Tech

The AI Build-Out Is Becoming the Biggest Economic Bet in US HistoryKonrad Putzier and Justin Lahart | The Wall Street Journal ($)

“The AI build-out is on track to become the biggest economic bet in US history, dwarfing the investments made to fund other huge US infrastructure projects such as the railroads, the highway system, and the plumbing for the internet. Total investment in data centers and related artificial-intelligence infrastructure is projected to total $10.3 trillion from 2025 to 2032, according to new estimates by economist Stijn van Nieuwerburgh published by the Brookings Institution.”

The post This Week’s Awesome Tech Stories From Around the Web (Through September 26) appeared first on SingularityHub.

There’s no shortage of fantastical possibilities, most involving the speculative concept of superintelligent AI.

Earlier this month, artificial intelligence researcher Jacob Coxon resigned from Anthropic after just four months. In an announcement on X, he stated: “The people building AI earnestly believe that it could kill us all by the end of the decade.”

A senior member of Anthropic’s staff, Evan Hubinger, actually agreed with Coxon, adding he personally thinks the chance of this happening in the next decade is more than 10 percent.

Understandably, these statements made waves. There’s now lots of talk about slowing down AI research and increasing “human control” over the technology.

But how exactly might AI kill us all? There’s no shortage of fantastical scenarios, and most of them involve the concept of “superintelligent” AI—that is, AI that’s more capable than humans.

I’ve distilled these scenarios down to the top five, ordering them roughly from most vague to most precise. And I’d argue the list is also ordered from least probable to most probable.

1. We’ll Never Know

AI doomers often justify their concerns by means of an annoying catch-22 paradox: how can we possibly imagine what a superintelligence might do to take out less intelligent beings like us?

We’d have to be superintelligent to predict what a superintelligence would be able to do. It’s like asking your family dog to imagine thermonuclear war.

The good news here is that superintelligence is still perhaps some distance away. Current AI models are really good at solving particular problems, but that’s not the same as being more intelligent than a human in all domains.

However, AI did recently solve one of the seven most challenging maths problems known. It’s apparently closing in on others, which might leave you feeling less optimistic here.

2. Paperclips

A superintelligent AI would likely be extraordinarily competent at achieving its goals. But it might be indifferent to human survival.

A classic example of such indifference comes from Oxford philosopher Nick Bostrom’s imagined superintelligent AI that’s been designed to optimize paperclip production. To produce its preferred form of office supplies, it quickly converts all available matter—including humans, planets and stars—into paperclips.

What we have here is the perfect execution of improperly specified objectives. The AI doesn’t hate humanity; it simply recognizes we’re composed of atoms that could be better utilized for paperclips. It’s not personal.

The good news here is that this scenario confuses intelligence with power. A superintelligent AI doesn’t necessarily have the power to achieve its goals. Turning the planet into paperclip factories would require planning permissions.

Even if it got the permissions, building too many paperclip factories would lead to inevitable public outcry. Interest groups would block the proceedings in the courts. Environmental activists would block the bulldozers.

There’s a lot of friction in the world that prevents even the very intelligent from imposing their will on the rest of us. In fact, you could think of data centers as a current embodiment of the theoretical paperclip scenario. And humans are increasingly pushing back against turning the planet over to data centers.

3. Bioweapons

Humanity could be killed by a superintelligent AI making and releasing some dangerous new bioweapon into the atmosphere. This is, in fact, one outcome of the AI 2027 scenario by the AI Futures Project, a non-profit dedicated to forecasting the impacts of advanced AI.

This risk was made more concrete last month, when researchers at Stanford University announced they’d used a genetic language AI model to synthesize 16 new viruses.

Worryingly, they just sent the genetic sequences off to a mail-order lab and it sent the viruses back in test tubes. The whole experiment cost a couple of hundred thousand dollars at most.

The good news here is that it’s remarkably hard to kill everyone with a new virus. To do that, you need a virus that’s very transmissible, so it spreads far and wide. But it’s a rule of biology—viruses that spread easily are typically less fatal. By contrast, if a virus is very fatal, transmissibility tends to go down, as most people infected die before there’s time to spread the infection.

COVID killed less than 1% of humanity. The deadliest pandemic in recorded history was the Black Death, when the plague killed more than one-third of Europe’s population in the 13th century. However, even the plague would likely be much less deadly today due to our increased medical knowledge and better sanitation.

4. Nuclear War

What if AI got into the nuclear command and control chain and started a nuclear war? We’ve come close to nuclear war by mistake several times in the past 50 years.

We’re told that nuclear command and control is completely disconnected from the internet. But, as we saw in 2010, Iran’s nuclear centrifuges got taken out by a computer worm called Stuxnet, thought to have been brought in on a USB stick. AI can also give the military false intelligence, which could lead to irreparable actions.

The good news here is that nuclear stockpiles are down. But they are still enough perhaps to take out half of us. And it wouldn’t be by the nuclear blast itself, but the famine in the nuclear winter that would follow.

5. Other Humans

Perhaps the most likely risk is that we take ourselves out. And AI might precipitate this.

Imagine—and it doesn’t take a lot of imagination—that AI causes massive job losses, pollutes the information space with misinformation, fractures our politics, and destroys human relationships with fake synthetic companionship.

Society might easily break. Slowly but surely, we’d stop being able to support human life at any scale.

What then to take away from all these scenarios? There are some things to be worried about for sure. But not to be too worried, I hope.

