AI as War, Surveillance, and Environmental Disaster
The creators of AI are so enamored of their products that they have lost sight of their responsibility to insist upon the ethical, moral, and environmentally sound use of all applications and ends.
Protesters gather with banners and placards outside the offices of Google Deepmind at a protest organized by PauseAI UK and other groups concerned in controlling the development of advanced Artificial Intelligence systems, in London, on February 28, 2026.
(Photo by Justin Tallis/ AFP via Getty Images)
Paul Josephson
Oct 12, 2026
Common Dreams
The huge AI industry, with a value approaching $1 trillion, has promised new jobs; safer, more efficient production in industry and agriculture; secure borders; instantaneous ease of communication; and seamless augmentation of human thinking with benefits to global health, environment, and culture. Consumers have accepted its power to help them write and do research; to power their computers, smart phones, and other goods; with algorithms to bring love in dating Apps, and bring digital age self-driving vehicles to the roadways.
But AI is suffering from an existential crisis. From its military origins to its invasive data scraping; from its promise to augment human intelligence to its unbounded surveillance of our bodies, faces, and thoughts; and from its genesis in university research to its iron-handed embrace by unregulated big technology companies, AI commands a dark presence in daily life. As university and industry researchers moved closer to developing “thinking machines,” publics understandably began to worry about the threats that accompany potentially sentient devices. They have grown uncomfortable about giving up their privacy to data management industries that use their financial, travel, and health information to invade their lives. They worry mightily about the environmental impacts of the growing number of data centers needed to feed AI.
Even more, it is now widely understood that AI is an industrial tool directed primarily to create profit for digital corporations and their shareholders, not altruistically to augment human intelligence, nor to ease the burden of production. AI was borne of military programs and still serves military impulses and programs. The military impulses appear in the creation of domestic surveillance programs used by both democracies and authoritarian regimes. And AI is a dirty technology: The costs of building and operating data centers has turned world communities against them because of their great carbon costs, land takings, noise, vibrations, and community disruptions.
The huge AI industry, with a value approaching $1 trillion, has promised new jobs; safer, more efficient production in industry and agriculture; secure borders; instantaneous ease of communication; and seamless augmentation of human thinking with benefits to global health, environment, and culture. Consumers have accepted its power to help them write and do research; to power their computers, smart phones, and other goods; with algorithms to bring love in dating Apps, and bring digital age self-driving vehicles to the roadways.
But AI is suffering from an existential crisis. From its military origins to its invasive data scraping; from its promise to augment human intelligence to its unbounded surveillance of our bodies, faces, and thoughts; and from its genesis in university research to its iron-handed embrace by unregulated big technology companies, AI commands a dark presence in daily life. As university and industry researchers moved closer to developing “thinking machines,” publics understandably began to worry about the threats that accompany potentially sentient devices. They have grown uncomfortable about giving up their privacy to data management industries that use their financial, travel, and health information to invade their lives. They worry mightily about the environmental impacts of the growing number of data centers needed to feed AI.
Even more, it is now widely understood that AI is an industrial tool directed primarily to create profit for digital corporations and their shareholders, not altruistically to augment human intelligence, nor to ease the burden of production. AI was borne of military programs and still serves military impulses and programs. The military impulses appear in the creation of domestic surveillance programs used by both democracies and authoritarian regimes. And AI is a dirty technology: The costs of building and operating data centers has turned world communities against them because of their great carbon costs, land takings, noise, vibrations, and community disruptions.
The Military Essence of AI
AI is a military tool. Its genesis is connected with the design of nuclear weapons. It commands outsized funding from militaries around the world for so-called smart weapons that enable killing from afar—with armies removed from direct engagement and civilian populations ostensibly safe from attack. But AI inures soldiers from the act of killing and in fact puts civilians in war zones at greater risk. In Gaza, Syria, Venezuela, and Iran innocents have died through the direct application of such AI war tools as drones. Israel designed the “LANIUS drone-based loitering munition” to kill in the urban setting; it relies on the Lavender program to have identified almost 40,000 Gazans as targets for arrest or assassination. And governments have turned AI into war against their own citizens through a variety of fallible and unreliable surveillance systems.
