<![CDATA[High-potency cannabis surges; psychiatry confronts psychosis risk, dependence, and data gaps—why clinicians must guide safer use now.]]>
Treating enterprise AI as an operating layer
There’s a fault line running through enterprise AI, and it’s not the one getting the most attention. The public conversation still tracks foundation models and benchmarks—GPT versus Gemini, reasoning scores, and marginal capability gains. But in practice, the more durable advantage is structural: who owns the operating layer where intelligence is applied, governed, and improved. One model treats AI as an on-demand utility; the other embeds it as an operating layer—the combination of operation software, data capture, feedback loops and governance that sits between models and real work—that compounds with use.
Model providers like OpenAI and Anthropic sell intelligence as a service: you have a problem, you call an API, you get an answer. That intelligence is general-purpose, largely stateless, and only loosely connected to the day-to-day operations where decisions are made. It’s highly capable and increasingly interchangeable. The distinction that matters is whether intelligence resets on every prompt or accumulates over time.
Incumbent organizations, by contrast, can treat AI as an operating layer: instrumentation across operations, feedback loops from human decisions, and governance that turns individual tasks into reusable policy. In that setup, every exception, correction, and approval becomes a chance to learn—and intelligence can improve as the platform absorbs more of the organization’s work. The organizations most likely to shape the enterprise AI era are those that can embed intelligence directly into operational platforms and instrument those platforms so work generates usable signals.
The prevailing narrative says nimble startups will out-innovate incumbents by building AI-native from scratch. If AI is primarily a model problem, that story holds. But in many enterprise domains, AI is a systems problem—integrations, permissions, evaluation, and change management—where advantage accrues to whomever already sits inside high-volume, high-stakes operations and converts that position into learning and automation.
The inversion: AI executes, humans adjudicate
Traditional services organizations are built on a simple architecture: humans use software to do expert work. Operators log into systems, navigate operations, make decisions, and process cases. Technology is the medium. Human judgment is the product.
An AI-native platform inverts this. It ingests a problem, applies accumulated domain knowledge, executes autonomously what it can with high confidence, and routes targeted sub-tasks to human experts when the situation demands judgment that the system can’t yet reliably provide.
But inverting human-AI interaction isn’t just a UI redesign—it requires raw material. It’s only possible when the platform is built on a foundation of domain expertise, behavioral data, and operational knowledge accumulated over years.
The three compounding assets incumbents already own
AI-native startups begin with a clean architectural slate and can move quickly. What they can’t easily manufacture is the raw material that makes domain AI defensible at scale:
- Proprietary operational data
- A large workforce of domain experts whose day-to-day decisions generate training signals
- Accumulated tacit knowledge about how complex work actually gets done
Services companies already have all three. But these ingredients aren’t moats on their own. They become an advantage only when a company can systematically convert messy operations into AI-ready signals and institutional knowledge—then feed the results back into operations so the system keeps improving.
Codifying expertise into reusable signals
In most services organizations, expertise is tacit and perishable. The best operators know things they cannot easily articulate: heuristics developed over the years, edge-case intuitions, and pattern recognition that operate below the level of conscious reasoning.
At Ensemble, the strategy for addressing this challenge is knowledge distillation. The systematic conversion of expert judgment and operational decisions into machine-readable training signals.
In health-care revenue cycle management, for example, systems can be seeded with explicit domain knowledge and then deepen their coverage through structured daily interaction with operators. In Ensemble’s implementation, the system identifies gaps, formulates targeted questions, and cross-checks answers across multiple experts to capture both consensus and edge-case nuance. It then synthesizes these inputs into a living knowledge base that reflects the situational reasoning behind expert-level performance.
Turning decisions into a learning flywheel
Once a system is constrained enough to be trusted, the next question is how it gets better without waiting for annual model upgrades. Every time a skilled operator makes a decision, they generate more than a completed task. They generate a potential labeled example—context paired with an expert action (and sometimes an outcome). At scale, across thousands of operators and millions of decisions, that stream can power supervised learning, evaluation, and targeted forms of reinforcement—teaching systems to behave more like experts in real conditions.
