, the team included simulated data that matches the Rubin Observatory's expected imaging parameters. The resulting tool produced images with 38.6% less error compared to classic methods for removing blur and 7.4% less error compared to modern methods.

When the Rubin Observatory officially opens next year, its telescopes will begin a decade-long deep survey across an enormous portion of the night sky. Because the researchers trained the new tool on data specifically designed to simulate Rubin's upcoming images, it will be able to help analyze the survey's highly anticipated data.

For astronomers interested in using the tool, the open-source, user-friendly code and accompanying tutorials are available online.

"Now we pass off this tool, putting it into the hands of astronomy experts," Alexander said. "We think this could be a valuable resource for sky surveys to obtain the most realistic data possible."

Read more at Science Daily

Artificial intelligence discovers secret equation for 'weighing' galaxy clusters

Astrophysicists at the Institute for Advanced Study, the Flatiron Institute and their colleagues have leveraged artificial intelligence to uncover a better way to estimate the mass of colossal clusters of galaxies. The AI discovered that by just adding a simple term to an existing equation, scientists can produce far better mass estimates than they previously had.

The improved estimates will enable scientists to calculate the fundamental properties of the universe more accurately, the astrophysicists reported March 17, 2023, in the Proceedings of the National Academy of Sciences.

"It's such a simple thing; that's the beauty of this," says study co-author Francisco Villaescusa-Navarro, a research scientist at the Flatiron Institute's Center for Computational Astrophysics (CCA) in New York City. "Even though it's so simple, nobody before found this term. People have been working on this for decades, and still they were not able to find this."

The work was led by Digvijay Wadekar of the Institute for Advanced Study in Princeton, New Jersey, along with researchers from the CCA, Princeton University, Cornell University and the Center for Astrophysics | Harvard & Smithsonian.

Understanding the universe requires knowing where and how much stuff there is. Galaxy clusters are the most massive objects in the universe: A single cluster can contain anything from hundreds to thousands of galaxies, along with plasma, hot gas and dark matter. The cluster's gravity holds these components together. Understanding such galaxy clusters is crucial to pinning down the origin and continuing evolution of the universe.

Perhaps the most crucial quantity determining the properties of a galaxy cluster is its total mass. But measuring this quantity is difficult -- galaxies cannot be 'weighed' by placing them on a scale. The problem is further complicated because the dark matter that makes up much of a cluster's mass is invisible. Instead, scientists deduce the mass of a cluster from other observable quantities.

In the early 1970s, Rashid Sunyaev, current distinguished visiting professor at the Institute for Advanced Study's School of Natural Sciences, and his collaborator Yakov B. Zel'dovich developed a new way to estimate galaxy cluster masses. Their method relies on the fact that as gravity squashes matter together, the matter's electrons push back. That electron pressure alters how the electrons interact with particles of light called photons. As photons left over from the Big Bang's afterglow hit the squeezed material, the interaction creates new photons. The properties of those photons depend on how strongly gravity is compressing the material, which in turn depends on the galaxy cluster's heft. By measuring the photons, astrophysicists can estimate the cluster's mass.

However, this 'integrated electron pressure' is not a perfect proxy for mass, because the changes in the photon properties vary depending on the galaxy cluster. Wadekar and his colleagues thought an artificial intelligence tool called 'symbolic regression' might find a better approach. The tool essentially tries out different combinations of mathematical operators -- such as addition and subtraction -- with various variables, to see what equation best matches the data.

Wadekar and his collaborators 'fed' their AI program a state-of-the-art universe simulation containing many galaxy clusters. Next, their program, written by CCA research fellow Miles Cranmer, searched for and identified additional variables that might make the mass estimates more accurate.

AI is useful for identifying new parameter combinations that human analysts might overlook. For example, while it is easy for human analysts to identify two significant parameters in a dataset, AI can better parse through high volumes, often revealing unexpected influencing factors.

