Singularity HUBModify

2026-03-21 22:00:00

The post This Week’s Awesome Tech Stories From Around the Web (Through March 21) appeared first on SingularityHub.

2026-03-20 22:00:00

Few terms are as closely associated with AI hype as artificial general intelligence, or AGI. But Google DeepMind researchers have now proposed a framework that could more concretely measure how close models are to this tech industry holy grail.

Artificial general intelligence refers to a mythical AI system that can match the general and highly adaptable form of intelligence found in humans. As the number of tasks that large language models can tackle has rocketed in recent years, there’s been a growing chorus of voices suggesting the technology is creeping ever closer to this threshold.

But so far, there’s been no clear way to assess progress toward AGI, leaving plenty of room for speculation and exaggeration. To address this gap, a team from Google DeepMind has introduced a new cognitively inspired framework that deconstructs general intelligence into 10 key faculties. More importantly, they propose a way to evaluate AI systems across these key capabilities and compare their performance to humans.

“Despite widespread discussion of AGI, there is no clear framework for measuring progress toward it. This ambiguity fuels subjective claims, makes it difficult to track progress, and risks hindering responsible governance,” the researchers write in a paper outlining their new approach. “We hope this framework will provide a practical roadmap and an initial step toward more rigorous, empirical evaluation of AGI.”

This isn’t DeepMind’s first attempt to clarify the term. In 2023, the company proposed separating AI systems into different levels of capability, in much the same way self-driving systems are categorized.

But the approach didn’t really propose a way to measure what level AI systems have reached. The new framework goes further by building a firmer conceptual footing for the key aspects underpinning model performance and a practical way to evaluate and compare systems.

Digging through decades of research in psychology, neuroscience, and cognitive science, the researchers identify eight basic cognitive building blocks that they say make up general intelligence.

These include the perception of sensory inputs and generation of outputs like text, speech, or actions. Add to those learning, memory, reasoning, and the ability to focus attention on specific information or tasks. Rounding out the list are metacognition—or the ability to reason about and control your own mental processes—and so-called executive functions, like planning and the inhibition of impulses.

The researchers also outline two “composite faculties” that require several building blocks to be applied together. These are problem solving and social cognition, which refers to the ability to understand and react appropriately to the social context.

To judge how well AI systems perform on each measure, the researchers suggest subjecting them to a broad suite of cognitive evaluations that target each specific ability. They also propose collecting human baselines for each task. This would involve asking a demographically representative sample of adults with at least a high school education to complete them under identical conditions.

The results of these tests can then be combined to create “cognitive profiles” that give a sense of a model’s strengths and weaknesses. And by comparing the results against the human baselines, it should be possible to determine when a system matches or surpasses the general intelligence of an average person.

Crucially, the framework focuses on what a system can do rather than how it does it, which means the evaluation is agnostic about the underlying technology. However, the researchers concede that there is currently no good way to measure many of the core cognitive capabilities identified.

While there are already well-established benchmarks for faculties like problem solving and perception, there are no reliable tests for things like metacognition, attention, learning, and social cognition. In addition, many of the best benchmarks are public, which means the testing criteria are easily accessible and may have already been included in model training data. So the authors say they’re working with academics to build more robust, non-public evaluations to fill the gaps.

How useful the new framework will be depends on several factors. First, it remains to be seen whether the criteria identified by the DeepMind team truly capture the essence of human general intelligence. Second, they need to prove that acing this test actually leads to better performance on practical problems compared to narrower, specialist AI systems.

But considering the hand-waving nature of the debate around AGI so far, any framework grounded in well-established cognitive theory and rigorous evaluation represents a significant step forward.

The post Google DeepMind Plans to Track AGI Progress With These 10 Traits of General Intelligence appeared first on SingularityHub.

2026-03-19 22:00:00

In the past few months, a wave of tech corporations have announced significant staff cuts and attributed them to efficiency gains driven by artificial intelligence.

