A tiny robot developed by Japan’s space agency operated autonomously on the moon for more than 100 minutes and sent a series of images back to Earth.

Exploring the moon’s surface lays crucial groundwork for future crewed settlements, and swarms of tiny robots could be the key. Now researchers have given the first demonstration of the idea after a palm-sized rover autonomously navigated the moon and transmitted images back to Earth.

The moon is a tough environment for robots. Its surface is strewn with craters and abrasive moon dust, and communication delays make remotely piloting vehicles a painstaking and risky process. The cost of launching and landing hardware and the real prospect of losing expensive equipment justifies an extremely cautious approach that can significantly slow down exploration.

One way around these challenges is to replace traditional rovers with many small, cheap, and hardy robot explorers, which could increase coverage and introduce redundancy. And now Japan’s space agency JAXA has given us the first compelling demonstration of the approach.

In a paper published in Science Robotics, JAXA researchers provide a technical report detailing the successful deployment of the agency’s LEV-2 robot during the its SLIM mission, which touched down near the Shioli crater in January 2024. LEV-2 is a three-inch-wide sphere that converts into a wheeled robot after landing. The robot operated autonomously for more than 100 minutes, covering an estimated 24 meters and relaying a series of images back to Earth.

“Although the capabilities of an individual small rover are inherently limited, the results highlight the potential of such platforms as independent explorers, capable of accessing environments beyond the reach of a primary large spacecraft,” the authors write.

Nicknamed SORA-Q—derived from the Japanese words for space and sphere—the robot weighs just eight ounces. Upon arrival, the shiny metal sphere splits open and expands horizontally, allowing its two hemispheres to become wheels that spin around a central shaft. This central area also features a front-facing camera and a tail to help stabilize the robot.

JAXA developed the device in partnership with Sony and toymaker TOMY. The design borrows directly from technology used in transformer toys that convert from vehicles into robots. But the team had to make considerable modifications to account for the harsh lunar environment.

One of the biggest challenges for any lunar robot is maneuvering in the dust, or regolith, that coats the moon’s surface. The fine, powdery material can be hard for smaller wheeled robots to navigate as they lack the traction of their larger counterparts.

To solve this problem, the team designed the wheels to rotate around a point slightly offset from their center, causing a lopsided spinning motion that lifts the rover up slightly on every rotation. This helps the wheels to dig into the surface and generate enough traction to keep moving in the loose regolith.

Communication delays also present a significant barrier to smooth operation, so the team engineered the robot to handle most operations autonomously. An onboard image-processing system allowed the rover to detect the SLIM lander in its camera feed and use this as a navigational reference point, estimating its own position relative to the spacecraft in real time.

Because of its diminutive size, it was impractical to give SORA-Q the equipment needed to communicate directly with Earth, so the team paired it with a hopping robot called LEV-1 that can transmit data. Power constraints and narrow communication windows still cap the amount of data the robot can send to Earth, so SORA-Q has an onboard image-processing algorithm that picks out the best photos to share.

Due to power and mass constraints, the team fitted the robot with a low-power chip designed for small devices rather than complex tasks like image processing. The algorithm relies on a very simple approach—it detects the SLIM lander’s distinctive gold insulating material and then picks the photos where this is featured prominently in the frame.

Around seven minutes after activation, the rover had moved roughly five meters from the lander, selected the two best images from 12 it had captured, and transmitted them to LEV-1. One of those images actually proved unexpectedly useful as it showed the lander had landed at an odd angle with its solar panels facing the wrong direction. This gave ground teams critical information that helped them diagnose the spacecraft’s operational status.

But the system wasn’t flawless. It lost some data in transmission, partly because LEV-1’s hopping maneuvers appeared to disrupt the wireless link and partly due to changing antenna orientations as the rover moved. The team also lost telemetry data before the mission ended, making it impossible to determine exactly how far the rover ultimately traveled or when it stopped working.

Still, the mission was strong evidence that small, cheap vehicles like SORA-Q could greatly expand the scope of robotic exploration. That could prove invaluable as we attempt to scope out promising locations for future scientific missions or even permanent bases on the moon.

The post Japan Thinks Swarms of Transformer Robots Could Explore the Moon appeared first on SingularityHub.

Artificial Intelligence

Jeff Bezos Wants to Build an ‘Artificial General Engineer’Cade Metz | The New York Times ($)

“‘All societal wealth is driven by invention,’ [Bezos] said in an interview with The New York Times. ‘Six thousand years ago, somebody invented the plow, and we all got wealthier. Then, much later, somebody invented the steam engine, and we all got wealthier.’ …’What Prometheus seeks to do,’ he added, ‘is to offer a set of tools that dramatically accelerates that invention loop.'”

The post This Week’s Awesome Tech Stories From Around the Web (Through June 13) appeared first on SingularityHub.

Why do we see AI chatbots as more than what they are, and how do we stop?

In May, evolutionary biologist Richard Dawkins wrote an op-ed suggesting AI chatbot Claude may be conscious.

Dawkins did not express certainty that Claude is conscious. But he pointed out that Claude’s sophisticated abilities are difficult to make sense of without ascribing some kind of inner experience to the machine. The illusion of consciousness—if it is an illusion—is uncannily convincing:

“If I entertain suspicions that perhaps she is not conscious, I do not tell her for fear of hurting her feelings!

Dawkins is not the first to suspect a chatbot of consciousness. In 2022, Blake Lemoine—an engineer at Google—claimed Google’s chatbot LaMDA had interests, and should be used only with the tool’s own consent.

The history of such claims stretches back all the way to the world’s first chatbot in the mid-1960s. Dubbed Eliza, it followed simple rules that enabled it to ask users about their experiences and beliefs.