"Right now, a lot of the machine-learning community focuses on deep neural networks," Wadekar explained. "These are very powerful, but the drawback is that they are almost like a black box. We cannot understand what goes on in them. In physics, if something is giving good results, we want to know why it is doing so. Symbolic regression is beneficial because it searches a given dataset and generates simple mathematical expressions in the form of simple equations that you can understand. It provides an easily interpretable model."

The researchers' symbolic regression program handed them a new equation, which was able to better predict the mass of the galaxy cluster by adding a single new term to the existing equation. Wadekar and his collaborators then worked backward from this AI-generated equation and found a physical explanation. They realized that gas concentration correlates with the regions of galaxy clusters where mass inferences are less reliable, such as the cores of galaxies where supermassive black holes lurk. Their new equation improved mass inferences by downplaying the importance of those complex cores in the calculations. In a sense, the galaxy cluster is like a spherical doughnut. The new equation extracts the jelly at the center of the doughnut that can introduce larger errors, and instead concentrates on the doughy outskirts for more reliable mass inferences.

The researchers tested the AI-discovered equation on thousands of simulated universes from the CCA's CAMELS suite. They found that the equation reduced the variability in galaxy cluster mass estimates by around 20 to 30 percent for large clusters compared with the currently used equation.

The new equation can provide observational astronomers engaged in upcoming galaxy cluster surveys with better insights into the mass of the objects they observe. "There are quite a few surveys targeting galaxy clusters [that] are planned in the near future," Wadekar noted. "Examples include the Simons Observatory, the Stage 4 CMB experiment and an X-ray survey called eROSITA. The new equations can help us in maximizing the scientific return from these surveys."

Read more at Science Daily

AI finds the first stars were not alone

By using machine learning and state-of-the-art supernova nucleosynthesis, a team of researchers have found the majority of observed second-generation stars in the universe were enriched by multiple supernovae, reports a new study in The Astrophysical Journal.

Nuclear astrophysics research has shown elements including and heavier than carbon in the universe are produced in stars. But the first stars, stars born soon after the Big Bang, did not contain such heavy elements, which astronomers call 'metals'. The next generation of stars contained only a small amount of heavy elements produced by the first stars. To understand the universe in its infancy, it requires researchers to study these metal-poor stars.

Luckily, these second-generation metal-poor stars are observed in our Milky Way Galaxy, and have been studied by a team of Affiliate Members of the Kavli Institute for the Physics and Mathematics of the Universe (Kavli IPMU) to close in on the physical properties of the first stars in the universe.

The team, led by Kavli IPMU Visiting Associate Scientist and The University of Tokyo Institute for Physics of Intelligence Assistant Professor Tilman Hartwig, including Visiting Associate Scientist and National Astronomical Observatory of Japan Assistant Professor Miho Ishigaki, Visiting Senior Scientist and University of Hertfordshire Professor Chiaki Kobayashi, Visiting Senior Scientist and National Astronomical Observatory of Japan Professor Nozomu Tominaga, and Visiting Senior Scientist and The University of Tokyo Professor Emeritus Ken'ichi Nomoto, used artificial intelligence to analyze elemental abundances in more than 450 extremely metal-poor stars observed to date. Based on the newly developed supervised machine learning algorithm trained on theoretical supernova nucleosynthesis models, they found that 68 per cent of the observed extremely metal-poor stars have a chemical fingerprint consistent with enrichment by multiple previous supernovae.

The team's results give the first quantitative constraint based on observations on the multiplicity of the first stars.

"Multiplicity of the first stars were only predicted from numerical simulations so far, and there was no way to observationally examine the theoretical prediction until now," said lead author Hartwig. "Our result suggests that most first stars formed in small clusters so that multiple of their supernovae can contribute to the metal enrichment of the early interstellar medium," he said.

