Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI – Page 63

The Hot Mess of AI: How Does Misalignment Scale with Model Intelligence and Task Complexity?

When AI systems fail, will they fail by systematically pursuing the wrong goals, or by being a hot mess? We decompose the errors of frontier reasoning models into bias (systematic) and variance (incoherent) components and find that, as tasks get harder and reasoning gets longer, model failures become increasingly dominated by incoherence rather than systematic misalignment. This suggests that future AI failures may look more like industrial accidents than coherent pursuit of a goal we did not train them to pursue.

Edge of Many-Body Quantum Chaos in Quantum Reservoir Computing

Reservoir computing (RC) is a machine learning paradigm that harnesses dynamical systems as computational resources. In its quantum extension—quantum reservoir computing (QRC)—these principles are applied to quantum systems, whose rich dynamics broadens the landscape of information processing. In classical RC, optimal performance is typically achieved at the “edge of chaos,’’ the boundary between order and chaos. Here, we identify its quantum many-body counterpart using the QRC implemented on the celebrated Sachdev-Ye-Kitaev model. Our analysis reveals substantial performance enhancements near two distinct characteristic “edges’‘: a temporal boundary defined by the Thouless time, beyond which system dynamics is described by random matrix theory, and a parametric boundary governing the transition from integrable to chaotic regimes.

Reproduction in space, an environment hostile to human biology

As commercial spaceflight draws ever closer and time spent in space continues to extend, the question of reproductive health beyond the bounds of planet Earth is no longer theoretical but now “urgently practical,” according to a new study published in the journal Reproductive Biomedicine Online.

“More than 50 years ago,” explains clinical embryologist Giles Palmer from the International IVF Initiative Inc, “two scientific breakthroughs reshaped what was thought biologically and physically possible—the first moon landing and the first proof of human fertilization in vitro.

Now, more than half a century later, we argue in this report that these once-separate revolutions are colliding in a practical and underexplored reality: space is becoming a workplace and a destination, while assisted reproductive technologies have become highly advanced, increasingly automated and widely accessible.

A programmable, Lego-like material for robots emulates life’s flexibility

Mechanical engineers at Duke University have demonstrated a proof-of-concept method for programming mechanical properties into solid Lego-like building blocks. By controlling the solidity of hundreds of individual cells in specific patterns, the approach could allow futuristic robotics to alter their mechanical properties and functionalities on the fly.

In their initial tests, the researchers showed how a tail-like 3D beam with various configurations can move a robotic fish through water along different paths with the same motor activity. The team envisions miniaturized versions of the technology that could, for example, maneuver through blood vessels to survey their health or even reconfigure to form an adaptive stent.

The research appears in the journal Science Advances.

AI systems could identify math anxiety from student inputs and change feedback

Math anxiety is a significant challenge for students worldwide. While personalized support is widely recognized as the most effective way to address it, many teachers struggle to deliver this level of support at scale within busy classrooms. New research from Adelaide University shows how artificial intelligence (AI) could help address challenges such as math anxiety by using a student’s inputs and identifying signs of anxiety or disengagement during learning.

Published in npj Science of Learning, the study suggests that when AI systems are designed to use the right data and goals, they can adapt their responses to help counteract negative emotional experiences associated with math, before these feelings escalate.

Lead researcher Dr. Florence Gabriel says AI has the potential to transform how math anxiety is supported, by offering timely, tailored interventions that step through learning and build student well-being.

Tiny droplets navigate mazes using ‘chemical echolocation,’ without sensors or computers

A recent study by a team of researchers led by TU Darmstadt has found that tiny amounts of liquid can navigate their way through unknown environments like living cells—without sensors, computers or external control. The tiny droplets can navigate autonomously, are able to detect obstacles from a distance and move reliably through complex mazes—without cameras or electronics. The reason for this is a mechanism that the research team refers to as “chemical echolocation.”

Here’s how it works: Instead of emitting sound waves like bats in dark caves, the droplets release small amounts of chemicals into their environment as they move. These chemicals spread throughout the environment and are reflected by nearby walls and dead ends. The returning “echo” subtly pushes the droplet away from blocked paths and toward open paths, thus guiding its movement.

When Familiar Faces Feel Better: A Framework for Social Neurocognitive Aging in a Rat Model

New in eNeuro from Dutta Gupta et al: Some older male rats prefer familiarity over new social situations, which can be reversed via transcranial magnetic stimulation without affecting hippocampus-mediated spatial memory.

▶️

Social cognition, central to emotional and cognitive well-being, is particularly vulnerable to aging, where impairments can lead to isolation and functional decline. Despite compelling evidence that altered social behavior is associated with cognitive decline and dementia risk, experimental strategies for testing causative links remain scarce. To address this gap, we aimed to establish a rat model for research on social neurocognitive aging. We conducted a large-scale behavioral study in 169 male young (6 months) and aged (24−25 months) Long-Evans rats. In order to explore potential relationships among aging outcomes, we first documented individual differences in a widely validated water maze test of hippocampal learning and memory. Sociability and social novelty were then evaluated in the same subjects using the three-chamber social interaction test. Aging induced a selective shift in social novelty preference, marked by a striking familiarity bias in a substantial subpopulation of old rats, while sociability remained entirely normal. Changes in social novelty preference were completely independent of individual differences in spatial memory, and unrelated to anxiety or sensorimotor function. Notably, neuromodulation via TMS enhanced social novelty preference selectively in aged rats that exhibited a social introversion phenotype before treatment, consistent with the possibility that this aging condition reflects a distinct and modifiable neural network state. Together, the results establish a valuable preclinical framework for developing a comprehensive neurobiology of social cognition in aging.

Significance statement Social behavior is a critical yet underexplored component of cognitive aging. While both human and animal studies report age-related narrowing of social networks, the behavioral and neurobiological underpinnings remain unclear. Using a well-powered rat model, here we demonstrate preserved sociability in aging alongside marked individual differences in social novelty preference. A subset of aged rats preferred familiar over novel conspecifics, resembling patterns observed in older humans and non-human primates. Social phenotypes were independent of hippocampal-dependent memory, suggesting a dissociation between these aging outcomes. This dissociation was further validated using transcranial magnetic stimulation, supporting the notion of distinct underlying neurobiological mechanisms. Collectively, the findings lay a powerful foundation for advancing the translational neurobiology of social behavior in cognitive aging and reserve.

AI that talks to itself learns faster and smarter

AI may learn better when it’s allowed to talk to itself. Researchers showed that internal “mumbling,” combined with short-term memory, helps AI adapt to new tasks, switch goals, and handle complex challenges more easily. This approach boosts learning efficiency while using far less training data. It could pave the way for more flexible, human-like AI systems.

/* */