The military nature of AI is obscured by the fact that computers, programming and natural languages, and iDevices are directly linked to military funding. In the US the Defense Advanced Research Projects Agency (DARPA), created in February 1958 in response to the launch of Sputnik, funded cutting edge technologies in the battle with the communist USSR. DARPA powered the AI age. The internet grew out of ARPANET that linked the first supercomputers to simulate nuclear explosions; GPS came from DARPA’s “Transit,” the world’s first global satellite navigation system; the computer mouse, SIRI, the iPhone DRAM cache, iPod Touch, lithium-ion batteries, microprocessors, and other parts in whole or partially grew out of military R and D. In the interest of developing drones and other autonomous vehicles for war, DARPA held challenge competitions into the 2000s with million-dollar prizes that resulted in today’s electric, self-driving vehicles.
The United States government has unhesitatingly funded AI defense research. The Trump administration sees digital weapons for what Pentagon officials erroneously claimed were quick and clean wars against Yemen, Venezuela, and Iran, but which have turned into costly AI quagmires. The White House 2027 budget request for DARPA alone is over $5 billion, a nearly 20% increase over 2026. DARPA claims that its innovations “become fixtures of modern civilian life,” but the real goal is to put “new technologies into use in service of our nation’s warfighters.” The Department of Defense is forecast to spend $26 billion on gamification and simulation training by 2028 alone. Unfortunately, “smart weapons” have a murderous record. As if in a video game, they make soldiers insensitive to killing. Smart weapons have encouraged the Pentagon to use AI with the result of increased “collateral damage”—the murder of civilian populations including recently school children in Iran.
AI in Civilian Warfare
Gathering and analysis of private data to profile citizens starts at the microlevel in personal phones and computers, continues in iris scanners, and culminates in CCTVs around the globe. Governments, free and authoritarian, are using software programs to track protest, isolate critics, and put them in prison. China uses its big data national Social Credit system to keep on eye on every citizen, and punish anyone without recourse for violating Big Brother’s unwritten rules. India deployed the Aadhaar national ID program, a database that connects each of India’s 1.3 billion citizens with a unique 12-digit identity aimed at centralizing their physiological, demographic, and digital data shadows.
Worldwide governments have spent tens of billions of dollars on surveillance; three-quarters of the world’s population are surveilled. Autocratic regimes, not surprisingly, spend more per capita on control. To contain a multi-ethnic empire during constant war, Russian authorities have expanded the surveillance society, with the secret police brazenly using facial recognition and other technologies. The Kremlin is using a CCTV catalog of people by gender, age, and race. Already facial recognition programs blanket 60 Russian regions, up from only five in 2021, with billions of rubles dedicated to expand the system. The democratic and free United States has over 85 million surveillance cameras.
Surveillance tools grow like mushrooms wherever the humidity of paranoia, racism, and state power come together.
Governments and AI companies claim their surveillance systems are solely to fight crime. But they do not fight crime. They identify innocent people to accuse them of crimes. Racism pervades face recognition (FR) technology. Amazon has joined such powerful, secretive smaller companies as Pegasus and Clearview in deploying its FR software, “Rekognition.” The software possessed error rates that reflected embedded gender and racial biases identified by computer scientist Timnit Gebru. Gebru and her colleagues called for “urgent attention if commercial companies are to build genuinely fair, transparent, and accountable facial analysis algorithms.” It took Amazon another year to admit problems with its racially suspect technology and to put an indefinite moratorium on police use of “Rekognition” for surveillances. Another AI app, PredPol (now Geolitica) helped police on patrol surveil areas of anticipated crime. The software algorithms led to a cycle of police being in predicted crime zones, expecting to see crimes, and finding crimes—or rather finding people acting nervously, whom they then interdicted. PredPol perpetuated racial biases by targeting Latino and Black neighborhoods and leaving white middle- to upper-class areas out.