For example, if an organization processes 50,000 cases a week and captures just three high-quality decision points per case, that’s 150,000 labeled examples every week without creating a separate data-collection program.
A more advanced human-in-the-loop design places experts inside the decision process, so systems learn not just what the right answer was, but how ambiguity gets resolved. Practically, humans intervene at branch points—selecting from AI-generated options, correcting assumptions, and redirecting operations. Each intervention becomes a high-value training signal. When the platform detects an edge case or a deviation from the expected process, it can prompt for a brief, structured rationale, capturing decision factors without requiring lengthy free-form reasoning logs.
Building toward expertise amplification
The goal is to permanently embed the accumulated expertise of thousands of domain experts—their knowledge, decisions, and reasoning—into an AI platform that amplifies what every operator can accomplish. Done well, this produces a quality of execution that neither humans nor AI achieve independently: higher consistency, improved throughput, and measurable operational gains. Operators can focus on more consequential work, supported by an AI that has already completed the analytical groundwork across thousands of analogous prior cases.
The broader implication for enterprise leaders is straightforward. Advantages in AI won’t be determined by access to general-purpose models alone. It will come from an organization’s ability to capture, refine, and compound what it knows, its data, decisions, and operational judgment, while building the controls required for high-stakes environments. As AI shifts from experimentation to infrastructure, the most durable edge may belong to the companies that understand the work well enough to instrument it and can turn that understanding into systems that improve with use.
This content was produced by Ensemble. It was not written by MIT Technology Review’s editorial staff.
<![CDATA[Faster aging may be linked to schizophrenia, according to new research.]]>
Making AI operational in constrained public sector environments
The AI boom has hit across industries, and public sector organizations are facing pressure to accelerate adoption. At the same time, government institutions face distinct constraints around security, governance, and operations that set them apart from their business counterparts. For this reason, purpose-built small language models (SLMs) offer a promising path to operationalize AI in these environments.
A Capgemini study found that 79 percent of public sector executives globally are wary about AI’s data security, an understandable figure given the heightened sensitivity of government data and the legal obligations surrounding its use. As Han Xiao, vice president of AI at Elastic, says, “Government agencies must be very restricted about what kind of data they send to the network. This sets a lot of boundaries on how they think about and manage their data.”
The fundamental need for control over sensitive information is one of many factors complicating AI deployment, particularly when compared against the private sector’s standard operational assumptions.
Unique operational challenges
When private-sector entities expand AI, they typically assume certain conditions will be in place, including continuous connectivity to the cloud, reliance on centralized infrastructure, acceptance of incomplete model transparency, and limited restrictions on data movement. For many state institutions, however, accepting these conditions could be anything from dangerous to impossible.
Government agencies must ensure that their data stays under their control, that information can be checked and verified, and that operational disruptions are kept to an absolute minimum. At the same time, they often have to run their systems in environments where internet connectivity is limited, unreliable, or unavailable. These complexities prevent many promising public sector AI pilots from moving beyond experimentation. “Many people undervalue the operating challenge of AI,” Xiao says. “The public sector needs AI to perform reliably on all kinds of data, and then to be able to grow without breaking. Continuity of operations is often underestimated.” An Elastic survey of public sector leaders found that 65 percent struggle to use data continuously in real time and at scale.
Infrastructure constraints compound the problem. Government organizations may also struggle to obtain the graphics processing units (GPUs) used to train and access complex AI models. As Xiao points out, “Government doesn’t often purchase GPUs, unlike the private sector—they’re not used to managing GPU infrastructure. So accessing a GPU to run the model is a bottleneck for much of the public sector.”
A smaller, more practical model
The many nonnegotiable requirements in the public sector make large language models (LLMs) untenable. But SLMs can be housed locally, offering greater security and control. SLMs are specialized AI models that typically use billions rather than hundreds of billions of parameters, making them far less computationally demanding than the largest LLMs.