Companies such as Atlassian, Block, and Amazon have announced they would lay off thousands of employees due to increased reliance on AI.

The narrative these companies offer is consistent: AI is making human labor replaceable, and responsible management demands adjustment.

The evidence, however, tells a more nuanced story.

The Automation Story Is Partly True

Genuine disruption is visible in specific corners of the labor market, though the scale of that disruption is commonly overstated. Research from Anthropic published earlier this month shows that although many work tasks are susceptible to automation, the vast majority are still performed primarily by humans rather than AI tools.

Moreover, some occupations are more exposed to displacement than others: Computer programmers sit at the top of the list, followed by customer service representatives and data entry workers. Yet even within the most exposed occupations, AI use is still limited.

The aggregate economic data reflects this reality. A 2025 Goldman Sachs report estimated that if AI were used across the economy for all the things it could currently do, roughly 2.5 percent of US employment would be at risk of job loss.

That’s not a trivial number. However, the report notes that workers in AI-exposed occupations are currently no more likely to lose their jobs, face reduced hours, or earn lower wages than anyone else.

The report does note early signs of strain in specific industries. Goldman Sachs identifies sectors where employment growth has slowed that align with AI-related efficiency gains. Examples include marketing consulting, graphic design, office administration, and call centers.

In the tech sector, US workers in their 20s in AI-exposed occupations saw unemployment rise by almost 3 percent in the first half of 2025. Anthropic’s research also found that job-finding rates (the chance of an unemployed person finding a job in a one-month period) for workers aged 22–25 entering AI-exposed occupations have fallen by around 14 percent since the launch of ChatGPT in 2022. This is a tentative but telling signal about where the pressure is being felt first.

These are meaningful signals, but they are sector-specific and concentrated—not the evidence of sweeping displacement that corporate announcements often imply. That gap between the evidence and the rhetoric raises an obvious question: What else might be driving these decisions?

What Is the Motive?

The timing and framing of the layoffs attributed to AI warrant closer examination. Corporate restructuring, over-hiring during the post-pandemic boom as demand for online services soared, and pressure from investors to demonstrate improved profit margins are all forces operating at the same time as genuine advances in AI.

While these are not mutually exclusive explanations, they are rarely acknowledged alongside one another in corporate communications.

There is a powerful financial incentive for companies to be seen to be embracing AI aggressively. Since the launch of ChatGPT, AI-related stocks have accounted for about 75 percent of S&P 500 returns.

A workforce reduction framed around AI adoption sends a signal to investors that a straightforward cost-cutting announcement does not. A company making AI-related innovations looks a lot better than one sacking staff due to declining revenues or poor strategic decisions.

It is also worth distinguishing between two kinds of workforce reduction. In the first, AI genuinely increases productivity to the point where fewer workers are needed to produce the same output. In the second, staff reductions are not a consequence of AI, but a way to fund it.

Meta illustrates this distinction. The social media giant is reportedly planning to lay off as much as 20 percent of its workforce, while simultaneously committing $600 billion to build data centers and recruit top AI researchers.

In this case, the workers being let go are not being replaced by AI today; they are subsidizing the AI bet their employer is making on the future.

The More Plausible Future

The big picture is likely one of transformation rather than elimination. According to a recent PwC report, employment is still growing in most industries exposed to AI, although growth tends to be slower than in less exposed sectors.

At the same time, wages in AI-exposed industries are rising roughly twice as fast as in those least touched by the technology. Workers with AI skills command an average wage premium of about 56 percent across the industries analyzed.

Together, the data points toward a flattening of the traditional workplace pyramid rather than mass displacement. Firms require fewer junior employees for routine analytical and administrative work, while experienced professionals who deploy AI tools effectively become more productive and command greater value.

AI is a consequential technology and will have a significant impact in the long term. What is in doubt is whether the dramatic, AI-attributed workforce reductions announced by individual companies accurately reflect that trajectory, or whether they conflate genuine technological change with decisions that would have been made regardless.

