 |
|
|
|||||
The flap of a butterflys wings in South America can famously lead to a tornado in the Caribbean. The so-called butterfly effector sensitive dependence on initial conditions, as it is more technically knownis of profound relevance for organizations seeking to deploy AI solutions. As systems become more and more interconnected by AI capabilities that sit across and reach into an increasing number of critical functions, the risk of cascade failureslocalized glitches that ripple outward into organization-wide disruptionsgrows substantially. It is natural to focus AI risk management efforts on individual systems where distinct risks are easy to identify. A senior executive might ask how much the company stands to lose if the predictive model makes inaccurate predictions. How exposed could we be if the chatbot gives out information it shouldnt? What will happen if the new automated system runs into an edge case it cant handle? These are all important questions. But focusing on these kinds of issues exclusively can provide a false sense of safety. The most dangerous AI failures are not the ones that remain confined to one particular area. They are the ones that spread. How Cascade Failures Work While many AI systems currently operate as isolated nodes, it is only when these become joined up across organizations that artificial intelligence will fully deliver on its promise. Networks of AI agents that communicate across departments; automated ordering systems that link customer service chatbots to logistics hubs, or even to the factory floor; executive decision-support models that draw information from every corner of the organizationthese are the kinds of AI implementations that will deliver transformative value. But they are also the kinds of systems that create the biggest risks. {"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/creator-faisalhoque.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/faisal-hoque.png","eyebrow":"","headline":"Ready to thrive at the intersection of business, technology, and humanity? ","dek":"Faisal Hoques books, podcast, and his companies give leaders the frameworks and platforms to align purpose, people, process, and techturning disruption into meaningful, lasting progress.","subhed":"","description":"","ctaText":"Learn More","ctaUrl":"https:\/\/faisalhoque.com","theme":{"bg":"#02263c","text":"#ffffff","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#ffffff","buttonHoverBg":"#3b3f46","buttonText":"#000000"},"imageDesktopId":91420512,"imageMobileId":91420514,"shareable":false,"slug":""}} Consider how quickly problems can multiply: Corrupted data at a single collection point can poison the outputs of every analytical tool downstream. A security flaw in one model becomes a doorway into every system it touches. And when several AI applications compete for the same computing resources, a spike in demand can choke performance across the boardoften at the worst possible moment. When AI is siloed, failures are contained. When AI is interconnected, failures can propagate in ways that are difficult to predict and even harder to stop. The 2010 Flash Crash in the U.S. stock markets showed how algorithms can interact in unexpected ways, causing problems on a scale that can be hard to imagine. On the morning of May 6th, more than a trillion dollars was wiped off the value of the Dow Jones Industrial Average in a matter of minutes as automated systems triggered a spiral of sell-offs. Despite several years of investigation, the exact cause of the crash is still unknown. What the Flash Crash revealed is that when autonomous systems interact, their combined behavior can diverge dramatically from what any single system was programmed to do. None of the algorithms were designed to crash the market and none of them would have done so if they were operating independently. But the interactions between themeach responding to signals created by othersproduced an unexpected result at the systemic level that was divorced from the goals of any one part of that system. This is the nature of cascade risk. The danger lies not in any individual AI system failing, but in the unpredictable ways that interconnected systems can amplify and spread failures across organizational boundaries. The Hidden Connections Several characteristics make AI systems particularly susceptible to cascading failures. Shared data dependencies create hidden connections between seemingly independent systems. Two AI applications might appear to be completely separate, but if they rely on the same underlying data sources, a corruption or error in that data may affect both simultaneously. And a simultaneous failure may have consequences that are more severe than the sum of the individual failures. These kinds of dependencies and their possible outcomes often go unmapped until a failure forces the organization to take notice. Shared infrastructure creates similar vulnerabilities. Multiple AI systems running on common cloud resources or the same on-site computational infrastructure can all be affected by a single point of failure. During high-demand periods, competition between systems for resources can degrade performance across the board in ways that are difficult to predict or diagnose. Feedback loops between AI systems can amplify small errors into large disruptions. When one systems output feeds into another systems input, and the second systems output then influences the first system, the potential for runaway effects increases. What begins as a minor anomaly can be magnified through successive iterations until it produces significant failures. Integration with critical operations also raises the stakes dramatically. When AI becomes embedded in systems that organizations depend onsupply chains, financial operations, customer service, manufacturingcascade failures dont just create technical problems. They disrupt the core functions that keep the business running. The Organizational Blind Spot Perhaps the greatest challenge in managing cascade risk is organizational rather than technical. The systems that interact to create cascade failures often span different departments, different teams, and different areas of expertise. No single person or group has visibility into all the connections and dependencies. This means that cascade risk management requires cross-functional coordination that cuts against traditional organizational structures. It requires mapping dependencies that cross departmental boundaries. It requires testing failure scenarios that involve multiple systems simultaneously. And it requires governance structures that can make decisions about acceptable risk levels across the organization as a whole, not just within individual units. Organizations that treat AI implementation as a series of independent projectseach managed by its own team, each evaluated on its own meritswill inevitably create the