The public sector does not need to build ever-larger models housed in offsite, centralized locations. An empirical study found that SLMs performed as well or better than LLMs. SLMs allow sensitive information to be used effectively and efficiently while avoiding the operational complexity of maintaining large models. Xiao puts it this way: “It is easy to use ChatGPT to do proofreading. It’s very difficult to run your own large language models just as smoothly in an environment with no network access.”
SLMs are purpose-built for the needs of the department or agency that will use them. The data is stored securely outside the model, and is only accessed when queried. Carefully engineered prompts ensure that only the most relevant information is retrieved, providing more accurate responses. Using methods such as smart retrieval, vector search, and verifiable source grounding, AI systems can be built that cater to public sector needs.
Thus, the next phase of AI adoption in the public sector may be to bring the AI tool to the data, rather than sending the data out into the cloud. Gartner predicts that by 2027, small, specialized AI models will be used three times more than LLMs.
Superior search capabilities
“When people in the public sector hear AI, they probably think about ChatGPT. But we can be much more ambitious,” says Xiao. “AI can revolutionize how the government searches and manages the large amounts of data they have.”
Looking beyond chatbots reveals one of AI’s most immediate opportunities: dramatically improved search. Like many organizations, the public sector has mountains of unstructured data—including technical reports, procurement documents, minutes, and invoices. Today’s AI, however, can deliver results sourced from mixed media, like readable PDFs, scans, images, spreadsheets, and recordings, and in multiple languages. All of this can be indexed by SLM-powered systems to provide tailored responses and to draft complex texts in any language, while ensuring outputs are legally compliant. “The public sector has a lot of data, and they don’t always know how to use this data. They don’t know what the possibilities are,” says Xiao.
Even more powerful, AI can help government employees interpret the data they access. “Today’s AI can provide you with a completely new view of how to harness that data,” says Xiao. A well-trained SLM can interpret legal norms, extract insights from public consultations, support data-driven executive decision-making, and improve public access to services and administrative information. This can contribute to dramatic improvements in how the public sector conducts its operations.
The small-language promise
Focusing on SLMs shifts the conversation from how comprehensive the model can be to how efficient it is. LLMs incur significant performance and computational costs and require specialized hardware that many public entities cannot afford. Despite requiring some capital expenses, SLMs are less resource-intensive than LLMs, so they tend to be cheaper and reduce environmental impact.
Public sector agencies often face stringent audit requirements, and SLM algorithms can be documented and certified as transparent. Some countries, particularly in Europe, also have privacy regulations such as GDPR that SLMs can be designed to meet.
Tailored training data produces more targeted results, reducing errors, bias, and hallucinations that AI is prone to. As Xiao puts it, “Large language models generate text based on what they were trained on, so there is a cut-off date when they were trained. If you ask about anything after that, it will hallucinate. We can solve this by forcing the model to work from verified sources.”
Risks are also minimized by keeping data on local servers, or even on a specific device. This isn’t about isolation but about strategic autonomy to enable trust, resilience, and relevance.
By prioritizing task-specific models designed for environments that process data locally, and by continuously monitoring performance and impact, public sector organizations can build lasting AI capabilities that support real-world decisions. “Do not start with a chatbot; start with search,” Xiao advises. “Much of what we think of as AI intelligence is really about finding the right information.”
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. This includes the writing of surveys and collection of data for surveys. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
The Download: cyberscammers’ banking bypasses, and carbon removal troubles
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.
Cyberscammers are bypassing banks’ security with illicit tools sold on Telegram
Inside a money-laundering center in Cambodia, an employee opens a banking app on his phone. It asks for a photo linked to the account, so he uploads a picture of a 30-something Asian man.
The app then requests a video “liveness” check. The scammer holds up a static image of a woman who doesn’t match the account. After 90 seconds, he’s in.