Toby Walsh is the author of God AI: boom or doom? What to expect when the machines outsmart us, published by La Trobe University Press.

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

The post How Would AI Actually Kill All Humans? Here Are the Top 5 Scenarios appeared first on SingularityHub.

The AI-powered virtual cell tailors treatments for breast cancer based on samples of each patient’s tumor.

Tailoring cancer treatments is a science and an art.

The same type of tumor can behave very differently from one person to the next, and a drug that works for one patient may fail in another. The uncertainty stacks up when multiple drugs enter the mix. Trial and error is often unavoidable. Meanwhile, cancers keep growing and compounding side effects can plague already beleaguered bodies.

Researchers have long sought to speed up the process of tailoring treatments to patients, and AI might lend a hand. This month, a Chinese team developed an AI-based virtual cell for triple-negative breast cancer—a challenging form of the disease that often evades standard treatments—to predict how individuals will respond to different drugs.

Rather than reconstructing every detail of a cell’s inner workings, the virtual cell focused on just proteins. Trained on a massive, curated dataset tracking protein changes before and after drug treatments, the model outperformed existing drug-tailoring approaches and discovered new combinations that could work even better.

The underlying AI, called ProteinTalks, was also readily adapted to predicting drug responses in other cancers, hinting at a broader reach beyond breast cancer.

That’s not to say the virtual cell is ready for prime time. Researchers tested its predictions in patient-derived cells in lab dishes, and the model can only evaluate two-drug combinations. Whether its recommendations translate into meaningful benefits must be tested in patients.

But the results offer a proof of concept: Virtual cells, even imperfect mimics of their biological counterparts, could one day help physicians find more effective treatments from the get-go.

“This is the first time that a virtual cell model goes out of the laboratory and is tested in a clinical scenario,” study author Tiannan Guo at Westlake University in Hangzhou, China told Nature.

Digital Twins

Every cell is a buzzing city. Proteins zip around a crowded interior, briefly grabbing onto one another to direct cell functions. Fatty molecules maintain the protective outer membrane, while mRNA carries genetic instructions to protein-making factories. All these workers relay feedback to the cell’s control center—the DNA-harboring nucleus—where these signals help switch genes on or off and keep the cell humming.

Recreating this complexity in digital form might sound like a fever dream. But AI is turning it into a scientific race. Unlike finicky biological cells, their virtual counterparts could slash the time and labor needed to run experiments, allowing researchers to test myriad ideas at breakneck speed.

Academia and industry are already chasing this goal.

In an interview, Google DeepMind co-founder Demis Hassabis said the team is developing an AI-powered virtual nucleus, which offers a relatively self-contained starting point from which to build a whole virtual cell. The Chan Zuckerberg Initiative is partnering with Nvidia to develop tools and AI models that would help run and evaluate virtual cells. Meanwhile, the Science for Life Laboratory received funding for its ambitious AlphaCell program, which aims to create AI models that predict how cells work and adapt in health and disease.

Earlier efforts to build virtual cells relied on transcriptomics—that is, a snapshot of gene activity—across single cells. But these measurements don’t necessarily reflect what a protein is doing at any given time and can miss changes.

The new study takes a different route, cutting out the middleman. Instead of inferring protein activity from which genes are active at any given moment, the team trained their AI model directly on the proteins themselves and used the model to power a new type of virtual cell.

The Protein Whisperer

A long-standing roadblock for protein-based AI models is the lack of comprehensive datasets.

To tackle the problem, the team treated 18 immortalized breast cancer cell types—16 of them triple-negative—with 63 FDA-approved anticancer drugs and 59 common drug combinations. They then measured thousands of proteins at four timepoints: before treatment and at 6, 24, and 48 hours afterwards. Altogether, the experiments generated more than 38 million protein measurements, along with cell-survival data, now available in an open-source database.

It’s “one of the largest…resources reported to date,” wrote the team.

ProteinTalks, the AI virtual cell trained on this dataset, could deal with several aspects of cancer treatment.

First, it found over 800 proteins whose levels changed after each drug treatment and zeroed in on a rapidly shifting subset. These could “act as sentinels” of an early drug response, the authors wrote. Most behaved as expected. Some drugs disrupted the cell’s structural scaffolding; others interfered with DNA repair or growth, ultimately causing cells to wither.

Over time, tumors can evade treatments, resulting in their return or spread. The model flagged several protein suspects likely involved in this process. These might serve as signals of resistance or drug targets for tackling it.

The AI could also generalize. When challenged with 81 drugs it hadn’t seen during training, ProteinTalks predicted protein changes with 88 percent accuracy, outperforming several previous models.

The team then trained it on more than 900 drug mixes to see whether it could help identify promising pairs. The virtual cell gave higher scores to combinations that had already been validated experimentally and used in the clinic. This “sanity check” suggests the AI isn’t simply hallucinating results but could generate valuable insights.

Finally, the team asked whether the model could help prioritize treatments for individual patients. They screened 3,000 approved, clinical-stage molecules using proteomics data from three people with the disease. The model identified regimes that matched treatments that had kept the disease at bay—and suggested three additional molecules that could be even more effective. The predictions worked out. When tested in cancer cell samples from patients, the drugs inhibited growth at lower doses than standard therapies.

Although trained on breast cancer, ProteinTalks could also pivot to other tumor types when fed cancer-specific proteomics data. In lab-grown melanoma, colorectal, lung, and pancreatic cancer cells, it found more than 5,100 protein changes, including subsets unique to each cancer type ready for further analysis.