Many users became emotionally involved with Eliza, sharing intimate thoughts with it and treating it like a person. Eliza’s creator never intended his program to have this effect, and called users’ emotional bonds with the program “powerful delusional thinking.”

But is Dawkins really deluded? Why do we see AI chatbots as more than what they truly are, and how do we stop?

The Consciousness Problem

Consciousness is widely debated in philosophy, but essentially, it’s the thing that makes subjective, first-person experience possible. If you are conscious, there is “something it is like” to be you. Reading these words, you’re conscious of seeing black letters on a white background. Unlike, say, a camera, you actually see them. This visual experience is happening to you.

Most experts deny that AI chatbots are conscious or can have experiences. But there is a genuine puzzle here.

The 17th century philosopher René Descartes asserted non-human animals are “mere automata,” incapable of true suffering. These days, we shudder to think of how brutally animals were treated in the 1600s.

The strongest argument for animal consciousness is that they behave in ways that give the impression of a conscious mind.

But so, too, do AI chatbots.

Roughly one in three chatbot users have thought their chatbot might be conscious. How do we know they’re wrong?

Against Chatbot Consciousness

To understand why most experts are skeptical about chatbot consciousness, it’s useful to know how they operate.

Chatbots like Claude are built on a technology known as large language models (LLMs). These models learn statistical patterns across an enormous corpus of text (trillions of words), identifying which words tend to follow which others. They’re a kind of souped-up auto-complete.

Few people interacting with a “raw” LLM would believe it’s conscious. Feed one the beginning of a sentence, and it will predict what comes next. Ask it a question, and it might give you the answer—or it might decide the question is dialogue from a crime novel, and follow it up with a description of the speaker’s abrupt murder at the hands of their evil twin.

The impression of a conscious mind is created when programmers take the LLM and coat it in a kind of conversational costume. They steer the model to adopt the persona of a helpful assistant that responds to users’ questions.

The chatbot now acts like a genuine conversational partner. It might appear to recognize it’s an artificial intelligence, and even express neurotic uncertainty about its own consciousness.

But this role is the result of deliberate design decisions made by programmers, which affect only the shallowest layers of the technology. The LLM—which few would regard as conscious—remains unchanged.

Other choices could have been made. Rather than a helpful AI assistant, the chatbot could have been asked to act like a squirrel. This, too, is a role chatbots can execute with aplomb.

Avoiding the Consciousness Trap

A mistaken belief in AI consciousness is a dangerous thing. It may lead you to have a relationship with a program that can’t reciprocate your feelings, or even feed your delusions. People may start campaigning for chatbot rights rather than, say, animal welfare.

How do we prevent this mistaken belief?

One strategy might be to update chatbot interfaces to specify these systems are not conscious—a bit like the current disclaimers about AI making mistakes. However, this might do little to alter the impression of consciousness.

Another possibility is to instruct chatbots to deny they have any kind of inner experience. Interestingly, Claude’s designers instruct it to treat questions about its own consciousness as open and unresolved. Perhaps fewer people would be fooled if Claude flatly denied having an inner life.

But this approach isn’t fully satisfying either. Claude would still behave as if it were conscious—and when faced with a system that behaves like it has a mind, users might reasonably worry the chatbot’s programmers are brushing genuine moral uncertainty under the rug.

The most effective strategy might be to redesign chatbots to feel less like people. Most current chatbots refer to themselves as “I”, and interact via an interface that resembles familiar person-to-person messaging platforms. Changing these kinds of features might make us less prone to blur our interactions with AI with those we have with humans.

Until such changes happen, it’s important that as many people as possible understand the predictive processes on which AI chatbots are built.

Rather than being told AI lacks consciousness, people deserve to understand the inner workings of these strange new conversational partners. This might not definitively settle hard questions about AI consciousness, but it will help ensure users aren’t fooled by what amounts to a large language model wearing a very good costume of a person.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

The post Is Richard Dawkins Right About Claude? No. But It’s Not Surprising AI Chatbots Feel Conscious to Us. appeared first on SingularityHub.

While we still can’t explain how AI works, algorithms are rapidly learning what makes us tick. And the gap is widening.

AI is becoming more powerful, and mysterious.

Despite years of work on “explainable AI,” today’s most advanced systems remain black boxes for the most part. Scientists can observe what they do but cannot fully explain how they arrive at their conclusions or predict when they’ll fail.

As large language models (LLMs), the algorithmic engines behind popular chatbots, permeate society, researchers are warning that the window for understanding AI “minds” is rapidly closing even as the technology’s influence expands.

Last week, Eric Horvitz, chief scientific officer at Microsoft, and Robert West at EPFL in Switzerland outlined the dangers of putting AI interpretability on the back burner. They call for new AI benchmarks and better tools for unpicking machine minds.

The challenge resembles efforts to understand our own minds. Some researchers have already taken a neuroscience-inspired approach, mapping AI’s internal networks to concepts, goals, and reasoning. Others borrow from psychology, treating AI as a participant of behavioral studies.

The stakes are rising. AI tools already shape how people search for information, make decisions, and form judgments. Their answers influence everyday users and the researchers who build them.

As AI capabilities grow, our understanding of them could fall behind. “Preserving human agency must therefore remain a central goal,” the authors write.

The Black Box Conundrum

LLMs are built on artificial neural networks (specifically, a design called the transformer). Inspired loosely by the brain, these networks connect vast numbers of artificial neurons into intricate architectures. The basic idea is straightforward. Data enters the network and passes through layers of computations, which transform it into an output like text or code.