The exploit relies on illicit hacking services sold on Telegram that break “Know Your Customer” (KYC) facial scans. MIT Technology Review found 22 channels and groups advertising these services. This is what we discovered.
—Fiona Kelliher
Is carbon removal in trouble?
—Casey Crownhart
Last week, news emerged that Microsoft was pausing carbon removal purchases. It was a bombshell—Microsoft effectively is the carbon removal market, single-handedly purchasing around 80% of all contracted carbon removal.
The report sparked fear across the industry, raising questions about the future of carbon removal and the role of Big Tech. Read the full story.
This story is from The Spark, our weekly newsletter exploring the technology that could combat the climate crisis. Sign up to receive it in your inbox every Wednesday.
The quest to measure our relationship with nature
—Emma Marris
Humans have done some destructive things to the ecosystems around us. But conservationists are learning that we can also be a force for good.
To understand how we work best with nature, a group of scientists, authors, and philosophers have developed new measurements of human-nonhuman relationships. Now, a team in the United Nations is continuing the work. Find out why—and what they hope to achieve.
This story is from the next issue of our print magazine, which is all about nature. Subscribe now to read it when it lands on Wednesday, April 22.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Ukraine says Russian troops have surrendered to robots
They claim a fully automated attack captured army positions for the first time in history. (404 Media)
+ Europe’s vision for future wars is full of drones. (MIT Technology Review)
2 Monkeys with BCIs are navigating virtual worlds using only their thoughts
The research could help people with paralysis. (New Scientist)
+ But these implants still face a critical test. (MIT Technology Review)
3 NASA wants to put nuclear reactors on the Moon
They could power lunar bases and extend spaceflight. (Wired $)
+ NASA is also building a nuclear-powered spacecraft. (MIT Technology Review)
4 Plans for online age verification in the US are raising red flags
Experts warn of compliance issues and potential data breaches. (NBC News)
+ In the EU, an age verification app is about to launch. (Reuters $)
5 An AI chip boom just pushed Taiwan’s stock market past the UK’s
It’s risen past $4 trillion to become the world’s seventh largest. (FT $)
+ Future AI chips could be built on glass. (MIT Technology Review)
6 The public backlash against data centers is intensifying in the US
Protests and litigation are blocking projects. (CNBC)
+ One potential solution? Putting them in space. (MIT Technology Review)
7 Five-minute EV charging is becoming a reality
China’s BYD has started rolling it out. (Gizmodo)
+ “Extended-range electric vehicles” are about to hit US streets. (Atlantic $)
8 Stealth signals are bypassing Iran’s internet blackout
Files hidden in satellite TV broadcasts keep information flowing. (IEEE)
9 Shoe brand Allbirds made a shock pivot to AI, sending stock up 700%
No bubble to see here, folks. (CNBC)
+ What even is the AI bubble? (MIT Technology Review)
10 The largest ever map of the universe is complete
It captures 47 million galaxies and quasars. (Space.com)
Quote of the day
“I like the internet as much as anybody, but we’ve got to go on an internet diet. We don’t need to pay for corporations to do their internet stuff.”
—Sylvia Whitt, a 78-year-old retiree based in Virginia, tells the Washington Post why they’re protesting against data centers.
One More Thing
ISRAEL VARGAS
AI and the future of sex
Some Republican lawmakers want to criminalize porn and arrest its creators. But what if porn is wholly created by an algorithm? In that case, whether it’s obscene, ethical, or safe becomes a secondary issue. The primary concern will be what it means for porn to be “real”—and what the answer demands from all of us.
Technological advances could even remove the “messy humanity” from sex itself. The rise of AI-generated porn may be a symptom of a new synthetic sexuality, not the cause. Read the full story.
—Leo Herrera
We can still have nice things
A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line.)
+ An animator turned his son’s drawings into epic anime characters.
+ Hundreds of baby green sea turtles made a spectacular first journey to the ocean.
+ You can now track rocket launches from take-off to orbit in real time.
+ These musical mistakes prove that even the classics aren’t perfect.
