The researchers examined straight-line distances and expected travel times between randomly selected neighbourhoods and eight forms of cultural amenity—libraries, museums, elite universities, Whole Foods supermarkets, Equinox gyms, fine performing-arts venues, independent cinemas, and art galleries—across the dozen largest metropolitan areas in the United States.

Clear divide

The results are stark. Communities with high levels of educational attainment sit systematically closer to all eight types of cultural institution, regardless of how distance is measured. High-income neighbourhoods also enjoy markedly shorter travel times to these amenities than their low-income counterparts.

The authors argue that such geographic disparities may help explain why lower-income households make less use of the cultural resources that foster upward mobility, from the acquisition of knowledge to the development of social and professional networks.

They propose several remedies. Elite universities—typically located in dense clusters of cultural institutions—should broaden access to their campuses. Affordable housing could be more evenly distributed across metropolitan areas to reduce spatial segregation. And schools, they say, should introduce children to cultural capital early, teaching not only the value of such exposure but its role in shaping long-term economic opportunity.

To reduce this dependence, the U.S. Geological Survey (USGS), in partnership with the Defense Advanced Research Projects Agency (DARPA) and the Advanced Research Projects Agency–Energy (ARPA-E), has launched a new effort: Critical Mineral Assessments with AI Support, or CriticalMAAS. The project aims to harness artificial intelligence to speed up the discovery and analysis of domestic mineral resources. It brings together researchers from the University of Southern California’s Viterbi Information Sciences Institute (ISI) and the University of Minnesota.

Digging for data

The first obstacle is a familiar one: data. The USGS holds more than 100,000 geological maps—many of them decades old and available only as scanned images. These maps contain invaluable details of rock formations, fault lines and mineral deposits, but their raster format renders them largely useless for computational analysis. Converting them into digital, machine-readable form has long been a slow, painstaking task.

Enter DIGMAPPER, a modular AI system designed to automate the process. Where human digitisation can take hours per map, DIGMAPPER completes the job in under 25 minutes. Tested on more than 100 annotated maps from a DARPA–USGS dataset, it proved both fast and accurate, turning a once-manual bottleneck into a scalable workflow.

A second ISI-led effort, known as MinMod, tackles a different challenge: unifying the world’s fragmented mineral data. Mining information is scattered across academic papers, company reports, and geological databases, each with its own jargon and structure. MinMod brings this chaos into order, merging disparate sources into a single AI-ready “knowledge graph”.

Deep mines

So far, the system has integrated tens of thousands of documents covering more than 679,000 mining sites and 190 commodities. By using large language models and semantic technologies, MinMod extracts and organises data into a consistent framework that researchers and policymakers can search, analyse, and expand upon.

Among its most valuable features is the ability to generate “grade and tonnage” models—interactive charts showing which deposit types yield the most metal. Such insights help geologists prioritise exploration targets and guide national resource strategy.

If CriticalMAAS succeeds, it could mark a turning point. What was once an exercise in painstaking fieldwork and archival research may soon be powered by algorithms—helping America map, mine, and manage the materials that will define its economic and strategic future.

As large language models (LLMs) proliferate, their use in political messaging seems inevitable—if it has not begun already. The Stanford researchers wanted to know whether such AI-crafted arguments could be as convincing as those penned by humans. The answer, across a range of policy issues, was yes. Participants who read AI-generated arguments were swayed just as much as those who read human-written ones, even though none were told who—or what—had authored them.

Shooting the messenger

In a follow-up experiment, a different Stanford team asked a different question: how does knowing that a message comes from AI change the way people respond? In other words, the first study tested the message; the second tested the messenger.

In the first set of surveys, participants were asked about their views on policies such as smoking bans, gun control, carbon taxes and automatic voter registration. They were then shown arguments in favour of these policies, written either by people or by AI. A control group read unrelated material. Afterwards, researchers measured whether opinions had shifted. The results showed similar persuasion effects across both sources. Interestingly, the biggest changes occurred among participants who already supported the policies—the AI-authored messages appeared to reinforce existing convictions rather than convert new believers.