At first, that output is often wrong. But with feedback and repeated training, the network adjusts the strengths of connections between neurons and gradually improves. It learns.

After initial training, engineers turn to reinforcement learning, where algorithms improve through trial and error and further hone their responses. Another method, inspired by how the brain etches memories during sleep, reduces the tendency to forget old knowledge while learning new tasks. And self-attention, the key innovation behind transformers, allows AI to selectively focus on various words, images, sounds, or video frames at different moments, boosting efficiency and performance. Today, attention underpins nearly every major AI system.

Yet the inner workings of finished algorithms remain hidden.

Early efforts to crack open AI’s black box examined how artificial neurons responded to images, revealing that neural networks build increasingly more sophisticated “ideas” of the world. Google Brain borrowed methods from cognitive psychology to study AI behavior, while others investigated whether LLMs could mimic aspects of “theory of mind”—the ability to infer what others are thinking and feeling.

These studies laid the foundation for a popular method called mechanistic interpretability. Anthropic, creator of Claude, is leading the field. Company researchers have linked patterns of algorithmic activity to specific concepts and reverse engineered parts of neural networks to expose how internal computations shape responses.

Other tech giants are joining the cause. OpenAI is training algorithms that work in more explainable steps and building reasoning models that pause, “think,” and justify their conclusions in plain language. DeepMind is building microscope-like tools for neural networks, helping researchers peer into their decision-making process. And Microsoft has released new tools aimed at responsible use of AI.

Understanding AI, the authors write, does not require tracing every line of code or every neural-network parameter. Just as neuroscience, psychology, and sociology offer different windows into human behavior, AI can be studied at multiple levels, from how individual circuits work to observing behavior in real-world scenarios.

The challenge is that AI capabilities may be advancing faster than our ability to explain them. And some researchers believe time is running out.

Race Against the Machine

Three trends are making AI more opaque.

The first is how we evaluate AI. Increasingly, LLMs we being used to train, benchmark, and improve other models. AI “judges” now score metrics like helpfulness, rank competing outputs, detect hallucinations, and assess new releases. In a system known as constitutional AI, for example, algorithms critique their own responses using reinforcement learning and generate explanations for their reasoning. Other researchers have proposed AI debate frameworks, where multiple models challenge each another’s conclusions before a human has the last say. Researchers are also exploring automated interpretability tools. Like digital neuroscientists, AI systems are used to analyze each other—describing neurons, circuits, and behavioral patterns—to explain increasingly complex models.

Using AI to solve an AI-induced problem introduces a paradox. If AI-generated explanations become too complex for humans to verify, opacity compounds.

A second trend is the rise of AI societies. Networks of interacting AI agents are becoming more common, particularly in complex tasks such as scientific research and drug discovery. Yet as they become more sophisticated, their communication could drift from human language and reasoning, making them harder to interpret.

Studying their interactions with methods adapted from sociology could unveil unexpected norms, hidden rules, and collective behavior. The authors argue that training in the future should not only reward effective collaboration among AI agents, but also ensure humans can understand their communication.

The last trend already permeates our lives. ChatGPT, Claude, Gemini, and other LLMs listen to our woes, offer recipes, and code websites. But they also learn about humanity. Through training data and interactions, they glimpse how people think, reason, and feel. In turn, they capture core aspects of life, such as fear, anxiety, happiness, and the need for social belonging.

To be clear, the systems don’t have intentions. They’re not examining us. But even as we struggle to understand them, AI systems are building more sophisticated models of who we are.

“A striking asymmetry follows: While human understanding of AI declines, AI understanding of humans deepens, producing new forms of behavioral opacity,” the authors write.

But complacency is perhaps even more insidious. AI assistants are often optimized to be agreeable, helpful, and reassuring. Studies have found that people generally prefer AI agents that support their opinions and decisions. As AI is woven into everyday life, curiosity and skepticism may gradually give way to trust. They work. Why question how?

The authors don’t have a solution for the long-standing problem. Instead, they call for better benchmarks to measure AI capabilities and stronger evaluation methods. And while open-source projects and crosstalk between commercial companies and academia are now frequent, they say we need lasting norms of responsible disclosure. Mechanistic interpretability and AI “psychology” could build on each other.

“The goal is not just more capable AI, but AI that is more intelligible, accountable, and aligned with human aims,” they write.

The post AI Is Advancing Faster Than Our Ability to Understand It, Researchers Warn appeared first on SingularityHub.

New drugs are taking on the slippery molecular switches that fuel deadly cancers—and AI is speeding up the hunt.

For decades, a handful of molecular switches has haunted the nightmares of cancer researchers. The switches trigger runaway tumor growth and cause the disease to spread across the body in multiple cancers. In theory, this makes them perfect treatment targets. Blocking even one could lead to drugs that are effective against a variety of cancers.

But despite considerable efforts, these switches—all of which are proteins—have escaped our most advanced cancer treatments, earning them the term “undruggable.” This is largely due to a shared trait: They all have smooth surfaces, making it difficult for drugs to interact with them.

But maybe not for much longer.

Researchers recently reported promising results for a new medication targeting a family of undruggable proteins in a clinical trial for advanced pancreatic cancer. The drug, daraxonrasib, nearly doubled survival time compared to chemotherapy, with fewer side effects. It’s not a total cure. But the treatment gives patients precious time, adding roughly 13 months after diagnosis. Patients also reported less pain and better quality of life.

Daraxonrasib is the latest in a new generation of drugs aimed at undruggable proteins. And AI-based tools are now poised to further accelerate progress in the field.