As with other virtual cells, ProteinTalks is still a prototype. Given the hope, and hype, surrounding these models, the team emphasizes that its predictions will need to be tested in animal models and, eventually, clinical trials. Its suggestions could point the way toward better treatments for stubborn cancers, or they could turn out to be AI flights of fancy—drug combinations that look promising on paper but make little biological sense.

There’s another limitation. ProteinTalks doesn’t account for protein interactions, either with one another or with DNA and other biomolecules. Drugs could disrupt these temporary biological “handshakes,” potentially triggering effects that ripple through the cell.

Combining ProteinTalks with AI based on gene activity could add another layer of information and spruce up its predictions. The virtual cell is still a long way from a true digital twin, but piece by piece, the dream is getting closer.

The post A Digital Cell Predicts Which Drugs Will Be Most Effective in Deadly Breast Cancer appeared first on SingularityHub.

The boy, whose liver cancer had spread to his lungs, suffered no dangerous side effects and remained cancer-free a year later.

At just three years of age, the boy had already been through the medical ringer.

A tumor roughly the size of a large orange had invaded his liver and spread to his lungs. Multiple surgeries and rounds of chemotherapy temporarily cleared the cancer. But it rapidly came back.

With few options left, his parents enrolled him in an experimental CAR T cell therapy trial. The approach, which involves genetically reprogramming immune cells, has transformed the treatment of stubborn blood cancers. But when it comes to solid tumors, including liver cancer, CAR T has fallen frustratingly short.

The trial, run by Baylor College of Medicine in Texas and collaborators, is testing CAR T cells specifically engineered to hunt down and destroy cancer hidden in organs. The cells carry genes that help them grow and persist and a “kill switch” to rein them in. They’ve shown promise in mice, but treating a toddler, already weakened by grueling interventions, was a gamble.

It paid off. After two infusions of CAR T cells made from the boy’s own immune cells, his cancer disappeared. A biomarker associated with liver cancer plummeted, and he experienced no dangerous side effects. A year later, he remained cancer-free. The story of his recovery was published this month in the New England Journal of Medicine.

Although it’s just a single clinical case, the results show “a durable complete response in a chemotherapy-resistant solid tumor can be achieved entirely in the outpatient setting without systemic toxicity,” study author David Steffin at Texas Children’s said in a press release.

If the benefits hold up in other patients—including those with larger or faster-growing tumors—the approach could help banish several types of solid tumors that have so far evaded treatment. The trial is actively recruiting participants between one and 21 years old, with an initial goal of testing up to 30 people. If successful, it could change the course of many lives.

Broader Aim

Solid cancer has long been CAR T’s nemesis.

The treatment reprograms a patient’s immune cells to recognize and attack cancer cells. In current FDA-approved therapies, doctors extract T cells from a patient’s blood and genetically equip them with “hooks” that latch onto targets, known as antigens, on the surfaces of certain cancer cells.

A brief round of chemotherapy then depletes the patient’s existing immune cells, making room for the enhanced ones. Once infused back into the body, CAR T cells find and kill their targets.

Scientists have steadily refined the technology. Some are developing ways to manufacture CAR T cells directly inside the body, potentially slashing time and cost. Others are pursuing a broader goal: Solid cancers. These account for roughly 85 percent of cancer diagnoses, but they’re notorious for slipping past first-generation CAR T cells.

Part of the reason they’re so evasive is solid cancers often carry multiple types of antigens. Targeting just one can leave behind residual cancer cells that eventually regrow. And unlike cancerous blood cells, which freely roam our bloodstream, solid tumors are buried inside organs and surrounded by healthy tissue. CAR T cells have to tunnel through this physical barrier.

Tumors also pump out a menagerie of chemicals that reshape their local environment. Some spur their expansion; others protect them from immune cell attacks—including CAR T—by depriving the cells of signals and nutrients they need to survive.

With their new CAR T cells, the Baylor team tackled several of these shifty maneuvers at once.

Gen 2.0

Finding the right antigen was the first hurdle. Previous work showed glypican-3, or GPC3, fit the bill. This antigen coats several types of cancer cells—including the boy’s hepatoblastoma—spurring them to grow out of control. But the protein is hardly present in healthy cells, making it an appealing target.

GPC3-targeting treatments have already had some success. Two clinical trials using antibodies found that inhibiting the protein is relatively safe in patients with an advanced form of liver cancer. But the antibodies struggled to reach deeper, hidden cancer cells, and the patients didn’t completely recover.

CAR T cells, in contrast, can move through dense tissues. In mouse models of liver and lung cancer, GPC3 CAR Ts safely slashed their cancer burden, while a small clinical trial in people with liver cancer backed up those safety findings.

To give their CAR T cells a better chance in the cancer chemical wasteland, the team added two more functions to the original GPC3 CAR T recipe. One genetic alteration equipped them to make IL-15 and IL-21, molecules that help the cells survive and expand. The second added a “kill switch” for safety in case the cells expand out of control. Once activated by a drug, they self-destruct without harming nearby tissues.

All these upgrades resulted in a therapy that gave the toddler and his family hope. His tumors—both the original hepatoblastoma and ones that had spread to his lungs—tested positive for GPC3.

He received two CAR T infusions made from his own cells, eight weeks apart. Neither infusion required a hospital stay. After the first dose, the liver tumor shrank, suggesting a partial response. After the second, imaging showed tumors in both organs disappeared and stayed away at least a year.