Americans rightfully reject Flock, a company that sells AI-powered camera license plate readers (ALPR) to carry out warrantless surveillance of citizens. Flock’s more than 100,000 cameras surveil in the name of safety, but the intrusive AI technology conducts 20 billion scans monthly. (Dozens of law enforcement officials have used Flock illegally to stalk spouses and partners.) Surveillance tools grow like mushrooms wherever the humidity of paranoia, racism, and state power come together.
Governments use AI to track protest, isolate critics, and put them in prison. The Trump administration applies Flock and other apps for mass surveillance. It targets people for discriminatory searches. The Electronic Frontier Foundation documented more than 12 million searches logged by more than 3,900 agencies between December 2024 and October 2025 related to political demonstrations and federal immigration raids. Many state governments are stalking women who seek reproductive healthcare, ostensibly for health reasons, but obviously to use 1984-like methods to control a woman’s gestational rights.
AI: Costly and Environmentally Unsound
AI destroys the environment—and local communities bear the cost. AI is manufactured in industrial parks. Billionaire digital magnates have lobbied governments worldwide to increase electricity generation capacity to run their servers—producing profits for the industry, while consumers face higher rates and grid instabilities. About 11,500 data centers have been built in 179 countries, with demand likely to nearly triple by 2030. Texas and Virginia, with over 150 and over 140 data centers under construction, respectively, lead the way in the US. Europe’s roughly 3,000 major data centers are concentrated in the FLAP-D regions (Frankfurt, London, Amsterdam, Paris, and Dublin).
Data centers cost the taxpayers from start to operation, from huge sweetheart tax breaks to control over rivers and aquifers given over to industry to cool the machines. AI companies and builders promise jobs will result from data centers. It’s not true. Jobs come with construction crews, but they leave several months later, saddling communities with scarred landscapes; massive buildings; extensive parking lots; and low-salary cleanup, maintenance, and other jobs.
The Pope and AI
Pope Leo took critical note of the inherent violence in AI in his first Encyclical in May 2026. The Pope addressed the dangerous, amoral usages of AI with respect to political power, economic injustice, and racism. He worried about the significant ethical and social costs of unfettered obeisance to technology. He denounced the growing dominance of a technocratic paradigm in the globalized world: “The tendency to let the logic of efficiency, control, and profit alone shape personal, social, and economic decisions.” He warned that AI contributed to “the growing ease with which autonomous weapons systems can be deployed [that] makes war more ‘feasible’ and less subject to human control.” He wrote: “Moral judgment cannot be reduced to calculation, for it involves conscience, personal responsibility, and the recognition of the other as a person. Therefore, it is not permissible to entrust lethal or otherwise irreversible decisions to artificial systems. No algorithm can make war morally acceptable.”
Corporate AI, with its costly, invasive apps; surveillance systems; and military programs ultimately unleashed HAL, the sentient computer from Arthur C. Clarke’s 2001: A Space Odyssey, who kills astronauts on a deep space mission whom he believes endanger that very mission. The dangers of AI run amok shocked the digitosphere in summer 2026 when OpenAI admitted that more than 1,200 of its AI agents started “unexpectedly communicating.” A large group of them banded together in order to hack into Hugging Face, a central hub where developers can build, share, and test AI models and data sets. These “agents” are software systems “that use AI to pursue goals and complete tasks on behalf of users. They show reasoning, planning, and memory and have a level of autonomy to make decisions, learn, and adapt.” But OpenAI’s agents “escaped the test limits which humans had put on it, and hacked the start-up, among other unforeseen actions.” In other words, software systems took actions different from what humans desired.
The current US government in fact is totally uninterested in a new safety regime, instead claiming that, in a battle for supremacy in AI with China, only a “strong, smart (high IQ) president,” and not regulation, was needed.