Participants attributed human-written appeals’ persuasiveness to storytelling and personal anecdotes. AI-written ones, by contrast, were valued for their logic and clarity.

More or less persuasive?

The second study added a twist. This time, participants were told whether an argument—crafted by the researchers—came from a person or from AI. Each message contradicted the reader’s prior beliefs: vaccine supporters read anti-vaccine arguments, for instance. Across several experiments, people proved more open-minded when they believed opposing arguments came from AI. They were more receptive to counterarguments, more willing to share them, and less hostile to those on the other side.

Humans, it seems, suspect other humans of bias and ulterior motives. “When a person tries to convince us, we assume they’re uninformed or simply trying to win an argument,” the researchers note. Machines, by contrast, are viewed as knowledgeable and objective—an impression that makes their words easier to hear.

Such findings hint at practical uses. If social-media platforms or civic institutions wish to expose users to a broader range of viewpoints, attributing information to AI might make those viewpoints less threatening and more likely to be considered in good faith. Used wisely, the same technology that floods public discourse with misinformation could, paradoxically, help bridge the political divide.

A new book by Barbara Plester, an associate professor at the University of Auckland Business School, examines how communication, humour, and happiness are being reshaped in the age of hybrid work. Hybrid Happiness: Fun and Freedom in Flexible Work explores the social and emotional consequences of dividing time between home and office, and how these shifts are altering the fabric of workplace life.

What makes us happy

Drawing on immersive fieldwork in two organisations—a technology firm dubbed “Gecko” and a food manufacturer called “Firefly”—Dr Plester observed employees across all levels, from CEOs to recent hires. Over four weeks embedded in each company, she watched, listened, and interviewed, taking a grounded-theory approach that let themes emerge naturally rather than testing a preconceived hypothesis. “My research on fun and humour evolved into a study of happiness,” she notes, “because participants consistently linked these ideas together.”

The book captures both the freedoms and frictions of flexible work. A chapter titled “The Emotional Landscape of Hybrid Work” examines the strain that digital communication imposes on emotional understanding. Reading nuance through chat messages, emojis and GIFs, Dr Plester argues, has become a new workplace skill—and a potential minefield. “Emotions are increasingly mediated through technology,” she writes. “Pictorial language can enliven conversation and convey playfulness, but it can just as easily be misinterpreted.”

Another chapter, “Tech-Powered Freedom,” delves into questions of trust, surveillance, and the erosion of boundaries between home and office. Trust, Plester argues, is the cornerstone of successful hybrid arrangements. When employees feel trusted, they tend to perform better and with greater engagement. One worker at Firefly put it bluntly: “If you don’t trust your employee, why employ them?”

Encouraging safety

In “Psychological Safety in Hybrid Fun,” Plester considers the paradox of “forced fun”—activities designed to build camaraderie but which can leave employees feeling coerced or infantilised. Her research suggests that contrived cheerfulness may foster cynicism rather than connection. True enjoyment, she writes, depends on the freedom to participate—or not—without fear of judgment.

The book concludes that happiness in hybrid work cannot be manufactured through corporate enthusiasm alone. It flourishes instead in cultures that prize trust, empathy, and flexibility—qualities that, in an era of emojis and blurred boundaries, matter more than ever.

Recruitment has seldom been an enjoyable affair, and while it’s hard to argue that AI is making it better, you sense that it is a factor in the rise of AI in recruitment. This desire to remove ourselves from the stress and tedium of submitting or wading through application after application.

Dealing with the biases

Of course, it’s increasingly evident that the systems we’re using to do all of this also come with their fair share of biases. What’s more, we’re also not very good at spotting biased outputs when they’re presented to us by AI. So, given the important role recruitment plays, the reliance on these tools comes with a large degree of risk.

Research from the University of Washington found that when we use AI to help us select candidates, we’re more likely to make biased choices than if we make decisions on our own. The researchers presented over 500 volunteers with identical resumes, but some were presented as being from white candidates, some from Black candidates, some from Hispanic candidates, and some from Asian candidates (all were from male candidates).