"Our new algorithm provides an excellent tool to interpret the big data we will have in the next decade from on-going and future astronomical surveys across the world" said Kobayashi, also a Leverhulme Research Fellow.

"At the moment, the available data of old stars are the tip of the iceberg within the solar neighborhood. The Prime Focus Spectrograph, a cutting-edge multi-object spectrograph on the Subaru Telescope developed by the international collaboration led by Kavli IPMU, is the best instrument to discover ancient stars in the outer regions of the Milky Way far beyond the solar neighborhood.," said Ishigaki.

The new algorithm invented in this study opens the door to make the most of diverse chemical fingerprints in metal-poor stars discovered by the Prime Focus Spectrograph.

Read more at Science Daily

AI draws most accurate map of star birthplaces in the Galaxy

Osaka Metropolitan University scientists identified about 140,000 molecular clouds in the Milky Way Galaxy from large-scale data of carbon monoxide molecules, observed in detail by the Nobeyama 45-m radio telescope. Using artificial intelligence, the researchers estimated the distance of each of these molecular clouds to determine their size and mass, successfully mapping the distribution of the molecular clouds in the Galaxy in the most detailed manner to date.

Stars are formed by molecular gas and dust coalescing in space. These molecular gases are so dilute and cold that they are invisible to the human eye, but they do emit faint radio waves that can be observed by radio telescopes.

Observing from Earth, a lot of matter lies ahead and behind these molecular clouds and these overlapping features make it difficult to determine their distance and physical properties such as size and mass.

So, even though our Galaxy, the Milky Way, is the only galaxy close enough to make detailed observations of molecular clouds in the whole universe, it has been very difficult to investigate the physical properties of molecular clouds in a cohesive manner from large-scale observations.

A research team led by Dr. Shinji Fujita from the Osaka Metropolitan University Graduate School of Science, identified about 140,000 molecular clouds in the Milky Way Galaxy, which are areas of star formation, from large-scale data of carbon monoxide molecules, observed in detail by the Nobeyama 45-m radio telescope. Using artificial intelligence, the research team estimated the distance of each of these molecular clouds, determined their size and mass and successfully mapped their distribution, covering the first quadrant of the Galactic plane, in the most detailed manner to date.

From Science Daily

New AI tool makes speedy gene-editing possible

An artificial intelligence program may enable the first simple production of customizable proteins called zinc fingers to treat diseases by turning genes on and off.

The researchers at NYU Grossman School of Medicine and the University of Toronto who designed the tool say it promises to accelerate the development of gene therapies on a large scale.

Illnesses including cystic fibrosis, Tay-Sachs disease, and sickle cell anemia are caused by errors in the order of DNA letters that encode the operating instructions for every human cell. Scientists can in some cases correct these mistakes with gene editing methods that rearrange these letters.

Other conditions are caused, not by a mistake in the code itself, but by problems in how the cellular machinery reads DNA (epigenetics). A gene, which provides the recipe for a particular protein, often partners with molecules called transcription factors that tell the cell how much of that protein to make. When this process goes awry, over- or underactive genes contribute to diabetes, cancer, and neurological disorders. As a result, researchers have been exploring ways to restore normal epigenetic activity.

One such technique is zinc-finger editing, which can both change and control genes. Among the most abundant protein structures in the human body, zinc fingers can guide DNA repair by grabbing onto scissor-like enzymes and directing them to cut faulty segments out of the code.

Similarly, zinc fingers can also hook onto transcription factors and pull them toward a gene segment in need of regulation. By customizing these instructions, genetic engineers can tailor any gene's activity. A drawback, however, is that artificial zinc fingers are challenging to design for a specific task. Since these proteins attach to DNA in complex groups, researchers would need to be able to tell -- out of countless possible combinations -- how every zinc finger interacts with its neighbor for each desired genetic change.