conditions for cascading failures. The connections between systems will emerge organically, without deliberate design or oversight. And when failures occur, they will propagate through pathways that no one fully understood. The alternative is to treat the entire AI ecosystem as an interconnected whole from the beginning. This means thinking about how systems will interact before they are built. It means maintining visibility into dependencies as systems evolve. And it means accepting that the reliability of any individual system is less important than the resilience of the system of systems. Four Ways to Protect Your Organization from AI Cascade Failures 1. Map your AI dependencies before they map themselves. Most organizations discover their system interdependencies only after a failure reveals them. Dont wait. Conduct a systematic audit of how your AI systems connectwhat data they share, what infrastructure they rely on, what outputs feed into other systems inputs. Create a visual map of these dependencies and update it as your AI ecosystem evolves. The goal isnt to eliminate connections (interconnection is often where value comes from) but to understand them well enough to anticipate how failures might propagate. 2. Design circuit breakers into your architecture. Financial markets use automatic trading halts to prevent cascading crashes. Your AI systems need equivalent mechanisms. Build monitoring systems that can detect unusual patternssudden spikes in error rates, unexpected resource consumption, anomalous outputsand automatically pause operations before small problems become large ones. These circuit breakers buy time for human operators to assess situations and intervene. The cost of brief pauses is far less than the cost of cascading failures. 3. Test failure scenarios across system boundaries. Traditional testing evaluates whether individual systems work correctly. Cascade risk requires testing how systems fail together. Run exercises that simulate failures in one system and trace the effects through connected systems. What happens to your customer service AI when your data pipeline delivers corrupted information? How does your inventory system respond when your demand forecasting model produces anomalous predictions? These cross-boundary tests reveal vulnerabilities that single-system testing will never find. 4. Establish cross-functional AI governance. Cascade risks emerge from the gaps between organizational silos. Managing them requires governance structures that span those silosa cross-functional team with visibility into AI implementations across departments and the authority to make decisions about system interactions, acceptable risk levels, and required safeguards. This team should own the dependency map, oversee cross-boundary testing, and ensure that new AI implementations are evaluated not just for their individual merits but for how they affect the broader ecosystem. The butterflys wings are already flapping. The organizations that thrive will be those that see the tornado comingnot by monitoring any single system, but by understanding how all their systems connect. {"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/creator-faisalhoque.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/faisal-hoque.png","eyebrow":"","headline":"Ready to thrive at the intersection of business, technology, and humanity? ","dek":"Faisal Hoques books, podcast, and his companies give leaders the frameworks and platforms to align purpose, people, process, and techturning disruption into meaningful, lasting progress.","subhed":"","description":"","ctaText":"Learn More","ctaUrl":"https:\/\/faisalhoque.com","theme":{"bg":"#02263c","text":"#ffffff","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#ffffff","buttonHoverBg":"#3b3f46","buttonText":"#000000"},"imageDesktopId":91420512,"imageMobileId":91420514,"shareable":false,"slug":""}}
Category:
E-Commerce
The days of Republicans hard stances against marijuana have seemingly gone up in smoke. On Thursday, President Trump signed an executive order to reschedule marijuana as a Schedule III substancean effective downgrade from Schedule I, the most dangerous classification, which includes substances like heroin. The change could allow for marijuana to be used in more medical research, and the order also authorized the creation of a pilot program to reimburse Medicare patients for CBD products. The reclassification does not legalize marijuana, and seemingly completes or finalizes a recommendation made by the Biden administration in 2022 that the drug be rescheduled. The Attorney General shall take all necessary steps to complete the rulemaking process related to rescheduling marijuana to Schedule III of the CSA in the most expeditious manner in accordance with Federal law, the executive order reads. Notably, Trumps reclassification order was signed days after Trump ramped up his administrations stance on another controlled substance, fentanyl. On Monday, he signed another executive order that called fentanyl closer to a chemical weapon than a narcotic, and designated it as a weapon of mass destruction. Cannabis stocks actually ended Thursday in the red, despite the news. Tilray Brands, for example, was down 4.3% for the day, and Canopy Growth was down almost 12%. Members of the cannabis industry are cautiously optimistic about what the executive order actually means, however. The Administrations order calling to remove the cannabis plant from its Schedule I classification validates the experiences of tens of millions of Americans, as well as those of tens of thousands of physicians, who have long recognized that cannabis possesses legitimate medical utility, said Paul Armentano, the Deputy Director at NORML, an organization that advocates for cannabis reform laws, in a statement. This directive certainly marks a long overdue change in direction. Others are much more skeptical of what it means for the legal market, but do think the changes mark a boon for medical cannabis. Cannabis is still federally illegal, said Ryan Hunter, the chief revenue officer at Spherex, which creates vapes and cannabis-infused gummies, in comments shared with Fast Company. The real win here is for medical cannabis, he added. At Schedule III, its much more practical for mainstream physicians to prescribe cannabis products. Still others are simply happy to see reclassification come to fruition. We welcome the decision to reschedule cannabis. This long-overdue step aligns regulation with science and public opinion, providing a necessary foundation for patient relief and compliant business growth, said Socrates Rosenfeld, the CEO and co-founder of Jane Technologies, which creates software for cannabis businesses, in a statement shared with Fast Company. We are hopeful this marks the beginning of real momentum toward the broader, systemic reform needed for a truly just and accessible industry.