Making this distinction is not merely an academic exercise. It shapes how policymakers, educators, and workers themselves understand the nature of the disruption they are navigating.

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

The post Tech Companies Are Blaming Massive Layoffs on AI. What’s Really Going On? appeared first on SingularityHub.

2026-03-18 05:54:35

It’s hard to picture a keyboard layout other than the one we know best. From laptops to smartphones, it’s an integral part of our digital lives.

Scientists at Massachusetts General Hospital have now restored the ability to communicate by keyboard to two people with paralysis—using their thoughts alone.

Both people already had brain implants that could record their minds’ electrical chatter. The new system translated brain signals in real time as each person imagined finger movements. The system then accurately predicted the character they were trying to type.

The system learned to translate brain activity to physical intent after just 30 sentences. Typing speeds reached 22 words per minute with few errors, nearly matching speeds of able-bodied smartphone users.

“To our knowledge, this system provides the fastest… [brain implant] communication method reported to date based on decoding from hand motor cortex,” wrote the team.

The participants are part of the BrainGate2 clinical trial, a pioneering effort to restore communication and movement by decoding neural signals in people who have lost the use of all four limbs and the torso. One of the participants previously used the implants to translate his inner thoughts into text, but with mixed success.

Controlling a digital keyboard is far more intuitive and familiar, which makes it easier to grasp. Once a person learns to use the system, they don’t have to look at the keyboard, giving their eyes a break as they type with their minds. It also allows users full control of when, or when not, to share their thoughts, preventing accidental leakage of private musings onto a screen or broadcasted with AI-generated speech.

All Hands on Deck

Parts of the brain hum with electrical activity before we speak. Over the past decade, brain implants—microelectrodes that listen in and decode signals—have translated these seemingly chaotic buzzes into text or speech, allowing paralyzed people to regain the ability to communicate.

Methods vary. Some hardware takes the form of wafer-thin disks sitting on top of the brain and gathering signals from vast regions; other devices are inserted into the brain for more targeted recordings.

These systems are life changing. In a recent example, an implant translated the neural activity controlling a man with ALS’s vocal muscles. With just a second’s delay, the system generated coherent sentences with intonation, allowing him to sing with an artificial voice. Another device turned a paralyzed woman’s thoughts into speech with nearly no delay, so she could hold a conversation without frustrating halts. People have also benefited from a method that uses the neural signals behind handwriting for brain-to-text communication.

Brain implants aren’t purely experimental anymore: China recently approved a setup allowing people with paralysis to control a robotic hand. It’s the first such device available outside of clinical trials.

Perhaps the most widely used clinical solution is eye-tracking. Here, patients move their eyes to focus on individual letters, one at a time, on a custom digital keyboard. But the pace is agonizingly slow and prone to error. And prolonged screen time strains the eyes, making extended conversations difficult.

“Those systems take far too long for many users,” said study author Daniel Rubin in a press release, causing them to abandon the technology.

Tapping Away

For people who already know how to type, the standard keyboard layout—known as QWERTY—feels familiar and comfortable. Fingers stretch to hit letters in the upper row, tap directly down for ones in the middle, and curl into a loose claw to hit bottom letters and punctuation.

As fingers dance across the keyboard, parts of the motor cortex that control their motion spark with activity, precisely directing each placement. Mind-typing using a familiar keyboard, compared to a custom one, could feel more intuitive and relaxing.

Two people with tetraplegia gave the idea a shot. Participant T17 was diagnosed with ALS at 30, a disease that slowly destroys motor neurons, weakening muscles and eventually impairing breathing. Three years later, when he enrolled in the study, he’d lost control of his vocal muscles and relied on a ventilator. He could move only his eyes, but his mind was still sharp. The second participant, T18, was paralyzed by a spinal cord injury 18 months before enrollment. Both had multiple brain implants in different areas. These were connected to cables that shuttled recordings to a computer system for real-time processing.

