Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI

Battleship-trained AI learns to ask sharper questions, boosting win rate from 8% to 82%

In 2026, the hype for artificial intelligence agents is louder than ever before. These semi-autonomous programs can “think” and execute well-defined tasks in areas like customer service and software development, typically using language models (LMs). But fields like medical diagnosis and scientific discovery require them to inquire about a vast range of solutions in uncertain environments which LMs struggle with.

Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Harvard University’s School of Engineering and Applied Sciences (SEAS) peered deeper into LMs to understand their main issues in high-stakes settings. Their test: Battleship, a classic guessing game that’s helped cognitive scientists study how humans seek information.

CSAIL and SEAS scholars added a twist by reframing the game around asking and answering natural language questions. In their “Collaborative Battleship” game, one participant is a “captain” who inquires about where hidden ships are, while their teammate plays the “spotter” by responding to those questions in real time.

AI-designed universal coronavirus vaccine passes first human trial

Because the method does not require a needle, it could offer an alternative for people who are uncomfortable with injections. Researchers also believe it may make large scale vaccination campaigns easier and faster, particularly in settings where traditional injections are more difficult to administer.

Before human testing began, animal studies showed the vaccine could generate strong immune responses against multiple coronaviruses.

New AI tools could help eye doctors diagnose retinal disease faster

Non-invasive eye scans allow doctors a zoomed-in, three-dimensional look beneath the eye’s surface without causing discomfort or pain to the patient. Used routinely in clinics worldwide, the scans produce detailed views of individual layers of the eye’s interior to help diagnose conditions that threaten vision. But with that level of precision comes a flood of data—hundreds of images per scan that physicians have to review manually, a time-consuming process that is vulnerable to human error.

Now, researchers at Washington University School of Medicine in St. Louis, in collaboration with colleagues at the University of Washington in Seattle and Genentech, Inc., have developed an experimental artificial intelligence (AI) system that can speed the scan review process and help doctors spot subtle signs of eye disease sooner. The technology, called OCTCube-M, includes a family of three AI models that are designed to read and interpret 3D images of the eye’s retina as well as other types of eye scans.

In a new study, the researchers found that, compared with older models, the new AI system more accurately identified eight different retinal diseases, including age-related macular degeneration, a common disease that damages the retina and is the leading cause of blindness in people over 50. It also was more accurate in its predictions of how fast a severe form of this condition, called geographic atrophy, would progress.

The Brain Health Accelerator Seeks to Revolutionize Neuroscience Research

For decades, researchers across institutions have peered into microscopes and dived into data to try to understand how diseases like Alzheimer’s disease, Parkinson’s disease, and amyotrophic lateral sclerosis (ALS) affect the brain. While scientists have made many important insights into these conditions, breakthrough therapies to cure or even treat them remain out of reach.

To expedite understanding of and treatments for neurodegenerative diseases, the Allen Institute launched the Brain Health accelerator. The project, announced today, is a global initiative that will leverage cutting-edge technology with the goal of improving modeling, therapeutic development, and the understanding of disease mechanisms. With funding support from the Allen Institute, the Bezos family, Amazon Web Services, the National Institutes of Health, EverythingALS, and other partners, the project financial contribution is $400 million.

One of the challenges in studying diseases in the human brain and identifying treatment strategies has been the scale and complexity of the organ. The brain consists of many distinct parts, and studying disease mechanisms requires samples from large numbers of individuals. Additionally, while technological advancements in transcriptomics, proteomics, neuroimaging, and AI have helped researchers study the brain in finer detail, researchers have not always integrated many of these approaches into the same project.

Semiconductors enter ‘multi-tasking’ era: New device cuts required components by 75% and quadruples processing speed

Less than two decades after smartphones fit into the palm of our hands, artificial intelligence is now running on devices worn on our wrists. The challenge is that while devices continue to shrink, the amount of data they must process and the number of functions they must perform are growing exponentially. A research team at POSTECH (Pohang University of Science and Technology) has found a promising way to address this contradiction.

A team led by Professor Byoung Hun Lee of the Department of Electrical Engineering and the Department of Semiconductor Engineering at POSTECH, together with Dr. Jae Hyeon Jun of the Department of Electrical Engineering, has developed a transistor technology that enables a single semiconductor device to perform multiple circuit functions simultaneously. The new approach significantly simplifies circuit design and increases data processing speed fourfold compared with conventional methods. The findings were published in Advanced Functional Materials.

One of the key challenges in the semiconductor industry is integrating more functions into smaller chips. As the number of functions increases, so do the number of circuits and transistors required. However, when adding new functions to previously fabricated semiconductor chips, back-end-of-line processing must be conducted at temperatures below 400 C to protect the existing chip structure.

AI fails classic attention test, with longer word lists triggering dramatic accuracy collapse

Giving AI a classic psychological test reveals an inherent weakness in LLM decision-making abilities. Suketu Patel and colleagues explored how transformer-based machine attention differs from human attention by testing AI models on the “Stroop task,” in which words for colors are printed in colored ink, and participants are asked to name the ink color of each word while ignoring its meaning.

The findings are published in the journal PNAS Nexus.

The task is clinically used to assess executive control, especially a person’s ability to inhibit an automatic response. Although humans generally take longer to answer correctly when words and colors are mismatched than when they match, they can still perform stably and with high accuracy even on long word lists.

/* */