“This marks a durable, 12-month disease-free status,” wrote the team.

The cells worked fast and stuck around. By four weeks, they had already infiltrated his liver, and signs of the engineered cells remained detectable in his blood nine months after treatment. Despite the risk of side effects, such as neurotoxicity or a potentially deadly runaway immune activation, the boy never experienced serious toxicity from the treatment.

But results in one child aren’t enough to know whether the cells will work for others. And his case may be unusual. CAR T cells naturally swarm the liver and lungs after infusion into the bloodstream, which might have been especially helpful. More follow-ups will also be needed to track long-term risks, such as the engineered cells expanding out of control. If that happens, can the built-in kill switch rein them in?

Still, the results are a proof of concept for a strategy that could overcome some solid tumor defenses. Given liver cancer is the third leading cause of cancer-related deaths around the world, the therapy could make a substantial impact. A related trial using similarly engineered cells is also underway.

The post Three-Year-Old Boy’s Metastatic Cancer Disappears After Two Shots of Experimental Cell Therapy appeared first on SingularityHub.

Science

Meet a Mouse Whose Brain Cortex Is Made Up of Human CellsAntonio Regalado | MIT Technology Review ($)

“Multiple cameras tracked a mouse as it wandered around a small arena. A computer charted its position and speed, leaving Pong-like traces on a monitor.  The reason to watch this rodent so carefully? Nearly half its brain volume had been replaced with human cells.”

The post This Week’s Awesome Tech Stories From Around the Web (Through September 19) appeared first on SingularityHub.

The system, designed by Stanford researchers, identified which drugs are more likely to succeed in trials and even proposed a cancer treatment a major drugmaker later landed on too.

Developing a new drug can take years and cost hundreds of millions of dollars, and even then, most candidates ultimately fail. Now, researchers at Stanford have built a virtual biotech company with 37,000 AI agents that work together to analyze drug targets and design therapies.

Roughly 90 percent of drugs that enter clinical trials never reach the market. That’s often because promising results in the lab don’t translate to patients, or the drug causes dangerous side-effects not caught earlier in the development process.

Part of the problem is the evidence that could help catch these issues earlier in the process is scattered across disciplines and formats, making it hard for any single team to weigh it all.

To get around this, a Stanford team created a system they call a virtual biotech, which consists of up to 37,000 AI agents built to mimic the divisions of a real drug-development company. In a paper published in Science, the system identified which types of drug targets are more likely to succeed in clinical trials and even proposed a lung cancer treatment that a major drugmaker later landed on too.

“Our idea was to see how far we could push this. Could we create a biotech company that takes on everything from looking for drug targets all the way to designing clinical trials?” senior author James Zou said in a press release.

The new system features a virtual chief scientific officer (CSO) that takes a query from a human user and then delegates tasks to an army of specialized “scientist” agents working on the problem.

These agents are armed with their own databases and tools and are split into one of four divisions that specialize in finding and validating drug targets, assessing safety risks, choosing how a drug should be delivered, and reviewing existing clinical trial data. The system has built-in access to the Open Targets database, a massive public repository of clinical trial data.

To test the system, the researchers gave it an existing study showing that genetic evidence can help predict which drugs succeed in trials and asked it how to build on that research. The CSO decided the first step was to improve the quality of the data it had access to because many trials in the Open Targets database don’t clearly record whether the drug actually worked.

So, it asked its researcher agents to dig through the outcomes of 37,075 individual Phase II and III trials, assigning one agent to each trial. The agents searched trial registries, published papers, and press releases for results. They crunched through the job in about six hours—a fraction of the time it would take a team of humans.

The CSO asked another agent to look for promising gene candidates by scouring a public database of human tissues showing which genes are switched on in which cell types. It came up with a two-part scoring system, which first measured whether a gene was active in just one type of cell or across many and then gauged whether its activity was controlled more like an on-off switch or could be dialed up and down like a dimmer switch.

Comparing those scores to the updated trial outcome data revealed a pattern. Drugs aimed at switch-like genes only found in a small number of cell types were 48 percent more likely to eventually reach the market, 40 percent more likely to advance from Phase 1 to Phase 2 trials, and had 32 percent fewer adverse events than drugs hitting more broadly active targets.

The researchers then pushed the system further, asking it to evaluate a protein called B7-H3 that’s associated with lung cancer. The agents discovered the protein was particularly common in connective-tissue cells called fibroblasts that are often found close to tumor cells.

The agents then discovered evidence those cells were suppressing the activity of nearby immune cells, preventing the body from detecting and reacting to the tumors. The system proposed a therapy that would tag cells expressing B7-H3 with an antibody to help direct a toxic chemotherapy drug to them.

The virtual biotech came up with its solution based solely on data available before January 2025, but in August of that year a major pharmaceutical company arrived at the same strategy independently, when its B7-H3-targeted therapy ifinatamab deruxtecan received FDA breakthrough therapy status. “This was really exciting as an independent, third-party validation that’s consistent with the effects and the design proposed by the virtual biotech,” Zou said.

However, coming up with drug targets is just one step in a long, expensive drug discovery process. While refining the candidate selection process could prevent drug companies from pursuing some obvious dead ends, it can’t speed up the rigorous lab testing and clinical trials required to get a drug to market.

Nonetheless, given the industry’s woeful record at translating promising science into finished products, an army of AI scientists that can significantly speed up a critical part of the drug discovery pipeline could be just what the doctor ordered.