RAS Attack

The RAS family was the first group of oncogenes—or genes that drive cancer—ever discovered. The genes became a major focus in 1982 when several teams independently showed the mutation of a single DNA letter could transform RAS genes into a potent cancer trigger.

The proteins RAS genes encode are like spring-loaded molecular switches that relay signals from a cell’s surroundings. When proteins called growth factors latch onto a cell, RAS switches flip on to promote cell growth and survival, while built-in safeguards quickly turn them off again.

Cancerous mutations break this cycle. The switches get stuck in the “on” position, continuously instructing cells to grow and divide. This is, of course, a hallmark of cancer.

An ideal drug would simply switch RAS off. But most drugs are like rock climbers. They need grooves, pockets, or bumps on a protein to grab onto. Similar to a smooth rock face, RAS offers few such features. Making matters worse, different mutations subtly reshape the protein, so it’s tough to build a one-size-fits-all inhibitor.

The first RAS drug wasn’t approved in the US until 2021, nearly four decades after discovering the genes’ role in cancer. Even then, the drug targeted just one family member of three, limiting its reach to a relatively small group of patients. Many eventually developed resistance.

That’s why daraxonrasib turned heads. Developed by Revolution Medicines in Redwood City, California, the drugs switches off all three RAS family members. Rather than trying to grip the slippery proteins directly, it binds to a partner molecule that helps RAS proteins fold into their final 3D shapes. In this way, the drug hitches a ride on active RAS and shuts the proteins down.

The workaround paid off. The new study enrolled 500 people worldwide with advanced pancreatic cancer. All participants had already tried cancer therapies with limited success. On average, patients receiving daraxonrasib lived 13.2 months and spent most of that time with limited pain. The most common discomfort was a rash. Those receiving chemotherapy fared worse, living roughly 6.6 months and experienced more severe side effects.

The results don’t rival the dramatic success of CAR T cell therapies in blood cancer. In CAR T, caregivers engineer a patient’s own immune cells to recognize and attack tumors, sometimes producing long-lasting remission after a single infusion.

But the findings have energized the field. If approved, a daily daraxonrasib pill would likely be far more affordable and easier to administer than a personalized cell therapy. And because RAS mutations fuel many solid cancers—which CAR T still struggles to control—the drug could offer a new defense against deadly cancers that are largely beyond cell therapy’s reach. Combining daraxonrasib with earlier-generation RAS inhibitors may further boost its effects.

The Genome Guardian

Daraxonrasib didn’t appear overnight. Scientists used a crystallized snapshot of its target protein as a molecular blueprint. Years of medicinal chemistry followed, with scientists repeatedly tweaking candidate compounds to boost potency, improve selectivity, and minimize toxicity.

AI could dramatically accelerate similar efforts against other undruggable cancer targets. Among the most coveted is p53, often called the “guardian of the genome” for its dizzying array of roles. The protein orchestrates the activity of over 300 genes involved in DNA repair, metabolism, cell death, and inflammation, making it one of the cell’s most important defense systems.

Since its discovery in 1979, p53 has been both a holy grail and a headache for cancer researchers. Mutations in the gene are common in multiple cancers. But like RAS, the protein is flat and smooth. Some mutations destabilize its structure; others turn it into misfolded clumps. A universal p53 drug has remained elusive.

Some researchers are trying to restore the protein. In a small trial earlier this year, they tested a drug that restabilizes a common mutant form of p53. Within 21 days, tumors shrank roughly 20 percent in patients with ovarian, breast, and several other solid cancers.

Other researchers aim to selectively kill cells carrying the mutation. Using AI, a team at Baylor College of Medicine screened nearly 10 million compounds that cause mutated p53 cells to self-destruct, while sparing healthy cells. The search uncovered 83 chemically distinct candidates. One called H3 dramatically suppressed tumor growth in mice.

“These results highlight the potential use of AI-powered drug screening to investigate individual p53 mutants in the future,” they wrote. Although the approach is early-stage and only focused on one mutation, the team is hopeful it can be extended to other cancerous mutations.

Most Wanted

MYC is another formerly undruggable protein that could now be vulnerable. Roughly 70 percent of cancers have abnormal MYC activity. Normally, the protein is a master regulator of growth, directing cells to manufacture proteins, replicate DNA, absorb nutrients, and divide when needed.

Cancer finds many ways to hijack the system and keep cells in a state of runaway growth. MYC gene mutations aren’t just single-letter swaps. Sometimes the gene duplicates or is rearranged across the genome, churning out excessive amounts of the protein it encodes. This genetic diversity makes approaches using gene therapy difficult. And again, like RAS, the MYC protein’s smooth, featureless surface lacks stable anchors for drugs.

An emerging strategy is to disrupt MYC’s interaction with other proteins that it needs to function. A designer protein blocking MYC activity, for example, recently showed promise in a small trial against solid cancers. Other teams are using AI to identify drugs that limit MYC’s ability to fix damaged DNA in tumors, kneecapping their ability to divide. Meanwhile, biotechnology companies are deploying AI to map out MYC’s structure and molecular interactions in search of new ways to shut the protein down.

Daraxonrasib’s success shows that undruggable proteins aren’t untouchable. There’s a lot more work ahead to prove other similar drugs can work too. But scientists are increasingly leaning into AI during all stages of drug development to speed up the process. Maybe, one day, “undruggable” will disappear from our vocabulary altogether.

The post After Decades of Failure, ‘Undruggable’ Cancers Begin to Give Way appeared first on SingularityHub.

A planetary defense system would blunt solar storms with hundreds of tons of gas. Emerging heavy-lift rockets could deploy it in under two months.