Spatial evolution in temporal dynamics of hemodynamic response function in human superior colliculi with ultra-high-resolution MRI at 9.4T
The superior colliculus (SC) plays a crucial role in multisensory integration, visual information processing, saccadic target selection, visual selective attention, and decision making. In particular, the SC has a key role in oculomotor coordination, following a rostro-caudal organization. The rostral SC, which corresponds to foveal representation, is linked to fixation, microsaccades, smooth pursuit, and vergence adjustments. In contrast, the caudal SC, representing more peripheral visual field, is associated with the large gaze shifts (saccades). However, evidence regarding whether this functional gradient is preserved in the human SC remains limited. In this study, we employed a sequence-following visual-motor task to specifically engage SC activity. We measured blood oxygenation level dependent (BOLD) functional magnetic resonance imaging (fMRI) responses to brief neural activity, known as hemodynamic response function (HRF). We showed a spatial gradient of the BOLD positive HRFs (pHRF) along the rostro-caudal axis of the SC. The pHRF was primarily located in the rostral SC, and it gradually weakened toward the caudal SC, where negative HRF (nHRF) was often observed. The systematic rostro-caudal evolution of HRFs were consistent both within and across subjects, consistent with results from previous electrophysiological studies. Our work showed the feasibility of using ultra-high-field fMRI to non-invasively examine neurovascular dynamics in a small and deeply located subcortical structures of the human brain.
MambaSSM: efficient segmentation of brain structures in anisotropic 3D EM images via state-space models
Accurate segmentation of brain structures from anisotropic 3D electron microscopy (EM) images remains challenging due to the trade-off between global context modeling and computational efficiency. While state-space models (SSMs) like Mamba have shown promise in capturing long-range dependencies, their direct application to anisotropic EM data has been limited. We introduce MambaSSM, a novel network that adapts SSMs to anisotropic 3D EM images via a tailored scanning strategy. Our method features two core modules: an SSM-based anisotropic adaptation module for early-stage feature learning and an SSM-based isotropic adaptation module for later-stage refinement. These modules are interleaved with convolutional layers to enable multi-scale feature extraction. Evaluated on two public datasets (SNEMI3D and MitoEM-R), MambaSSM achieves superior segmentation accuracy with significantly lower memory usage compared to CNN, Transformer, and Mamba based baselines.
Role of TRPC1 in the pathogenesis of depression induced by traumatic brain injury
BackgroundTraumatic brain injury (TBI) is one of the leading causes of mortality and disability, with many patients developing long-term sequelae. Depression is among the most common psychiatric complications following TBI, yet its underlying mechanisms remain unclear. Transient receptor potential canonical 1 (TRPC1) has been implicated in neurological disorders, but its role in post-TBI depression is not well understood.MethodsA controlled cortical impact (CCI) model was used to induce moderate TBI in mice. At 4 weeks post-injury, depressive-like behaviors were assessed using the tail suspension test (TST), forced swim test (FST), and sucrose preference test (SPT). Subsequently, reactive astrocytes and microglia were quantified, along with the expression of inflammatory cytokines, in the ipsilateral hippocampus. Synaptic function was also evaluated.ResultsBehavioral tests revealed that TBI mice exhibited significant depressive- and anxiety-like behaviors at 4 weeks post-injury. Concurrently, TRPC1 expression was downregulated in the ipsilateral hippocampus, accompanied by reduced levels of synaptic-associated proteins, elevated pro-inflammatory cytokines, and increased reactive astrocytes and microglia. Further experiments demonstrated that TRPC1 overexpression attenuated neuroinflammation, restored synaptic function, and ameliorated depressive-like behaviors in TBI mice.ConclusionThis study suggests that TBI may trigger depression by downregulating TRPC1, thereby promoting neuroinflammation and synaptic dysfunction. Conversely, TRPC1 overexpression mitigates these effects, highlighting its potential as a therapeutic target for post-TBI depression.