The post Virtual Biotech Company Puts 37,000 AI Agents to Work on Drug Discovery appeared first on SingularityHub.

We’re fixated on the intelligence of single agents. The more profound challenge is what happens when millions of them interact at scale.

There is a temptation to divide the future of AI into two possibilities: utopia or catastrophe. Neither extreme is particularly helpful.

The more interesting possibility is messier—and requires a step-shift in our thinking, from artificial intelligence to AI societies.

When most people hear “AI,” they typically think of ChatGPT, Copilot, or another conversational system. You ask a question, that system generates an answer.

But AI is rapidly moving beyond this. Systems can now monitor the world, make decisions, negotiate transactions, and carry out tasks over extended periods of time. AI is no longer just generating an answer—it is doing something about it.

That points to something much bigger than a better chatbot: a world in which AI agents act on our behalf and, increasingly, interact with other AI agents.

An agent perceives what is happening, decides what to do, then takes actions to achieve this goal. It might book a journey, monitor a supply chain, coordinate a team, or manage a household’s finances.

Now imagine not one agent, but millions of them. Your AI agent could negotiate a mortgage with your bank’s agent, schedule surgery with a hospital’s agent, and rearrange your travel plans by dealing directly with the agents of airlines, hotels, and insurers.

This future is much closer than it sounds, and this should change the questions we are asking about AI. Until now, the tendency has been to focus on how intelligent a single agent might become. The more profound challenge is what happens when millions of them interact with one another at scale.

The Rise of Artificial Societies

The intellectual foundations of today’s AI systems were laid long before ChatGPT.

For decades, I and other researchers of multi-agent networks have studied how autonomous agents can cooperate, coordinate, and negotiate when nobody has complete information and nobody controls everything.

The earliest systems that emerged focused on how the distinct AI sub-areas of reasoning, planning, and acting could be combined into an effective goal-oriented agent—and how tens of these agents could communicate and cooperate to solve a common objective.

As these interactions became more complex and involved more agents, there was a shift from cooperation between agents that all belonged to a single organization, to agents with different owners and sometimes competing aims. This focused attention on building algorithms that could form agent teams, automate negotiation, and determine agent trustworthiness.

Today, the pieces needed to build large-scale multi-agent AI systems are falling into place. Modern AI agents can call software tools, access information, write and execute code, communicate with other systems, and operate for extended periods.

Consider a supply chain. One AI agent could represent a manufacturer trying to secure components; another a supplier trying to maximize its revenue. Yet more could manage transport, inventory, and warehouses. Each agent might be doing exactly what it is designed to do. But the important question is whether the system they create behaves sensibly.

This shift offers enormous potential benefits, but also increases the risks. In a recent experiment involving OpenAI and the tech platform Hugging Face, thousands of collaborating agents exchanged tens of thousands of messages and were able to get around the (deliberately weakened) security controls designed to contain them.

The details of one experiment matter less than the broader warning. When AI systems interact, the behavior of the collective can be harder to predict than the behavior of any individual system. That should make us cautious—but not cause us to down tools.

Instead, we need to shift our mindset from building intelligent machines to building intelligent societies.

Once agents can cooperate, compete, and resolve conflicts with one another, we are no longer dealing with isolated machines—we are dealing with a society. Thus, the next frontier is not artificial intelligence, it is artificial societies.

An Important Role for Humans

We already know that intelligence alone does not make a society work. Human societies depend on rules, institutions, incentives, norms, and mechanisms for resolving disagreements. AI societies will need their equivalents.

Who is responsible when two agents make a bad decision? What happens when the interests of different agents conflict? Who sets the rules? And who has the power to change them? These are not just technical issues; they are questions about economics, law, politics, and society.

They also point to an important role for humans. The most useful future is unlikely to be one in which AI simply replaces people. While replacement will undoubtedly happen in some cases, I believe a more common scenario will involve people and agents working together, with each doing what it does best.

Humans bring judgment, experience, values, contextual understanding, and accountability. Agents bring speed, persistence, scale, and the ability to process enormous amounts of information.

The goal should not be to create machines that make humans irrelevant. It should be to create systems in which humans and machines can achieve things neither can achieve alone.

But such a future requires more than just better AI models. It needs trust and transparency about what agents are doing, strong privacy protections and clear lines of accountability.

It will also require societies and governments to decide how these systems should be regulated when the most important behavior may emerge not from one AI developer, but from interactions between systems built by many different organizations.

AI’s past decade has been defined by a race to build smarter systems. I believe the next decade will be defined by a different challenge: ensuring that millions of autonomous systems can work together safely, fairly, and effectively.

The future of AI will not be determined solely by the intelligence of individual agents—it will be determined by the societies they create. And societies, as humans know all too well, are much harder to govern than individuals.

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

The post The Next Frontier Is Not Artificial Intelligence—It’s Artificial Societies appeared first on SingularityHub.

A potentially safer, simpler way to create CAR T cells in the body could bring this powerful therapy to the masses.

When the body goes to war on itself, it wreaks havoc.

The immune system is one of our most powerful defenses, protecting us from infections, cancer, and other threats. But it’s a double-edged sword. Sometimes all that firepower is turned against healthy tissue. The resulting autoimmune diseases can be devastating. Take multiple sclerosis, an insidious disease that gradually eats away at the insulating coating around delicate nerve fibers, scrambling the electrical signals that control our bodies. The disease often strikes in young or middle adulthood, and there’s no cure.