Category:
E-Commerce
Dont beat yourself up if you do some serious damage on a cheese plate during holiday festivities this year: You just may do your future self a favor. A new study has found that eating nearly 2 ounces or more of high-fat cheese each day has been associated with a 16% lower risk of dementia, according to the study published this week in Neurology. Lest you think this is some sort of propaganda by Big Cheese, the study followed nearly 28,000 adults in Malmö, Sweden for roughly 25 years. The studys findings indicate that Swedes who ate more cheese with a fat content exceeding 20%which includes many varieties of cheddar, gouda, and blue cheese, among othershad a lower risk of all-cause dementia. The researchers didnt find a similar link with other high-fat dairy products and noted that further confirmation of these findings in diverse populations is warranted. While the amount of cheese in questionequivalent to less than a handful of diced cubesmay not seem significant, scientists are keen to identify even something small that could raise or lower the risk of dementia. More than 6 million Americans are currently living with dementia, and 42% of Americans over the age of 55 could eventually develop such declines in mental abilities, according to figures from the National Institutes of Health. QUESTIONING THE FINDINGS It might be tempting to give yourself permission to go wild on full-fat cheese for your brain, though your waistline could pay the price. The studys authors said their findings require caution in interpretation, something that other experts were quick to do. The researchers only captured the dietary habits of participants at one point in 1991 and didnt follow up with the majority of them over the course of the next 25 years. This sort of approach raises questions about the robustness of the studys conclusions, Dr. Tian-Shin Yeh, a physician and nutritional epidemiologist at Taipei Medical University in Taiwan, wrote in an editorial published alongside the study. Whats more, the benefits of eating high-fat cheeses were most evident when participants swapped cheese for other foods, like processed or high-fat red meat, which might just reveal the difference of better options, according to Yeh. It is not so much that high-fat cheese is inherently neuroprotective, but rather that it is a less harmful choice than red and processed meats, she wrote. BENEFITS OF CHEESE The findings may not apply to somewhere like the U.S., where much of our cheese is processed, according to Emily Sonestedt, who led the new study and is a senior lecturer and associate professor of nutrition at Lund University in Sweden, Still, its possible that there are benefits from certain healthful components of cheese, like vitamins K or B12, or minerals like calcium, she told The New York Times. As with any of these sorts of studies, its also important for people to remember that correlation doesnt imply causationsomething Sonestedt reinforced. This does not prove that cheese prevents dementia, but it does challenge the idea that all high-fat dairy is bad for the brain, Sonestedt said in an email to CNN. HIGH-FAT FOODS IN FOCUS Some people dont need purported brain benefits to convince them to eat foods high in saturated fats. These foods have been embraced in the keto diet, among others, in recent years, despite long-standing nutrition guidelines that recommend people limit their consumption of foods high in saturated fats because of the evidence that they raise LDL cholesterol levels, along with the risk of heart attack or stroke. But those guidelines are likely to see a shake-up in 2026 as Robert F. Kennedy Jr., the U.S. secretary for health and human services, has said that the next edition of the federal dietary guidelines will instead stress the need to eat saturated fats, dairy, fresh meat, and vegetables. And even if the results of this study are appealing to cheese lovers, the quirks of how the research was conducted mean that some experts arent exactly buying the results. NOT BUYING IT In fact, because the link between cheese consumption and dementia risk was at the margin of statistical significance, it could be due to just chance, notes Dr. Walter Willett, professor of epidemiology and nutrition at Harvard T.H. Chan School of Public Health and professor of medicine at Harvard Medical School in Boston. Im not running out to buy a block of cheese, Willett said in an email to CNN.
Category:
E-Commerce
All news |
||||||||||||||||||
|
||||||||||||||||||