The participants used a simplified QWERTY digital keyboard containing all 26 letters, a space key, and three types of punctuation—a question mark, comma, and period. To train the system, the volunteers imagined stretching, tapping, or curling their fingers to type text prompts, while implants captured and isolated neural signals for each finger. After training, a deep learning model predicted intended characters, and a language model continuously attempted to autocomplete the sentence.

After practicing just 30 sentences, both participants could copy on-screen text or type whatever they wanted. When asked “what was the best part of your job,” T18 cheekily replied “the best part of my job was the end [of] the day.” Meanwhile, T17, a fan of The Legend of Zelda video games, told the researchers “you should try oracle of ages and seasons…another is skyward sword…the music in those games is great.”

Their typing speeds broke records. T18 communicated at 110 characters or roughly 22 words per minute, which is 20 characters more than a previous state-of-the-art method based on handwriting, wrote the team. The rate is nearly on par with able-bodied smartphone users similar to his age. Typing errors were consistently low and neared perfection after practice.

T17, with incomplete locked-in syndrome due to ALS, typed 47 characters a minute at a higher error rate. He had full use of his vocabulary, unlike with previous systems that imposed word restrictions, and communicated much faster.

The performance differences could be due to where their implants are located. T18’s microarrays are on both sides of the brain, with some covering an area that controls all four limbs. T17’s implants are on only the left half of his brain, with less coverage of finger motor areas.

The team is now tweaking the system for longer use tailored to individuals. As disease progresses, the link between brain signals and keyboard characters may drift and produce more errors. But updating the algorithm is easy. The system needs only a few sentences to learn, so users could start each day mind-typing some thoughts to keep things dialed in.

Updates to the digital keyboard, like adding numbers or the return and delete keys, are in the works. Temporarily disabling the language model could also let participants type strong gibberish passwords, internet slang (ikr, btw, lol), and other non-standard words without being autocorrected.

The brain implant “is a great example of how modern neuroscience and artificial intelligence technology can combine to create something capable of restoring communication and independence for people with paralysis,” said study author Justin Jude.

The post Brain Implants Let Paralyzed People Type Nearly as Fast as Smartphone Users appeared first on SingularityHub.

2026-03-17 07:20:21

Five years ago, scientists watched in wonder as synthetic bacteria grew and split into daughter cells. The bacteria’s extremely stripped-down genome still supported its entire life cycle. It was a crowning achievement in synthetic biology that shed light on life’s most basic processes.

These processes can now be viewed digitally. This month, a team at the University of Illinois at Urbana-Champaign developed a virtual model of the bacteria tracking nearly all of a cell’s molecules down to the nanoscale. The researchers made this digital cell by combining several large datasets covering thousands of molecules and then animating them as the bacteria split in two.

The model is the latest in a growing effort to make digital twins of living cells. Mimicking diseases or treatments in the digital world offers a bird’s-eye view of cellular changes and could speed up drug discovery and help researchers tackle complex diseases like cancer.

“We have a whole-cell model that predicts many cellular properties simultaneously,” study author Zan Luthey-Schulten said in a press release. The model could provide “the results of hundreds of experiments” at the same time, she said.

Digitizing Life

Every cell is a bustling metropolis. Proteins orchestrate a vast range of cellular responses. RNA molecules carry instructions from genes to the cell’s protein-building factories. Fatty acids in a cell’s membrane rearrange themselves to admit nutrients or ward off invaders. Working in tandem, they all keep the cell humming along.

This complexity makes cells hard to simulate. But with large datasets charting the genome, gene expression, and proteins alongside sophisticated AI, scientists have built static virtual cells that paint a near-complete picture with atomic-level resolution. More recent models can even predict molecular movements for a short period of time (often less than a second).