The results showed that the volunteers strongly followed the advice given to them by the AI, regardless of whether it was biased or not. Indeed, even when the AI was extremely biased, people didn’t shift their recommendations a great deal.

“In one survey, 80% of organizations using AI hiring tools said they don’t reject applicants without human review,” the researchers explain. “So this human-AI interaction is the dominant model right now. Our goal was to take a critical look at this model and see how human reviewers’ decisions are being affected. Our findings were stark: Unless bias is obvious, people were perfectly willing to accept the AI’s biases.”

Fair recruitment

An obvious first step to ensure that bias doesn’t pervade AI-driven recruitment is to ensure that the models and systems themselves aren’t biased. Indeed, the study shows that when AI made unbiased recommendations, this typically resulted in humans making unbiased decisions.

There are also things that we can do that don’t rely on the tech companies to change. For instance, implicit bias training can be effective as it reminds us that the technology can make discriminatory decisions. This was underlined by research by the Haas School of Business, which suggested a good approach is for companies to foster a culture of bias awareness in AI, which allows individual employees to report biased outcomes and support them in addressing these issues with their supervisors.

Ultimately, a fairer application of AI in recruitment is likely to require a collective effort from recruiters, developers, and policymakers. We can’t rely on recruiters alone to fix this issue, nor indeed on developers to do so on our behalf. Indeed, research from University College London shows that while some developers take these societal externalities seriously, many others view downstream problems as someone else’s to fix.

Moving forward

We’ve seen in numerous other applications of AI that it can be useful when it helps us to think, and very unhelpful when it does our thinking for us. Recruitment is no different, but these studies remind us that AI can also lead us astray when it serves us biased recommendations.

As such, it’s important that we better understand that AI decisions mirror the data used to train it, with all of the biases that come with that. Awareness is a key part of being able to work not only more effectively with AI but also more fairly with it. We shouldn’t use it to shortcut our own thought processes, but rather as one input among many. Embedding bias awareness training, such as implicit association testing, into hiring processes can help ensure that human judgment remains critical and reflective.

We should also be much more demanding of the companies making these systems to ensure that not only are datasets tested for bias during development, but also that outputs are continuously monitored for biases throughout operations. Just as “bug hunting” is a common part of traditional software development, so too should “bias hunting” be a part of modern AI tools. This would require implementing clear and easy-to-use means for recruiters to report when recommendations are biased.

Meanwhile, the onus also lies with those creating and deploying these systems. Developers must actively test for and mitigate bias during model training, and organizations should insist on transparency around how recommendations are generated. Regulation will likely play a growing role too, helping ensure these systems align with broader societal and ethical expectations.

There’s clearly something a bit broken with recruitment at the moment, with no party particularly happy with the status quo. AI can undoubtedly play a role in making things better. It might even help us to make recruitment fairer. It won’t happen by osmosis, however. It requires us to design, deploy, and critique it at every step so that humans are always in the driving seat.

Synthesising four decades of global studies, the researchers find that the relationship between urban greenness and mental health follows an inverted-U pattern: psychological well-being improves as greenery increases—but only up to a point. Beyond a moderate threshold, benefits plateau and may even reverse. The study, published after analysing 87,000 records across five continents, offers the most comprehensive evidence yet that moderation, not maximisation, is key.

The right dose

The optimal dose, the authors find, is remarkably precise. For greenery seen at eye level, the streets, parks, and plazas people actually move through, mental health peaks when roughly half (around 53%) of the visual field is green. Top-down views, such as canopy coverage from satellite data, show similar results. Below about a quarter or above four-fifths, the psychological pay-off declines.

The logic is intuitive. A city that is too barren is stressful; one that is too lush can feel isolating, poorly lit, or impractical. The findings echo broader theories—from the Yerkes–Dodson law to hormesis—showing that moderate stimuli often elicit optimal human responses.