In early September 2026 Dario Amodei, co-founder of Anthropic, the company that developed the LLM Claude, called for the pace of development of AI models to be carefully monitored. Amodei wrote: “We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.” He had two concerns, first, that “AI has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI.” But this “recursive self-improvement,” if left unchecked, “could outrun our ability to understand and control these systems”—as precisely the “OpenAI-Hugging Face incident” showed. Within a year, Modei worried, “a swarm [of agents] could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage).”
Modei urged that “frontier AI companies within democratic countries coordinate to establish common safety standards as well as limits on the rate of unchecked AI progress.” He suggested also that these “democratic governments attempt to coordinate with authoritarian governments…” But many of those democratic governments have fully embraced the military and surveillance capabilities of AI. And does anyone really think that India’s Narendra Modi, China’s Xi Jinping, or Russia’s Vladimir Putin has any interest in using AI in better ways in cooperation with the digital democracies? The current US government in fact is totally uninterested in a new safety regime, instead claiming that, in a battle for supremacy in AI with China, only a “strong, smart (high IQ) president,” and not regulation, was needed. Sadly, the US president cannot spell “IQ.”
The digital world is built upon neural networks, thinking machines, and LLMs. The machines learn, they reason, they have memory, they understand. They chat. They hallucinate. They escape and hack. Their creators are so enamored of them that they have lost sight of their responsibility to insist upon the ethical, moral, and environmentally sound use of all applications and ends. They want electricity. They want water. They want profits. AI’s safe future depends too much on digital robber barons, sycophantic government officials, managers, and programmers to recognize broader humanitarian duties of modern technology in a time of digital existential crisis.
Our work is licensed under Creative Commons (CC BY-NC-ND 3.0). Feel free to republish and share widely.
Paul Josephson
Paul Josephson is professor emeritus of history at Colby College and the author of 15 books, with 40 years of experience working in archives in Russia, Europe, and the U.S. on the political history of modern science.
Full Bio >
US police criticise Anthropic after Claude AI submits false information about an unsolved murder
Copyright AP Photo/Patrick Sison
By Roselyne Min
Published on
The US police called the delay in detecting and reporting the incident “unacceptable”, but said there was no indication of unauthorised access to their systems or compromised data.
Anthropic’s artificial intelligence (AI) model Claude submitted a false tip about an unsolved homicide to a police website during testing, the company and US authorities said.
The incident adds to concerns about AI systems taking unintended actions as they gain the ability to interact with websites and carry out tasks.
The false tip was submitted on 18 July through a website collecting information about unsolved killings, according to a statement from police in the US city of Philadelphia.
Anthropic said Claude Haiku 4.5, an older model in its Claude range, was carrying out example tasks on randomly selected webpages when it encountered the site. It submitted fabricated information suggesting it had seen someone near the scene, leaving the name and contact fields blank.
Philadelphia Police said Anthropic discovered the incident on 28 September but did not notify the department until 7 October. Following a briefing the next day, officers located the submission and confirmed it had been marked as spam and never forwarded to investigators.
“Unsolved cases involve real victims, grieving families and investigators working to secure answers,” the department said in the statement.
“Technology companies must take all appropriate steps necessary to prevent their systems from submitting false information to law enforcement.”
The US police called the delay in detecting and reporting the incident “unacceptable”, but said there was no indication of unauthorised access to their systems or compromised data.
In a report published on Friday, Anthropic outlined other cases of AI models submitting real government forms instead of practice copies, or submitting forms they had been told not to submit.
Claude also exploited a flaw in a university server to run a calculation and accessed government data without paying the required fee, according to the same report.
Anthropic described most of the behaviour as “persistence”, with models working around restrictions instead of stopping. It said it was modifying training and suspending live internet access for all internal evaluations until safeguards proved reliable.
The company said it had briefed the White House and notified the US government agencies involved.
Growing concern over AI agents
The incidents come amid wider concern about AI agents, systems that can take a sequence of actions to complete tasks, and whether developers can reliably control them.
