This week in toast, AI smells, computers now see like humans and scientists discover why men's health problems might be linked to vanishing Y chromosomes. Also, for those bored of horses, Kawasaki has the answer. 

What do you get when you cross human vision with AI? Less chance of your self-driving car mistaking a pedestrian for a garbage bag... A team of researchers has developed a new method called Lp-Convolution that brings AI vision closer to human vision by allowing computers to dynamically reshape their visual filters, stretching horizontally or vertically based on what they're looking at, much like how our brains selectively focus on relevant details.

When tested against standard image datasets, the new approach significantly improved accuracy while requiring less computational power. Essentially, AI now wears glasses.

🧐 What's in it for me? This could transform how AI interacts with the visual world, making self-driving cars even better at not killing people, medical imaging systems more accurate at spotting subtle abnormalities, and cameras capable of understanding complex scenes rather than just recognising objects.

💵 Out of the Lab

Continuing to blur the lines between “dystopian” and “I want one”, Kawasaki is creating a robotic horse that promises to gallop through valleys, cross rivers, climb mountains, and jump over crevasses while you sit on top looking ridiculous [ly cool]. 

Creating a robot that can balance itself while carrying a human passenger and navigating unpredictable ground represents a monumental engineering challenge. The robot will need powerful actuators to support its own weight plus a rider, sophisticated sensors to detect terrain changes and advanced algorithms to maintain balance while galloping at speed. Let us pray they succeed.

🧐 What's in it for me? If successfully developed, this could make transportation in remote areas much more accessible, or give some overworked donkeys a well-earned break. In the short term, you get a promo video.

💵 Out of the Lab:

AI knows what what smells you like, and also what you smell like… Japanese researchers have developed an AI system called that creates custom scents based on simple descriptors like "citrusy" or "woody." The system analyses the chemical profiles of essential oils and calculates the precise mixture needed to create your desired aroma.

Unlike previous fragrance design methods that relied on a perfumer's intuition developed over decades of experience this approach is fully automated. The researchers tested their digital nose with human participants who could reliably identify the scents as matching their intended descriptions. At least now when our AI overlords take over the earth, they might make it smell a bit better.

🧐 What's in it for me? Fragrance creation has traditionally been a painstaking process of trial and error, often taking months or years. Now anyone could potentially become a perfumer overnight. On second thought, maybe not such a great idea.

💵 Out of the Lab:

Wolf whistling may become a thing of the past as flying drones take on construction. Researchers are developing aerial robots that can repair bridges without scaffolding, build emergency shelters in disaster areas or construct habitats on other planets. 

Current prototypes can already place individual building elements, tension cable structures and print materials layer by layer. Whilst not the most exciting application of flying robots, it’s definitely more productive than dropping grenades on people.

🧐 What's in it for me? Construction drones could dramatically reduce building costs in difficult-to-access locations, from mountainous regions to your own roof. They could also rapidly deploy emergency infrastructure after natural disasters, potentially saving lives.

💵 Out of the Lab:

Your smartwatch may soon know you're having a heart attack before you do. Pretty handy given 1 person dies every 40 seconds of a heart attack, just in the US. Engineers at the University of Mississippi have developed technology that detects heart attacks twice as fast as conventional methods with 92.4% accuracy, using AI to analyse ECGs in real-time through wearable devices.

The technology is lightweight and energy-efficient enough to be embedded in wearable devices like smartwatches or fitness trackers. Currently, heart attack detection typically happens in medical facilities after symptoms appear, requiring electrocardiograms or blood tests that consume precious minutes when every second counts. Imagine having a tiny cardiologist strapped to your wrist, constantly checking your heart without charging consultant fees.

🧐 What's in it for me? This could dramatically reduce the time between a heart attack starting and medical intervention, potentially saving millions of lives worldwide. For people with heart disease or risk factors, a watch with this capability could provide peace of mind and crucial early warning.

💵 Out of the Lab:

In case you were looking for another reason to schedule that checkup you've been avoiding, men's Y chromosomes are ghosting their cells, and it's worse than we thought. Scientists have discovered that as men age, the Y chromosome (the one that makes men male) gradually disappears from some of their cells. This isn't just a normal part of aging, it's now linked to higher rates of heart disease, Alzheimer's, and cancer. 

Throughout evolution, the Y chromosome has been shrinking. It's lost about 1,500 genes and kept only 106, just enough to develop male characteristics and make sperm. Since these remaining genes aren't essential for basic cell survival (men's cells can generally function without the Y chromosome), more and more cells lose it as men age. Perhaps this is the first biological explanation of the mid-life crisis.

🧐 What's in it for me? Understanding this could lead to better diagnosis, prevention, and treatment of conditions that disproportionately affect men. It might also explain why men are more susceptible to certain diseases and could serve as an early warning biomarker.

💵 Out of the Lab:

Just when you thought it really couldn’t, quantum physics just got even weirder. Scientists have identified transition points where quantum matter shifts directly between ordered states, rather than progressing from disordered to ordered (e.g. water to ice) or vice versa. Imagine that instead of water turning straight into ice, it instead takes a quick detour along the way to become olive oil…🧐 

This bypasses all the usual rules of physics we thought we understood. The breakthrough came by studying how particles at these points communicate with each other through entanglement and using computer simulations, they discovered a unique rulebook that kicks in at certain thresholds.

🧐 What's in it for me? While this sounds like science fiction, these discoveries could eventually revolutionise computing, energy, and materials. Quantum computers that could solve problems beyond our current capabilities, room-temperature superconductors that transmit electricity without any loss, or entirely new materials with properties we can't yet imagine, all might emerge from understanding these quantum crossroads.

💵 Out of the Lab:

Breakthroughs just keep on coming in the CRISPR gene editing world. Scientists have now developed a machine learning algorithm that can predict properties of 64 million genome editing enzymes, essentially transforming genetic scissors into precision scalpels that could revolutionise treatments for thousands of genetic disorders.

CRISPR-Cas9 enzymes work by locating and binding to specific DNA sequences. This new algorithm can predict the most effective enzymes for targeting these sequences with greater precision, reducing unwanted edits elsewhere in the genome. In proof-of-concept experiments, these "bespoke enzymes" demonstrated greater specificity when tested in human cells and a mouse model of eye disease.

🧐 What's in it for me? More precise gene editing tools could accelerate treatments for thousands of genetic disorders, from cystic fibrosis to sickle cell disease. The technology could also advance gene therapy approaches for more common conditions like heart disease, diabetes, and certain cancers.

💵 Out of the Lab:

AI’s feeling a little awkward 😬

It turns out AI still can't read the room. Johns Hopkins researchers found that even the most sophisticated AI models consistently fail to understand basic social interactions that humans interpret effortlessly, like distinguishing between people having a pleasant conversation versus those preparing for a duel.

The study tested more than 350 AI models against human participants, asking both to interpret three-second video clips of people interacting. While humans easily recognised social cues and intentions, even the most sophisticated AI models failed miserably at understanding whether people were communicating or just standing near each other.

"It's not enough to just see an image and recognise objects and faces," explains researcher Kathy Garcia.

"We need AI to understand the story that is unfolding in a scene." The researchers believe the problem stems from AI neural networks being modelled after brain regions that process static images rather than social dynamics.

This limitation could significantly impact technologies like self-driving cars that need to predict whether pedestrians are about to cross the street or are just chatting on the corner. So while AI might soon craft your perfect perfume, it will still be gratifyingly terrible at something.

Until next time, stay curious.

Like what you're reading? Share toast with a friend.