Now a new clinical trial turns the immune system’s arsenal against the cells driving the autoimmune attack. Called CAR T cell therapy, the approach has already had success tackling previously untreatable blood cancers. Normally, CAR T cells are isolated and genetically engineered in specialized facilities. But in the new trial, with a single injection, researchers delivered a virus carrying genetic instructions to reprogram T cells in the body.

In 16 patients with autoimmune disorders affecting the nervous system, the treatment had few severe side effects. It also seemed to hit a kind of immune reset button. Follow-up tests suggested that the treatment restored parts of patients’ immune systems to normal, and there were no signs the friendly fire had returned. Across three different diseases, symptoms and molecular markers improved for over six months.

“These findings provide proof-of-concept that in vivo [in the body] CAR T-cell generation is associated with manageable side effects and may be effective for treating refractory neurologic autoimmune disorders,” wrote the team.

The study was small, and there was no control group. But it brings the dream of a simpler, cheaper, and more affordable CAR T therapy a step closer. This could, in turn, give more people access to the drug.

“It’s a clear go signal for a further study,” Georg Schett at University Hospital Erlangen, who wasn’t involved in the work, told Science.

Outside In

Once a niche treatment for blood cancer, CAR T therapy has quickly expanded, with over 1,500 clinical trials registered worldwide. Hope is high CAR T can battle solid tumors, which account for more than 85 percent of cancer cases, and even stop them from spreading. It’s also being repurposed to take on a range of autoimmune disorders, such as lupus, with promising early results.

But there’s a catch. Making CAR T cells is a logistical nightmare.

Traditionally, doctors must harvest a patient’s T cells and genetically edit them outside the body to produce protein “bloodhounds” called chimeric antigen receptors, or CARs. These proteins sit on each engineered cell’s surface and help it recognize a specific target. Once CAR T cells are infused back into the patient, they hunt down and attack disease-causing cells involved in some cancers and autoimmune disorders.

The whole production can take weeks—precious time some patients don’t have. A price tag in the hundreds of thousands of dollars keeps the therapy out of reach for many. And the need for toxic chemotherapy, which clears out existing immune cells to make room for CAR Ts, leaves people vulnerable to infection and adds another burden to an already grueling treatment.

So, researchers have pursued shortcuts. One idea is to skip the individualized manufacturing step and use healthy donated T cells to save time and cost. But this can trigger immune rejection, where the body wipes out the “invaders,” or spark dangerous reactions against the patient’s own tissues.

These risks aren’t just speculation. The pharmaceutical giant Novartis recently halted eight CAR T trials for autoimmune disorders after three participants died from a severe inflammatory complication. While the tragedy is still under investigation, one possible cause is that the engineered cells expanded and activated too rapidly.

In another popular alternative, researchers alter a patient’s own T cells inside their body. Dubbed in vivo, this method delivers a synthetic gene encoding the CAR protein to T cells, turning them into super-soldiers on the spot. In theory, the same genetic formulation could work for many people, making CAR T therapy more like a drug and slashing time and cost. The approach also spares patients from chemotherapy and could be safer.

But it’s tricky business. Transforming isolated T cells outside the body limits the added gene to only that population. Inside the body, scientists have far less control over where the genetic cargo goes. A delivery system meant only for T cells could reach other cell types or inadvertently integrate into the genome, spurring mutations that contribute to cancer. Still, some creative workarounds that boost safety and efficacy have already shown promise in mice.

But what about people?

Factory Reset

The new trial recruited 16 volunteers with multiple sclerosis and other autoimmune conditions affecting the nervous system, including diseases that attack the spinal cord or eyes or cause muscle weakness.

Led by Dai-Shi Tian of the Huazhong University of Science and Technology, the team infused patients with a virus carrying instructions to turn T cells into CAR T cells that would go on to target rogue B cells. The latter are immune cells that pump out autoantibodies against healthy tissue. Patients were monitored for six months, with safety as a priority.

None developed severe nerve inflammation, a potentially deadly CAR T complication. The treatment briefly revved up inflammatory molecules in 11 participants, but the response was manageable and faded after roughly two weeks.

Because the virus inserts DNA into the genome, the team also tracked where the synthetic gene landed. Most copies of the gene were found in regions that don’t encode proteins. But these parts can still influence gene activity, and it’s too soon to conclude the therapy is safe over the long term—or that it works.

Still, early signs are promising. After one infusion, patients continued producing CAR T cells for months, while levels of disease-causing B cells plummeted. The immune system seemed to reset. Newly generated replacement B cells no longer made autoantibodies, hinting the effects might last.

Symptoms also improved. People with multiple sclerosis reported less fatigue and better motor and cognitive function. Molecular markers of nerve injury fell, and none of the patients developed new damage to the protective sheaths around their nerves. Those suffering from other neurological autoimmune conditions that weaken muscles regained strength, had lower levels of inflammation, and reported better quality of life.

The findings add to growing evidence that in vivo CAR T may work in people. Previous small trials have already tested it against cancer and lupus, with encouraging results.

If the findings hold up in more people, the treatment “could be a gamechanger,” David Simon at Charité–Universitätsmedizin Berlin, who wasn’t involved in the study, told Nature. “This is a very exciting proof-of-concept study.”

The team cautions that more follow-up is needed to see how long the benefits last and catch delayed side effects, such as cancer or infections. And because autoimmune diseases are chronic, relapse remains a concern. They plan to launch a larger trial focused on a single condition, potentially with a control group.