But they can’t simulate “the mechanics and chemistry that take place over minutes to hours in processes such as gene expression and cell division,” wrote the University of Illinois team.

Other efforts use physics to predict how molecular changes affect behavior in bacteria, yeast, and human cells. These treat cells as a “well-stirred system”—that is, a cup of molecular soup lacking details about where each molecule sits and how molecules vary from cell to cell.

But location is key. As cells divide, some proteins gather around DNA to help copy it; others assemble near the membrane to recruit fatty molecules for its growth as the cell splits in two.

Simulating everything, everywhere, all at once during human cell division is beyond even the most powerful supercomputers. Minimal bacteria offer an alternative. These synthetic bacteria are stripped-down versions of the parasite Mycoplasma mycoides. The team focused on one of these known as JCVI-syn3A. Its 493-gene genome—roughly half the original—is the smallest set of DNA instructions to boot up a living bacteria that can still grow and divide.

In 2022, the team developed a 3D model of the bacteria’s metabolism, genes, and growth. But the software, Lattice Microbes, struggled to track division.

Life in 4D

The new study added more data to the software. This included membrane changes and information about how ribosomes, the cell’s protein-making machines, assemble and move inside the cell’s gooey interior. They also added stochasticity, or unpredictability, to the model.

Changes to the location of chromosomes, which house DNA, are random as the cell divides, which makes them difficult to predict. But their position influences DNA replication and gene expression.

The first update nearly broke the software. It could map molecules involved in cell division, such as an enzyme critical for DNA copying. But adding chromosome location predictions slowed the model to a crawl, even when running on advanced GPUs. Most of the cells died before their simulations were complete.

Several tweaks helped. One was to add more computational power. The team used a GPU dedicated to chromosomes, while all other details were processed on a separate chip. The model also ran faster by rendering some proteins as inert spheres that could be largely ignored.

The upgrades worked. Leaving the model running over Thanksgiving, the team returned to find it had completed the bacteria’s whole life cycle. “All of a sudden, it was just this huge leap,” study author Zane Thornburg told Nature.

The simulation matched many real-world experiments, such as how the cells elongate and bubble into dumbbell-like shapes during division. The model also accurately predicted the length of a cell cycle and captured a wide range of cellular activity.

“I can’t overstate how hard it is to simulate things that are moving—and doing it in 3D for an entire cell was…triumphant,” said Thornburg.

Every cell is like a snowflake: Although containing similar molecules, the amounts and locations differ. The model easily handled this diversity. Repeated simulations of the bacteria, each starting with slightly different genetic, molecular, and metabolic makeup, resulted in a similar cycle length and movement of chromosomes during division.

The results came at a cost: Simulating the cell’s 105-minute cycle took up to six days on a supercomputer. But the virtual cell could lend insights into the molecular dance that causes all cells to grow and divide. JCVI-syn3A doesn’t have the smallest genome. Its predecessor holds the record, but it also struggles to make normally shaped and functional daughter cells—suggesting some genes are essential for division. Simulation could help us understand why.

Other efforts using generative AI to build virtual cells are in the works. But because this study’s model was grounded in strict physical and biochemical rules, results could be easily verified in the lab. AI-generated virtual cells, however, are commonly trained on gene expression data alone, which is a snapshot of a cell’s state and often fails to predict complex cell responses.

The two approaches could inspire each other by homing in on principles that make a virtual cell run like the real deal. For example, they could show that capturing each molecule in space and time, rather than as a soup, vastly improves the model.

Although the model can’t simulate a cell atom-by-atom, the team wrote, it could “illuminate the interwoven nature of the biology, chemistry, and physics that govern life for cells.”

The post Digital Twin of a Cell Tracks Its Entire Life Cycle Down to the Nanoscale appeared first on SingularityHub.

2026-03-14 22:00:00

The post This Week’s Awesome Tech Stories From Around the Web (Through March 14) appeared first on SingularityHub.