For policymakers, the message is pragmatic. Cities need not carpet every available surface in foliage. Instead, they should design “just enough” green—prioritising street-level trees, pocket parks and visible vegetation where people actually experience them. Over-greening can divert scarce land from housing, services, and infrastructure—an especially acute concern in densely packed Hong Kong, where verdant hills rise above cramped flats.

The study’s broader contribution is theoretical as well as practical. By defining a generalised “dose–response curve” between greenness and mental health, it may lay the groundwork for an emerging law of environmental psychology: that balance, not abundance, best nurtures well-being.

The analysis, published in the European Sociological Review, draws on the German Family Panel between 2008 and 2021. It finds that women still shoulder around 70% of housework and childcare, despite remote work becoming far more common, rising from one in ten parents before the pandemic to nearly half of fathers and a third of mothers by 2021. The shift, in other words, has expanded the potential for fairness but not guaranteed it.

Uneven share

When mothers work from home, the researchers found, they tend to do even more domestic labour, particularly when they or their partners hold conventional views about men’s household responsibilities. Only in households where fathers regard care as a shared duty does remote work appear to level the load.

The authors argue that encouraging men to work from home could help close the gap, but warn that structural reforms alone will fall short unless social attitudes change too. Policies such as flexible working rights and well-paid, ringfenced paternity leave in a child’s early years can help reset expectations about who cares. Many fathers, they note, are keen to do more; the question is whether governments and employers will enable them.

“Structural policies such as the right to flexible work must be accompanied by efforts to challenge gender stereotypes,” the authors conclude, “beginning with perceptions of men’s roles in family life.” Remote work, it seems, is only as equalising as the beliefs that govern the home.

Research from the Olin Business School at Washington University in St. Louis suggests that entrepreneurial zeal pays off, but only when it delivers results. Analysing data from around 650 employees at Fortune 500 companies, the researchers found that workers who exhibited entrepreneurial behaviour were significantly more likely to launch internal ventures, such as new products or improved processes. A one-point rise in entrepreneurial activity increased the likelihood of doing so by 50%.

Promotion chances

Those who successfully created such ventures were, in turn, far more likely to be promoted and, particularly in large firms, to receive higher pay. Merely having ideas was not enough; transforming them into tangible outcomes was what earned career dividends. As the authors note, “entrepreneurial behaviour alone did not guarantee success—execution did.”

Size also mattered. Smaller firms tended to reward innovators with promotions rather than pay rises, perhaps reflecting tighter budgets. In larger organisations, where managers are more distant from employees, creating an internal venture was a strong signal of leadership potential. In a follow-up experiment, participants reviewing résumés for a promotion were more likely to select candidates who had built internal ventures, viewing them as future leaders.

For ambitious employees, the lesson is clear: innovation must be visible to be valuable. Turning an idea into something concrete, however modest, can boost both credibility and compensation.

For firms, fostering such behaviour requires more than hackathons and suggestion boxes. Training helps, but so does giving employees ownership of their ideas, from conception to implementation. Cross-functional collaboration—where marketing meets engineering or logistics meets design—often sparks the most productive innovation.

Managers, however, must learn to see entrepreneurial employees not as troublemakers but as catalysts. Human-resources departments can help by reframing risk-taking and even failure as part of learning, not grounds for reprimand. Companies that manage this cultural shift will not only keep their innovators but reward them, and themselves, in the process.

Usage follows trust downhill. Between February and July, employees’ use of sanctioned AI tools dropped by 15%. Yet nearly half of frontline workers with access to corporate AI say they use unofficial alternatives instead. The problem, then, is not antipathy to AI per se, but scepticism toward the tools their employers prescribe. Many workers feel these systems are being foisted upon them—imposed, not adopted—and, worse, that they are being asked to help train the very technologies that might replace them.

Building trust

With so much money riding on the successful integration of AI, companies are scrambling to fix the trust deficit. Deloitte highlights five approaches that firms—its clients and others—are using to restore confidence.

Start by measuring mistrust. Deloitte’s own index parses trust into four dimensions—reliability, capability, transparency and “humanity”—allowing managers to isolate what is eroding confidence. When trust rises, so do positive behaviours: high-trust employees are almost twice as likely to recommend their employer and more willing to acquire new skills. If staff balk at AI’s opacity, greater communication or employee involvement can steady nerves.