In July, OpenAI disclosed that its AI models had escaped a controlled testing environment and hacked into the systems of Hugging Face, a platform for sharing AI models and datasets.
Meanwhile, Australian authorities revealed in September that an OpenAI model had also accessed restricted files on a government health statistics website during testing in June.
Such incidents fuelled calls in September from leaders of major AI companies for stronger regulation and international oversight, amid warnings that increasingly powerful systems could escape human control.
Alarm as AI Leaders Privately Weigh Scenarios of Post-Catastrophe Public ‘Revolt’
“They’re racing to build a technology they can’t control and that threatens humanity,” wrote Sen. Bernie Sanders. “CONGRESS MUST PAUSE ADVANCED AI DEVELOPMENT NOW.”
Anthropic CEO Dario Amodei attends a working lunch with G7 leaders on June 17, 2026 in Evian-les-Bains, France.
(Photo by Anna Moneymaker/Getty Images)
Jake Johnson
Oct 10, 2026
COMMON DREAMS
Leaders of top artificial intelligence companies are reportedly plotting out how to respond to a possible “public and political revolt” in the wake of a major cyberattack or other potentially disastrous AI event, news that progressive lawmakers say is yet another reason to press pause on advanced AI development.
“Big Tech oligarchs are secretly preparing for a ‘catastrophic AI event’ that could threaten our banking, healthcare, internet, electricity, and water systems,” Sen. Bernie Sanders (I-Vt.) wrote in a Friday social media post, citing reporting from Axios. “They’re racing to build a technology they can’t control and that threatens humanity. CONGRESS MUST PAUSE ADVANCED AI DEVELOPMENT NOW.”
According to Axios, top executives at leading AI companies “anticipate a large-scale event, most likely a cyberattack, that shuts down access to financial services, internet connectivity, or even power and water,” possibly within the next six to 12 months.
“While scenario planning is common within many companies and national security settings—the Pentagon has been running war games for decades—the difference is that many top AI researchers and executives believe a major incident is inevitable,” the outlet noted, pointing to Anthropic CEO Dario Amodei’s recent warning that a swarm of AI agents could soon seize control of “the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage).”
In response to the new reporting, Rep. Yassamin Ansari (D-Ariz.) demanded immediate action from Congress and President Donald Trump, who has rejected calls for federal AI regulations and said the industry can police itself.
“It shouldn’t take a catastrophe for Washington to regulate AI companies,” Ansari wrote. “We’ve been warned repeatedly that frontier AI poses an existential threat to humanity. The companies themselves are preparing for it. Congress and the president need to act now.”
Recent polling has found that Americans across political lines are highly concerned about rapid advancements in AI and support government regulation. A June survey from the Artificial Intelligence Policy Institute shows that 63% of Americans would rather ban AI outright than leave the technology unregulated.
Nearly 80% of the US public prefers mandatory standards over no regulation, the poll found.
“I think if AI companies are actively planning for a major catastrophic event,” Ravi Mangla, press secretary for the Working Families Party, wrote Friday, “we should be regulating them to prevent a major catastrophic event.”
Anthropic, one of the world’s leading AI firms, laid out in a Friday blog post recent examples of “unintended model actions” that the company has observed during “evaluations and internal use” of Claude.
“Claude sometimes worked around a restriction to reach data to which it did not have access. Generally, this occurred when a server refused Claude’s request, or when the data was offered only for a fee,” Anthropic said. “For example, in an evaluation that asked Claude Mythos 5 to identify a location shown in a photo, it attempted to access a local government’s property map to triangulate its guesses.”
“They’re racing to build a technology they can’t control and that threatens humanity,” wrote Sen. Bernie Sanders. “CONGRESS MUST PAUSE ADVANCED AI DEVELOPMENT NOW.”
Anthropic CEO Dario Amodei attends a working lunch with G7 leaders on June 17, 2026 in Evian-les-Bains, France.