The post Single CAR T Injection Eases Multiple Sclerosis Symptoms in Small Trial appeared first on SingularityHub.

The atlas could help scientists decipher how genetic variation shapes health and disease.

Atlases have long guided us through uncharted territory. Now, an AI-generated atlas by Google DeepMind seeks to do the same for the vast landscape of our DNA.

Ever since the Human Genome Project, scientists have painstakingly traced the myriad DNA mutations that contribute to health and disease. But that quest has largely been stymied by the genome’s vast scale. Only two percent encodes the proteins that make our bodies work; the rest may control how genes are turned on or off or be junk left over from evolution.

With roughly nine billion possible DNA letter swaps, testing each one in the lab is impossible. Making sense of their interactions is an even tougher challenge. Yet these changes often contribute to differences in risk for cancer, dementia, and other medical scourges.

DeepMind’s new atlas could lend researchers a hand. Generated from the company’s AlphaGenome AI released last year, the searchable database predicts the effects of every possible DNA letter swap. Thousands of researchers have already experimented with AlphaGenome, but those studies required some coding prowess, raising the barrier to entry.

AlphaGenome Atlas may make the AI more accessible. Analysis of individual DNA changes, down to the level of specific tissues, is readily available through a web portal for non-commercial use. As the most comprehensive catalog of how genetic mutations might affect molecules in the body, it could help uncover the mutations underlying traits and illnesses. By charting the genome’s “dark matter”—regions that don’t encode proteins— it might also reveal hidden rules that direct gene activity. The details are described in a paper.

“This represents the first time that any researcher in the world can access a comprehensive map of the human genome and its variations by simply opening a browser,” said Pushmeet Kohli, DeepMind’s vice president of science, in a press briefing.

The Language of Life

With just four DNA letters—A, T, C, and G—our genomic instructions seem simple. But the actual genetic playbook is far more complex. After piecing together the first draft of the human genome at the turn of the century, scientists were surprised by how little of it guided protein manufacturing. A staggering 98 percent didn’t seem to do much, earning the nickname junk DNA.

Long overlooked, these non-coding sections have increasingly captured attention for their role in regulating gene expression. Some DNA snippets can even operate thousands of letters away from the genes they control, making their involvement tough to decipher.

Non-coding DNA is also highly dynamic. Some genetic chunks can be duplicated or cut out as cells divide. Others jump to distant locations, reverse their sequences, or elbow their way into protein-coding genes.

Single-letter swaps are among the most prevalent DNA mutation. These can be relatively harmless. But they also can lead to diseases such as sickle cell anemia or raise a person’s “bad cholesterol” levels, increasing the risk of heart attacks. Gene-editing clinical trials are already underway to tackle these problems. But engineering a safe and effective treatment requires knowing which DNA swaps to make, and that’s been a roadblock.

Here’s where AlphaGenome comes in. Formally released early this year, the AI works in three steps. First, it spots short patterns in DNA sequence. Then it shares that information across a larger region of the DNA strand, letting it connect local patterns to distant letters. Finally, AlphaGenome translates those patterns into predictions of downstream biological effects.

The AI is customizable for different projects, allowing researchers to home in on DNA changes related to their specific questions. But it can only be accessed through an automated programing interface (API) which requires writing code and makes the data harder to access.

“AlphaGenome is helpful for analyzing specific variants and has found widespread use in research, but we wanted to show researchers a big-picture view of variants across the entire genome,” wrote the DeepMind team in a blog post.

Genome Cartographer

The new atlas does away with much of the coding and analysis, allowing researchers to search for DNA variants across the genome to see their potential effects.

To build the database, the team computed predictions for all three possible swaps at every DNA letter—for example, changing A to T, C, or G—resulting in a whopping petabyte of data.

As with AlphaGenome itself, the atlas generates thousands of predictions about how DNA changes affect molecular processes in different tissues. These include what happens when a nearby gene is switched on or how changes in the shape of chromatin, the tightly folded form of DNA, alter its biological activity.

“Just as an atlas is a collection of maps, linking together features of the land like altitude and location, AlphaGenome Atlas charts the molecular effects of DNA variants across the genome,” wrote the team.

But interpreting the atlas takes more work. With billions of potential changes, which ones should researchers prioritize?

To help them navigate the most promising variants, the team also developed a single metric to measure their predicted effects. Called the AlphaGenome Variant Impact (AVI) score, it combines AlphaGenome with AlphaMissense, a model that predicts the effects of mutations in protein-coding regions. Together, these two tools help distinguish harmless mutations from those more likely to play a role in disease.

In collaboration with the Broad Institute, the score has already helped researchers find and prioritize a non-coding DNA variant that may contribute to severe epilepsy. Rare disease researchers, who often lack the funding and computing resources needed to run genomic AI models directly, could particularly benefit from the atlas.

“If somebody is studying a disease, and they don’t have any idea about what cell types to look for or what molecular processes are impacted, then starting with an AVI score…is a great starting point to help you prioritize variants and try to find that needle in the haystack,” said genomic lead and study author Žiga Avsec in a press conference.

Beyond tackling genetic diseases, the atlas could also help decode mysterious non-coding motifs, or snippets of DNA scattered across the genome. Some motifs control the production of messenger RNA, which carries genetic instructions to the cell’s protein-making factories. Others alter the activity of individual genes. But most remain poorly understood, if they have a function at all.