The study authors' new technology, called ZFDesign, overcomes this obstacle by using artificial intelligence (AI) to model and design these interactions. The model is based on data generated by the screen of nearly 50 billion possible zinc finger-DNA interactions in the researchers' labs. A report on the tool is publishing online Jan. 26 in the journal Nature Biotechnology.

"Our program can identify the right grouping of zinc fingers for any modification, making this type of gene editing faster than ever before," says study lead author David Ichikawa, PhD, a former graduate student at NYU Langone Health.

Ichikawa notes that zinc-finger editing offers a potentially safer alternative to CRISPR, a key gene-editing technology with applications that range from finding new ways to kill cancer cells to designing more nourishing crops. Unlike the entirely human-derived zinc fingers, CRISPR, which stands for clustered regularly interspaced short palindromic repeat, relies on bacterial proteins to interact with genetic code. These "foreign" proteins could trigger patients' immune defense systems, which may attack them like any other infection and lead to dangerous inflammation.

The study authors add that besides posing a lower immune risk, the small size of zinc-finger tools may also provide more flexible gene therapy techniques compared with CRISPR by enabling more ways to deliver the tools to the right cells in patients.

"By speeding up zinc-finger design coupled with their smaller size, our system paves the way for using these proteins to control multiple genes at the same time," says study senior author Marcus Noyes, PhD. "In the future, this approach may help correct diseases that have multiple genetic causes, such as heart disease, obesity, and many cases of autism."

To test the computer's AI design code, Noyes and his team used a customized zinc finger to disrupt the coding sequence of a gene in human cells. In addition, they built several zinc fingers that successfully reprogrammed transcription factors to bind near a target gene sequence and turn up or down its expression, demonstrating that their technology can be used for epigenetic changes.

Noyes, an assistant professor in the Department of Biochemistry and Molecular Pharmacology at NYU Langone, cautions that, while promising, zinc fingers can be difficult to control. Since they are not always specific to a single gene, some combinations can affect DNA sequences beyond a particular target, leading to unintended changes in genetic code.

As a result, Noyes says the team next plans to refine their AI program so it can build more precise zinc-finger groupings that only prompt the desired edit. Noyes is also a member of NYU Langone's Institute for System Genetics.

Read more at Science Daily

Even smartest AI models don't match human visual processing

Deep convolutional neural networks (DCNNs) don't see objects the way humans do -- using configural shape perception -- and that could be dangerous in real-world AI applications, says Professor James Elder, co-author of a York University study published today.

Published in the Cell Press journal iScience, Deep learning models fail to capture the configural nature of human shape perception is a collaborative study by Elder, who holds the York Research Chair in Human and Computer Vision and is Co-Director of York's Centre for AI & Society, and Assistant Psychology Professor Nicholas Baker at Loyola College in Chicago, a former VISTA postdoctoral fellow at York.

The study employed novel visual stimuli called "Frankensteins" to explore how the human brain and DCNNs process holistic, configural object properties.

"Frankensteins are simply objects that have been taken apart and put back together the wrong way around," says Elder. "As a result, they have all the right local features, but in the wrong places."

The investigators found that while the human visual system is confused by Frankensteins, DCNNs are not -- revealing an insensitivity to configural object properties.

"Our results explain why deep AI models fail under certain conditions and point to the need to consider tasks beyond object recognition in order to understand visual processing in the brain," Elder says. "These deep models tend to take 'shortcuts' when solving complex recognition tasks. While these shortcuts may work in many cases, they can be dangerous in some of the real-world AI applications we are currently working on with our industry and government partners," Elder points out.

One such application is traffic video safety systems: "The objects in a busy traffic scene -- the vehicles, bicycles and pedestrians -- obstruct each other and arrive at the eye of a driver as a jumble of disconnected fragments," explains Elder. "The brain needs to correctly group those fragments to identify the correct categories and locations of the objects. An AI system for traffic safety monitoring that is only able to perceive the fragments individually will fail at this task, potentially misunderstanding risks to vulnerable road users."

Read more at Science Daily