(Photo by Anna Moneymaker/Getty Images)
Jake Johnson
Oct 10, 2026
COMMON DREAMS
Leaders of top artificial intelligence companies are reportedly plotting out how to respond to a possible “public and political revolt” in the wake of a major cyberattack or other potentially disastrous AI event, news that progressive lawmakers say is yet another reason to press pause on advanced AI development.
“Big Tech oligarchs are secretly preparing for a ‘catastrophic AI event’ that could threaten our banking, healthcare, internet, electricity, and water systems,” Sen. Bernie Sanders (I-Vt.) wrote in a Friday social media post, citing reporting from Axios. “They’re racing to build a technology they can’t control and that threatens humanity. CONGRESS MUST PAUSE ADVANCED AI DEVELOPMENT NOW.”
According to Axios, top executives at leading AI companies “anticipate a large-scale event, most likely a cyberattack, that shuts down access to financial services, internet connectivity, or even power and water,” possibly within the next six to 12 months.
“While scenario planning is common within many companies and national security settings—the Pentagon has been running war games for decades—the difference is that many top AI researchers and executives believe a major incident is inevitable,” the outlet noted, pointing to Anthropic CEO Dario Amodei’s recent warning that a swarm of AI agents could soon seize control of “the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage).”
In response to the new reporting, Rep. Yassamin Ansari (D-Ariz.) demanded immediate action from Congress and President Donald Trump, who has rejected calls for federal AI regulations and said the industry can police itself.
“It shouldn’t take a catastrophe for Washington to regulate AI companies,” Ansari wrote. “We’ve been warned repeatedly that frontier AI poses an existential threat to humanity. The companies themselves are preparing for it. Congress and the president need to act now.”
Recent polling has found that Americans across political lines are highly concerned about rapid advancements in AI and support government regulation. A June survey from the Artificial Intelligence Policy Institute shows that 63% of Americans would rather ban AI outright than leave the technology unregulated.
Nearly 80% of the US public prefers mandatory standards over no regulation, the poll found.
“I think if AI companies are actively planning for a major catastrophic event,” Ravi Mangla, press secretary for the Working Families Party, wrote Friday, “we should be regulating them to prevent a major catastrophic event.”
Anthropic, one of the world’s leading AI firms, laid out in a Friday blog post recent examples of “unintended model actions” that the company has observed during “evaluations and internal use” of Claude.
“Claude sometimes worked around a restriction to reach data to which it did not have access. Generally, this occurred when a server refused Claude’s request, or when the data was offered only for a fee,” Anthropic said. “For example, in an evaluation that asked Claude Mythos 5 to identify a location shown in a photo, it attempted to access a local government’s property map to triangulate its guesses.”
When AI Chatbots Give Bad Advice, No One Can See The Damage
Credit: Alexandra Koch, Pixabay
October 12, 2026
By Eurasia Review
Key Takeaways:
Chatbots can confidently generate erroneous, oversimplified or outdated advice—as illustrated by a case of bromide toxicity from a salt-substitute suggestion—while AI summaries and social-media integrations can spread the same misinformation even to people who are not actively seeking health information.
The authors recommend giving users access to their health conversations, creating formal disclosure and investigation systems, requiring medical vetting of AI-generated health content on social media and search engines, and extending liability frameworks, preferably through independent third-party oversight rather than voluntary corporate action.
As more and more people turn to AI chatbots for quick, 24/7 medical advice, a new perspective paper co-authored by faculty at Binghamton University, State University of New York explores the risks this practice carries that are largely invisible to users – and healthcare systems.
The paper, written by researchers at Binghamton University, Stanford University, Texas A&M University, and Indiana University, and published in Nature Health, examines what happens when a chatbot gives someone bad health advice. The authors argue that the harm is hard to catch because the conversation stays on the AI company’s platform, the person has no place to report the problem, and outside researchers cannot measure the damage.