Diffusion tensor imaging-functional MRI fusion reveals disrupted white matter structure–function coupling in HIV-associated asymptomatic neurocognitive impairment
ObjectiveConventionally, blood oxygen level-dependent (BOLD) signals derived from resting-state functional magnetic resonance imaging (rs-fMRI) are attributed to gray matter, but recent evidence confirms stable low-frequency oscillations within white matter. While structure–function coupling is pivotal in neuropsychiatry, it remains underexplored in HIV-associated neurocognitive disorders (HAND). Focusing on Asymptomatic Neurocognitive Impairment (ANI), the earliest stage of HAND, this study establishes a white matter skeleton-based fusion framework integrating diffusion tensor imaging (DTI) and rs-fMRI to investigate underlying mechanisms.MethodsWe enrolled 47 patients with ANI and 48 matched healthy controls. Fractional anisotropy (FA) images from DTI and BOLD signals derived from rs-fMRI were projected onto a unified white matter skeleton to achieve structure–function spatial alignment. FA, skeleton-based white matter amplitude of low-frequency fluctuations (SWALFF), and its dynamic variability (dSWALFF) were calculated. Group differences in white matter structure and function were assessed, with structure–function coupling examined in regions showing overlapping FA-SWALFF and FA-dSWALFF alterations. Additionally, a novel White Matter Dys-coupling Index (WDI) was proposed to quantify the deviation between structural integrity and functional activity and evaluate its clinical relevance.ResultsCompared to controls, ANI patients exhibited widespread FA reductions and increased mean diffusivity (MD) and radial diffusivity (RD), indicating diffuse demyelination. Functionally, a spatial dissociation emerged: SWALFF was reduced in posterior occipital pathways (left vertical occipital fasciculus, forceps major), whereas SWALFF and dSWALFF were elevated in prefrontal pathways (forceps minor). Overlapping regions revealed complex coupling patterns, ranging from concordant decline to compensatory upregulation and decoupling. The interaction between FA and dSWALFF further highlighted instability in dynamic regulation. The WDI was significantly correlated with infection duration, immune status, and cognitive domain scores.ConclusionThis study identifies a characteristic “coupling imbalance” in the white matter of ANI patients, defined by the coexistence of structural degeneration and functional reorganization. We propose the WDI as a quantitative metric for this deviation. Its significant associations with clinical and cognitive metrics suggest its potential as a neuroimaging biomarker for the early identification and mechanistic understanding of HAND.
The relationship between impulsivity and non-suicidal self-injury in adolescents: the chain-mediated effects of parenting style and distress tolerance
ObjectiveThe purpose of this study is to explore the related risk factors and protective factors of adolescent non suicidal self injury (NSSI).MethodsUtilizing the experience sampling method, we recruited 311 adolescents engaging in NSSI, all without other mental disorders, from five public high schools in a specific city. Questionnaire surveys were administered, employing the Chinese version of the Short-Form Egna Minnen av Barndoms Uppfostran (s-EMBU-C), Distress Tolerance Scale-Revised (DTS-CR), Adolescent Self-harm Behavior Questionnaire, Barratt Impulsiveness Scale version 11 (BIS-11).Results1) The findings indicate that NSSI in adolescents is positively correlated with impulsivity and negative parenting styles (P < 0.01), while it is negatively correlated with distress tolerance and positive parenting styles (P<0.01). Impulsivity is negatively correlated with distress tolerance and positive parenting styles (P < 0.01) and positively correlated with negative parenting styles (P < 0.01). Furthermore, distress tolerance is negatively correlated with negative parenting styles (P < 0.01) and positively correlated with positive parenting styles (P < 0.01). 2) This study reveals that both negative and positive parenting styles serve as complete mediators in the relationship between impulsivity and NSSI behavior in adolescents, with distress tolerance as a significant factor.ConclusionImpulsivity significantly influences NSSI behavior in adolescents through the mediation of parenting styles (both negative and positive) and distress tolerance.