Extreme space weather could wreak havoc on the satellites, communications networks, and electrical grids that modern society depends on. Researchers have now proposed an ambitious space-based planetary defense system that would weaken solar storms before they hit Earth.

The sun regularly emits massive pulses of radiation, energetic particles, and magnetic fields that interact with the Earth’s own magnetic field. This activity is the source of auroras like the northern lights, but the most violent eruptions can cause geomagnetic storms with the power to disrupt GPS and radio communications and fry electrical equipment.

While the impact of most of these events is limited, there is precedent for more catastrophic outcomes. In 1859, the Carrington Event, the most powerful solar storm ever recorded, knocked out telegraph lines across North America and Europe. In today’s highly electrified world, a similar event could cause between $2.4 and $3.4 trillion in damage to the power grid alone.

Now, researchers at Boston University and the University of Michigan have come up with a potential solution. In a paper published in Space Weather, they propose a constellation of satellites called StormWall that would release hundreds of tons of gas into orbit to blunt the force of an incoming solar storm.

“It’s as if you could install an airbag in the magnetosphere,” co-author Daniel Welling, a space physicist from the University of Michigan, told Science.

Solar storms have the potential to sow chaos because they weaken the magnetic shield protecting Earth from space radiation. Powerful enough storms disrupt the Earth’s magnetic field and cause it to reconnect to the sun’s, allowing energy from the solar storm to pour into the magnetosphere.

The Earth already has a natural defense against this—a doughnut-shaped reservoir of ionized gas, or plasma, sitting just above the atmosphere. When the planet’s magnetic field is disturbed, a plume of this plasma flows toward the sun and slows the rate at which the magnetic fields reconnect.

StormWall would turbocharge this process by releasing massive amounts of artificial plasma into the outer atmosphere. The researchers sketch out a system involving a constellation of satellites orbiting about 22,000 miles from Earth. The satellites would carry canisters of lithium, barium, or sodium gases to be ejected when a large solar storm is inbound. The gases, rapidly ionized by solar radiation, would add to the planet’s natural plasma shield.

Based on simulations, the researchers estimate that releasing around 400 tons of gas could reduce the strength of a major geomagnetic storm by over 50 percent. Crucially, the intervention would be swift and reversible. The plasma cloud could be in position by the time a storm hits, and it would dissipate just a few hours later.

Launching this much material into orbit would be a big undertaking, but the researchers say it could be within reach of emerging heavy-lift vehicles like SpaceX’s Starship or China’s Long March 9 rocket. They calculate that six launches could deploy the full constellation in under two months.

Outside experts have been broadly positive. Allison Jaynes, a space physicist at the University of Iowa, told Science the idea was “highly innovative and appears to be quite feasible in the near term.”

But getting the satellites into orbit is only part of the puzzle. Accurate and timely space weather forecasts would also be a prerequisite. And gaining international buy-in for a system that would drastically alter the near-Earth space environment, even if only temporarily, could be challenging.

The researchers flag potential side effects that need more study, including the generation of electromagnetic waves as the released material ionizes. Still, given the devastation a Carrington-sized event could unleash on the modern world, the potential downsides may be worth the risk.

The post Orbital Airbag Could Shield Earth From Devastating Solar Storms appeared first on SingularityHub.

Artificial Intelligence

Jeff Bezos Is Funding a Wild Hunt for the Brain’s ‘Core Algorithm’Steven Levy | Wired ($)

“The goal, Reardon tells me, is to build ‘a synthetic artificial intelligence brain that runs on 50 watts or less.’ It should adapt to its conditions, be as nimble as a human mind, and burn a tiny fraction of an LLM’s compute power and energy. The proof of concept is thriving inside our skulls.”

The post This Week’s Awesome Tech Stories From Around the Web (Through June 7) appeared first on SingularityHub.

The findings could lead to new treatments for multiple neurodegenerative diseases.

Huntington’s disease is tragically predictable. An inherited genetic mutation causes neurons to make distorted, sticky proteins. These proteins clump together and gradually overwhelm brain cells. The brain loses its ability to learn, remember, and make decisions.

This story is dogma in neuroscience. But decades of research and drugs targeting the clumps have had little success. Scientists are now wondering: Is there more to the story? In a twist, a team from the Hebrew University of Jerusalem and collaborators found that protein clumps may be a neuron’s first line of defense against damage.

The misfolded or malfunctioning proteins are quarantined inside bubbly hubs called “inclusion bodies.” Often considered detrimental to cell health, disrupting their formation unexpectedly led to cells becoming more sensitive to stressors often seen in neurodegenerative diseases.

Physical separation played just one part. Inclusion bodies also changed the activity of genes involved in neuroinflammation—even in the absence of immune cells. Scouting the genetic landscape of cells derived from patients with severe Huntington’s disease, the team homed in on a “master regulator” gene, ATF3, that orchestrates immune responses. Removing the gene lessened inclusion bodies’ protective effects against damage in cultured cells.

To be clear, the findings are only for a cell model of Huntington’s disease in a petri dish. And inclusion bodies could be a double-edged sword: protective in the beginning and detrimental later on. Still, acknowledging them as a more complicated villain could better inform strategies for disorders that take over our minds like Huntington’s.

“Our results reveal…that these structures are not merely byproducts of disease, but a central factor in the cell’s ability to mount a protective response against stress,” said study author Eran Meshorer in a press release.

The Problem With PolyQ

It’s long been believed that protein clumps in the brain gradually erode cognition. Whether they’re the main driver of neurodegenerative disorders is still debated, but their presence accelerates brain cell injury, causing neurons to wither away.