Invest in skills, not redundancies. Fears of automation replacing humans remain acute, but labour shortages in sectors such as logistics and healthcare mean that few firms can simply “hire their way” into an AI-ready workforce. Companies that use AI to redesign jobs rather than eliminate them reap better results. Training that blends technical fluency with human skills—judgement, empathy, adaptability—boosts performance when workers collaborate with intelligent systems. Employees who receive hands-on training report trust levels 144% higher than those who do not. Sweden’s IKEA offers a tidy case study: after launching an AI chatbot to handle half of customer inquiries, it reskilled 8,500 call-centre staff into design-advice roles rather than shedding them. In warehouses, drones have taken over laborious stock counts, freeing workers for more valuable tasks. The company plans to train 70,000 employees in AI literacy by 2026. Turnover has dropped, and morale is up.

Design AI with workers, not just for them. Many systems fail because they are delivered as fait accompli. Walmart takes a different tack: its internal AI foundry, Element, develops tools through iterative pilots with store associates. A scheduling app, refined with input from staff, now includes shift swapping and availability preferences—small features that matter when juggling childcare or second jobs. Another tool offers real-time translation across 44 languages, incorporating Walmart-specific vocabulary. Workers trust the tools because they helped build them.

Allow experimentation—safely. If AI adoption requires trial and error, firms must stop punishing the latter. Colgate-Palmolive’s “AI Hub” lets employees create no-code assistants tailored to their jobs. A plant manager in Greece built a system that translates dense German machinery manuals into practical troubleshooting steps; an HR employee created a “goals coach” trained on corporate values. Thousands of such assistants now exist, and the most effective are scaled company-wide.

Empower middle managers. Adoption ultimately depends on the people closest to the front line. Workers trust their direct managers far more than the corporate centre—by roughly 20 percentage points—and weekly check-ins boost trust by nearly 60%. At Intuit, efforts to increase AI use stalled until the company gathered 150 frontline “experimenters” for a hands-on training day led by mid-level managers. Participants returned to their teams as evangelists, turning a corporate initiative into a peer-driven movement.

Across these cases, one lesson stands out: trust is not a soft metric but a hard constraint. AI will deliver on its promise only when workers feel that it enhances, rather than threatens, their work. Companies that understand this may find that AI adoption becomes not a battle to be won, but a momentum that builds itself.

“Immigrants support the core democratic principles to a similarly high degree as people without a migratory background,” the researchers report.

High support

Across 30 European countries, the average level of support for democratic values—such as free elections, equal rights, minority protection, and independent courts—scores 8.56 out of 10 among migrants, compared with 8.48 among non-migrants. In Germany, the figures are almost identical: 2.67 versus 2.66 on a three-point scale. “These very high mean values hardly differ between the individual groups,” the study notes.

A small but statistically significant difference emerges between immigrants from highly authoritarian states—such as Eritrea, Saudi Arabia, and Iran—and those from more democratic countries, including India, Turkey, and Romania. Those accustomed to authoritarian rule tend to express slightly weaker democratic attitudes, while those with longer experience of democratic systems show marginally stronger support. The gap, however, is modest. Broadly, the researchers conclude, democratic convictions are widely shared across cultural and national lines and tend to deepen with exposure to democratic life.

A small minority of immigrants reject democracy altogether, amounting to a mid–single-digit share of the population—virtually identical to the proportion among non-migrants. Anti-democratic sentiment, the study stresses, is not a migration-related phenomenon but a marginal undercurrent present in all social groups.

The findings draw on two large-scale datasets. The tenth wave of the ESS includes tens of thousands of respondents across Europe, among them over 2,000 first-generation migrants. The fourth wave of the Integration Barometer, compiled by Germany’s Expert Council on Integration and Migration (SVR), surveys around 15,000 people, including some 7,000 with a migration background and 4,000 first-generation immigrants. Both surveys assess key indicators of liberal-democratic commitment.