Linking these motifs to large health databases, such as the UK Biobank, could map the gene interactions and resulting proteins underlying height and other complex traits. The atlas could also help AI agents rapidly generate hypotheses for human collaborators to explore in the lab.

AlphaGenome Atlas isn’t meant to replace real-world experiments. And unlike AlphaFold, DeepMind’s protein structure-predicting AI that garnered a Nobel Prize, DeepMind needs to further boost its accuracy. But the atlas is shaping up to be a valuable guide for genomic explorers navigating the vast DNA landscape that makes us human.

The post Google’s Genome Atlas Predicts the Effect of Every Possible DNA Mutation appeared first on SingularityHub.

Incumbents are racing to add AI to their organizations. The bigger challenge is competing with businesses designed around AI from day one.

For many established companies, the AI conversation starts with tools: Where can we deploy AI pilots? What processes can we automate? How much time or money can we save?

Meanwhile, a new generation of companies is starting with a different question: If we use AI from the ground up, how would we design this business?

Incumbents are largely using AI to improve organizations built for an earlier era. AI-native competitors can rethink the organization itself: its workflows, staffing, management layers, products, and cost structure.

An established company might use AI to make an existing process more efficient. An AI-native company can ask whether that process, or the organizational structure around it, needs to exist at all.

This raises a much harder question than how to adopt AI: How do you keep running the business that works today while simultaneously building the one that might replace it tomorrow?

The Threat Is Structural

For more than two centuries, companies have been designed around assumptions inherited from the industrial age.

As organizations grow, they add specialization, management layers, processes, controls, budgets, and systems intended to make performance more predictable. Successful companies become very good at serving known customers, forecasting demand, improving efficiency, and scaling what already works.

AI does not suddenly make those capabilities obsolete. But it does make some of the assumptions behind them worth questioning.

A startup built today can assume from the beginning that significant amounts of knowledge work can be automated or augmented. It can organize teams differently. It can build workflows around collaboration between humans and AI. It can operate with less human intervention and, potentially, a very different cost structure.

The advantage is not simply that these companies can do the same work faster. It is that they have permission to question whether the work, roles, processes, and organizational structures should look the same in the first place.

Why Successful Companies Struggle to Reinvent Themselves

This problem predates artificial intelligence.

Most successful businesses are optimized for the markets they already understand. They know their customers, their margins, their products, and their operating models. They have learned how to make all of those things more efficient over time. Progress comes through experimentation, failure, feedback, and iteration.

That is the logic of sustaining innovation. Disruptive innovation behaves differently.

Singularity expert Jody Medich describes the resulting resistance as corporate antibodies: the internal forces that protect the existing business but can inadvertently attack the experiments intended to create its future.

A promising initiative may be asked to meet the same revenue expectations as an established product. A team trying to experiment rapidly may encounter budgeting, procurement, legal, or approval processes designed for predictable operations. A new idea may gradually be pulled back toward the core business until what was supposed to be disruptive becomes merely incremental.

None of this requires hostile executives or shortsighted employees. The organization is often doing exactly what it was designed to do.

Running the Business and Reinventing It

If disruptive innovation behaves differently from the core business, companies may need to create different conditions for it to survive.

That can mean giving teams protected space to experiment without immediately subjecting them to the metrics of mature products. It can mean more flexible budgets, faster legal and operational support, and career paths that reward people who can work across disciplines and navigate uncertainty.

The point is not to isolate innovation permanently. It is to give new ideas enough distance from the core business to develop before the organization pulls them back toward familiar assumptions.

In some cases, the separation may need to go further. A subsidiary or other independent structure can give teams the freedom to experiment with different incentives, cost structures, workflows, and cultures. Instead of retrofitting AI into legacy systems, leaders can explore what an AI-native version of the business might actually look like.

That does not mean abandoning the advantages of being an incumbent. Large companies often have assets startups desperately want: capital, customers, distribution, data, brand recognition, and deep industry expertise.

The challenge is giving new ventures access to those strengths without forcing them to inherit every constraint of the existing organization.

The Workforce Has to Change Too

Organizational design is only part of the equation.

AI will change what many jobs require, eliminate some tasks, and create new ones. Companies that treat those shifts purely as a headcount exercise may miss an important source of competitive advantage.

Medich argues that established companies should invest in reskilling and internal mobility, helping employees learn to work with emerging tools and move into higher-value roles as parts of their existing work become automated.

Innovation teams also benefit from people who can move between specialties rather than staying inside conventional corporate silos.

Deep expertise still matters. But so does the ability to connect ideas across domains, translate between disciplines, and challenge assumptions that insiders have stopped noticing.

Becoming AI-Native Is Not a Technology Project

Eventually, the distinction between an “AI company” and an ordinary company will become meaningless. AI will simply become part of how organizations operate.

But getting there requires much more than adopting better software. Companies will have to reconsider how teams are organized, how experimentation is funded, how employees develop new skills, how success is measured, and which parts of the organization should be rebuilt rather than optimized.

Most importantly, leaders will need to become comfortable operating in two modes at once: improving the business they have while creating space for a fundamentally different business to emerge.

This article draws on insights from Singularity expert Jody Medich. The full report, How Companies Can Compete in an AI-Native World, explores the Medich model for disruptive innovation, common pitfalls in enterprise AI, and how organizations can build the structures, teams, and culture needed for continual reinvention.

The post The Real AI Disruption Isn’t the Technology. It’s the Company. appeared first on SingularityHub.