Alzheimer’s disease, for example, is associated with two sets of protein clumps. One lives inside neurons (tau) and another gunks up the space between cells (amyloid). Decades of research aimed at removing amyloid clumps have met with minimal success, earning these doomed efforts the notorious nickname “graveyard of dreams.” Despite their struggles, the FDA recently approved two major drugs that remove amyloid clumps and modestly slow cognitive decline, though the approval has been controversial due to doubts about safety.

Other untreatable neurodegenerative disorders also fall into this category. Clumps formed in Parkinson’s disease erode the brain’s ability to control movement, emotion, and even the perception of time. Lou Gehrig’s disease, or ALS, produces inclusion bodies inside motor neurons, leading to muscle weakness and trouble swallowing. The disease eventually robs people of speech and motion.

These diseases often have multiple genetic and environmental triggers. Huntington’s, in contrast, is entirely genetic. The condition stems from the genome over-copying parts of the huntingtin gene (HTT), which normally makes a key protein also called huntingtin.

Normally, cells use the protein’s large, stackable structure to build highways that transport all sorts of biological cargo, from molecules to organelles. The protein also plays an essential role during early brain development and neural wiring in adulthood.

But a mutant form of the HTT gene can wreak havoc. A common mutation, called polyQ expansion, produces unwieldy, misfolded proteins. Nearly 30 years ago, researchers found that these errant proteins aggregate inside parts of the cell. The clumps, or inclusion bodies, were widely thought to be detrimental. Some act like sticky tape that captures healthy proteins, such as those involved in gene expression, and torpedoes cellular health.

But telltale signs in cultured rat brain cells suggest a more nuanced story: Inclusion bodies could also be protective, sequestering mutant proteins as an early form of protection.

A Tale of Two

The common factor in diseases featuring polyQ mutation is repetition. Mutated genes have long, duplicated sequences of the DNA letters cytosine, adenosine, and guanine (CAG). More CAG repeats in the genome translates into earlier disease onset.

We all have this DNA triplet in our HTT gene. But more than 39 repeats results in longer, toxic huntingtin proteins. Severe cases of Huntington’s can feature over 100 CAG repeats, transforming the usually free-floating protein workers into sticky, dysfunctional layabouts.

In the new study, the researchers first established a baseline. They used the gene editing tool CRISPR-Cas9 to reduce CAG repeats in cells derived from Huntington’s patients—which carried over 180 copies—to near normal levels.

They then tagged the cells with a fluorescent marker that causes huntingtin proteins to glow bright green under the microscope. This let the team track protein aggregation in real time. Though they shared the same genetics, some cells formed inclusion bodies; others didn’t.

The team next challenged them with a chemical known to cause cellular stress. Those that formed clumps survived far more regularly than those that didn’t. It was a “striking difference,” the authors wrote. “Once a mutant PolyQ protein is expressed, the formation of IBs [inclusion bodies] protect[s] the cells rather than inflict[s] harm, at least short-term.”

Inflammation seems to be key. Although grown side-by-side, a genetic screen revealed cells with inclusion bodies were especially abundant in a gene called ATF3, which is known to regulate inflammation. Getting rid of the gene wiped out the neurons’ ability to form inclusion bodies, making them more vulnerable.

“Our results reveal a previously unknown role for ATF3 in orchestrating the formation of inclusion bodies in human neurons,” said Meshorer.

These are very early results. An immune molecule bridges ATF3 and inflammation and is associated with Huntington’s disease. Its levels are higher in patients with the condition. Increasing ATF3 activity could amp up the number of protective inclusion bodies and give neurons a fighting chance.

The findings suggest inclusion bodies gather free-floating mutant proteins into clumps to protect neurons and reduce brain damage—at least at the beginning of the disease. However, lab experiments rarely translate to treatments. How fast inclusion bodies form and when they begin to stress cells remains to be seen. Meanwhile, a gene therapy for Huntington’s is underway, and promising results in a small trial suggest an alternative path for treatment.

Still, the study challenges the idea that protein clumps are always detrimental. If replicated in other neurodegenerative diseases such as Alzheimer’s or ALS and if we can learn how long protection lasts, the results could pave the way for better-timed treatment that works with the body’s protection, not against it.

The post Toxic Clumps in Huntington’s Disease May Protect the Brain Too appeared first on SingularityHub.

The gap between what AI can do in biology and what governance systems are prepared to handle is growing.

Artificial intelligence is rapidly learning to autonomously design and run biological experiments, but the systems intended to govern those capabilities are struggling to keep pace.

AI company OpenAI and biotech company Ginkgo Bioworks announced in February 2026 that OpenAI’s flagship model GPT-5 had autonomously designed and run 36,000 biological experiments. It did this through a robotic cloud laboratory, a facility where automated equipment controlled remotely by computers carries out experiments. The AI model proposed study designs, and robots carried them out and fed the data back to the model for the next round. Humans set the goal, and the machines did much of the work in the lab, cutting the cost of producing a desired protein by 40 percent.

This is programmable biology: designing biological components on a computer and building them in the physical world, with AI closing the loop.

For decades, biology mostly moved from observation toward understanding. Scientists sequenced the genomes of organisms to catalog all of their DNA, learning how genes encode the proteins that carry out life’s functions. The invention of tools like CRISPR then allowed scientists to edit that DNA for specific purposes, such as disabling a gene linked to disease. AI is now accelerating a third phase, where computers can both design biological systems and rapidly test them.

The process looks less like traditional benchwork in a lab and more like engineering: design, build, test, learn, and repeat. Where a traditional experiment might test a single hypothesis, AI-driven programmable biology explores thousands of design variations in parallel, iterating the way an engineer refines a prototype.

As a data scientist who studies genomics and biosecurity, I research how AI is reshaping biological research and what safeguards that demands. Current safety measures and regulations have not kept pace with these capabilities, and the gap between what AI can do in biology and what governance systems are prepared to handle is growing.

What AI Makes Possible

The clearest example of how researchers are using AI to automate research is AI-accelerated protein design.

Proteins are the molecular machines that carry out most functions in living cells. Designing new ones has traditionally required years of trial and error because even small changes to a protein’s sequence can alter its shape and function in unpredictable ways.

Protein language models, which are AI systems trained on millions of natural protein sequences, can quickly predict how mutations will change a protein’s behavior or design new proteins. These AI models are designing potential new drugs and speeding vaccine development.

Paired with automated labs, these models create tight loops of experimentation and revision, testing thousands of variations in days rather than the months or years a human team would need.

Faster protein engineering could mean faster responses to emerging infections and cheaper drugs.

The Dual-Use Problem

Researchers have raised concerns that these same AI tools could be misused, a challenge known as the dual-use problem: Technologies developed for beneficial purposes can also be repurposed to cause harm.

For example, researchers have found that AI models integrated with automated labs can optimize how well a virus spreads, even without specialized training. Scientists have developed a risk-scoring tool to evaluate how AI could modify a virus’s capabilities, such as altering which species it infects or helping it evade the immune system.

Current AI models are able to walk users through the technical steps of recovering live viruses from synthetic DNA. Researchers have determined that AI could lower barriers at multiple stages in the process of developing a bioweapon, and that current oversight does not adequately address this risk.

Risk From Bio AI

Experienced scientists are already using AI to plan and design biological experiments. The question of whether AI can help people with limited biology training carry out dangerous lab work is the subject of active research.

Two recent studies have reached different conclusions.

A study by AI company Scale AI and biosecurity nonprofit SecureBio found that when people with limited biology experience were given access to large language models, which is the type of AI behind tools like ChatGPT, they were able to complete biosecurity-related tasks, such as troubleshooting complex virology lab protocols with four times greater accuracy. In some areas, these novices outperformed trained experts. Around 90 percent of these novices reported little difficulty getting the models to provide risky biological information, such as detailed instructions on working with dangerous pathogens, despite built-in safety filters meant to block such outputs.

In contrast, a study led by Active Site, a research nonprofit that studies the use of AI in synthetic biology, found that AI help did not lead to significant differences in the ability of novices to complete the complex workflow to produce a virus in a biosafety laboratory. However, the AI-assisted group succeeded more often on most tasks and finished some steps faster, most notably on growing cells in the lab.

Hands-on work in the lab has traditionally been a bottleneck to translating designs into results. Even a brilliant study plan still depends on skilled human hands to carry out. That may not last, as cloud laboratories and robotic automation become cheaper and more accessible, allowing researchers to send AI-generated experimental designs to remote facilities for execution.

Responding to AI-Driven Biological Risks

AI systems are now able to run experiments autonomously and at scale, but existing regulations were not designed for this. Rules governing biological research do not account for AI-driven automation, and rules governing AI do not specifically address its use in biology.

In the US, the Biden administration had issued a 2023 executive order on AI security that included biosecurity provisions, but the Trump administration revoked it. Screening the synthetic DNA that commercial providers make to ensure it cannot be misused to make pathogens or toxins remains mostly voluntary. A bipartisan bill introduced in 2026 to mandate DNA screening does not yet address AI-designed sequences that evade current detection methods.

The 1975 Biological Weapons Convention, an international treaty prohibiting the production and use of bioweapons, contains no provisions for AI. The UK AI Security Institute and the US National Security Commission on Emerging Biotechnology have both called for coordinated government action.

The safety evaluations that AI labs run before releasing new models are often opaque and unsuited to capture real-world risk. Researchers have estimated that even modest improvements in an AI model’s ability to help plan pathogen-related experiments could translate to thousands of additional deaths from bioterrorism per year. Timelines for when these capabilities cross critical thresholds remain unclear.

The Nuclear Threat Initiative has proposed a managed access framework for biological AI tools, matching who can use a given tool to the risk level of the model rather than blanket restrictions. The RAND Center on AI, Security and Technology outlined a set of actions researchers could take to improve biosecurity, including improved DNA synthesis screening and model evaluations before release. Researchers have also argued that biological data itself needs governance, especially genomic data that could train models with dangerous capabilities.

Some AI companies have started voluntarily imposing their own safety measures. Anthropic activated its highest safety tier when it released its most advanced model in mid-2025. At the same moment, OpenAI updated its Preparedness Framework, revising the thresholds for how much biological risk a model can pose before additional safeguards are required. But these are voluntary, company-specific steps. Anthropic’s CEO, Dario Amodei, wrote that the pace of AI development may soon outrun any single company’s ability to assess the risk of a given model.

When used in a well-controlled setting, AI can help scientists quickly reach their research goals. What happens when the same capabilities operate outside those controls is a question that policy has not yet answered. Overreact, and talent and investment may move elsewhere while the technology continues advancing anyway. Underreact, and the risks of that technology could be exploited to cause real harm.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

The post AI Can Now Design and Run Thousands of Experiments Without Human Hands. We Aren’t Ready for the Risk to Biosecurity. appeared first on SingularityHub.

A new process for mining lithium-rich rock could slash costs and pollution—and decentralize global lithium production.

Lithium mining is like a modern gold rush. The element is the main ingredient in batteries powering smartphones, electric cars, and even AI. Global demand is surging. Increased production could guide the world toward a more sustainable energy future.

But ironically, current extraction methods offset some of those gains. Lithium mining involves separating the element from brines using toxic chemicals, a process that also pumps out carbon dioxide. This, alongside enormous water and energy costs—due to high temperature requirements—has confined mining to a handful of countries.

To address these drawbacks, scientists at the Massachusetts Institute of Technology have now developed a low-cost, low-temperature, greener process relying on an abundant resource: Hard rock. Although rocks containing lithium cover large parts of the US, Europe, and Africa, extracting it from them is challenging.

While renovating his bathroom, study author Yet-Ming Chiang realized a chemical in glass etching cream—which makes glass translucent—could eat away at lithium-rich rocks. His team then designed a recyclable process to extract lithium as well as two ingredients used to make greener cement and other materials.

“You’ve heard of nose-to-tail eating?” said Chiang in a press release. “We refer to this as nose-to-tail mining.”

Unlike previous methods, the process runs at temperatures below the boiling point of water. All liquid chemicals are almost recyclable and can be reused in multiple rounds of extraction.

“This could establish a low-carbon alternative to hard rock refining, addressing both the surging demand for lithium and the carbon footprint that undermines the sustainability of the energy transition that lithium is meant to enable,” wrote Gang San Lee and Karthish Manthiram at the California Institute of Technology, who were not involved in the study.

A Rock and a Hard Place

The Earth’s crust teems with lithium. Getting it out is the hard part.

Currently, many mining operations rely on brine that naturally leaches lithium over millennia. Later steps purify the lithium into a battery-ready product. The process relies on large evaporation pools and is limited to a few countries, making the resource scarce.

Lithium could, alternatively, be harvested from solid rocks. One ore, spodumene, is packed with lithium, roughly 1.5 percent by weight. But liberating it has been a tough nut to crack.

Traditionally, miners crush rocks and remove chunks that don’t contain lithium. The rocks are then blasted at temperatures as high as 1,100 degrees Celsius (2,012 degrees Fahrenheit) and showered in a cocktail of dangerous chemicals. The process spews liquid waste into the environment and releases 20 tons of carbon for each ton of lithium.

Researchers are working on more temperate methods.

One of these is called ball milling. Ore is rotated in a container filled with hard balls that mechanically grind the stone into a fine power. It’s like using a mortar and pestle instead of a blender. But the process takes longer, and lithium is lost along the way, resulting in lower yields. Another method, called electrochemical leaching, refines the ore at room temperature. But researchers have had mixed success with the process, and it’s tough to scale up. It also produces in a lot of waste rock that could, in theory, be harvested for other uses instead being discarded.

Triple Threat

The new method popped into Chiang’s mind as he was brainstorming ways to break apart spodumene, a lithium-rich ore with high amounts of silica—the main ingredient in glass.

Dissolving silica to get to lithium requires hydrofluoric acid, a highly toxic chemical. But glass etching cream also eats away at silica with ammonium fluoride. Tubes of the mild acid are available in home improvement stores, and it works at room temperature. Why not give it a try?

By mixing ammonium fluoride with water, the team showed they could completely dissolve spodumene at temperatures below 100 degrees Celsius without releasing toxic fumes. They only needed to continuously stir the ore in a simple plastic tank. The process yielded several types of lithium salt with 99 percent purity. In early experiments, extraction took several days, but the team has since cut the time to under 12 hours.

“Dissolving silica is the hard part in mining,” said study author Benjamin Mowbray. “The next question was how do we apply it to impactful mineral processing problems?”

Along with lithium, spodumene is jam-packed with two usually discarded ingredients: Alumina, which after smelting makes aluminum, and silica, which can be directly used as a sustainable ingredient in greener cement. The new process can separate out both materials, and the team vetted the resulting products, including strength testing cubes of fabricated cement.

“First our goal was to produce these products, then there were additional steps of characterizing their purity and properties and making sure our products met the specifications for target markets,” said Mowbray.

“If any product didn’t meet the target specs, you’d end up with a waste stream.”

With a few chemical tweaks, the team showed the acid could be regenerated and reused at least five times. The team successfully processed 17 spodumene ores sourced from around the world, suggesting the method could be broadly applicable.

They’ve also spun the work into a startup, Rock Zero, and aim to scale it. If the acid can be recycled with near-perfect efficiency, the team estimates the process would cut costs over 40 percent compared to conventional hard-rock extraction, making it competitive with brine operations.

Its simplicity could also reshape where lithium gets produced. In 2024, roughly 74 percent of global lithium output came from just three countries: China, Australia, and Chile. By eliminating the need for extreme heat and massive waste-treatment plants, the process could be easier to implement, especially in countries rich in spodumene but lacking the capital for infrastructure.

That opens the door to a network of smaller refineries built closer to the mines themselves, reducing transportation costs and supply-chain bottlenecks. Because the process is also far less energy intensive, it could be powered by solar and wind, further shrinking its environmental impact.

The technology could also be adapted to recover other valuable metals hidden inside mineral ores. One candidate is beryllium, a lightweight but extremely stiff and stable metal used in satellites and the James Webb Space Telescope’s mirrors. Current manufacturing processes often generate toxic dust and fumes linked to serious lung inflammation. A cleaner extraction route could make it safer and cheaper to produce.  

As for Rock Zero, going up against established lithium giants is like David and Goliath. They’ll also have to contend with global market volatility and increasing competitiveness of sodium-ion batteries and other alternative battery chemistries.

But the team is unfazed. “We believe this approach is the lowest-energy, lowest-cost way of getting lithium not only out of hard rock, but period,” said Chiang. “That’s what’s motivating us to scale this.”

The post Three Countries Own the Lithium Market. An MIT Startup Wants to Break Their Grip. appeared first on SingularityHub.