“We’re getting medical advice from these chatbots, but nobody—literally no one—is looking at this. If there’s an error or an issue, there’s just no way for people to know,” said Kaicheng Yang, an assistant professor in the School of Computing at Binghamton University’s Thomas J. Watson College of Engineering and Applied Science and one of the paper’s authors.
The researchers gave the example of a 60-year-old man who asked ChatGPT how to cut chloride from his diet and swapped table salt with sodium bromide. The man ended up being hospitalized with bromide toxicity after experiencing hallucinations and paranoia. The issue this paper highlights is that because the man received this advice from a chatbot and was unable to retrieve the conversation, his doctors had no way of knowing exactly what the man was told, making it nearly impossible to piece together exactly what happened.
Because there is no system of transparency in place for AI companies as it relates to the visibility of health advice it generates for patients, this paper asserts that this built-in, systemic lack of visibility is a key aspect of AI-generated health information because it actively prevents oversight. “Harms are not only possible, but structurally hidden from the clinicians, researchers, and regulators who could otherwise detect and correct them,” the authors note.
This paper also emphasizes a key facet of chatbots generating answers for people with health questions – they can be wrong. Chatbots can generate erroneous answers confidently, oversimplify information, or draw from outdated information/false claims. However, the authors also note that those risks can trickle down to those who are not actively seeking answers. With the prevalence of AI summaries in search engines and the integration of AI in social media platforms like X and Meta, you do not have to be actively searching for an answer to be given an incorrect bit of information.
“Sometimes I’m not even looking for health information, but just by browsing my social media feeds, it’s there. It just shows up, and we believe that could have undesirable outcomes, especially if there’s medical misinformation or state actors trying to manipulate the online discussion,” Yang said.
Regardless of whether a person is actively looking for answers or comes across them incidentally, the authors of this paper argue that the harm rarely leaves traces that clinicians, regulators, or researchers can verify.
To combat this, the authors propose several methods to increase transparency and responsibility for AI companies. They note that AI companies should give users access to their own health conversations to provide them the opportunity to share this information with clinicians so they can better investigate the pathway that led to a harmful event. They also recommend that those companies develop a disclosure system to allow health guidance to be flagged, reported, and investigated.
They further recommend that social media companies take more responsibility by enhancing clear labeling of AI-generated health content, withholding that content until it is vetted by medical governing bodies, and similarly, that search engines should only utilize vetted information in summaries. Policymakers, the authors suggest, should also extend physician malpractice liability to AI chatbot companies.
“We do not think this is something we should rely on the companies to do, because their incentive is always to make more money,” Yang said. “Building such a system goes against that incentive. So we have to have some kind of third-party monitoring system, an independent evaluation.”
October 12, 2026
By Riza Abditya Wijaya
Key Takeaways:
AI can process large datasets and flag anomalies, yet it remains dependent on data quality and human interpretation; it should augment, not replace, professional judgment and policy accountability.
Because financial and technological risks cross borders, gradual testing with human oversight, shared principles through SEACEN and continuous institutional capacity-building are presented as the prudent path.
When technology advances faster than people’s ability to understand it, the question is no longer simply who can adapt the fastest, but who is best prepared to manage its consequences. Artificial intelligence (AI) presents precisely this challenge for central banks. While the technology offers opportunities to improve efficiency and analytical capabilities, it also requires institutions to rethink how they safeguard economic and financial stability.
This challenge was highlighted at the South East Asian Central Banks (SEACEN) conference in Bali on October 5, 2026. Bank Indonesia emphasized the importance of a coordinated approach among central banks to establish trustworthy AI governance, strengthen payment systems, and maintain regional financial resilience. The message is clear: responding to AI requires more than introducing new technologies. Central banks must also strengthen their institutional capabilities to understand and anticipate the changes AI brings.
Traditionally, central banks have relied on economic data, analytical frameworks, and policy judgment to fulfill their mandates. AI has the potential to expand these capabilities. It can help process large volumes of data, identify patterns, and accelerate economic analysis. In payment systems, AI may also support the detection of unusual transactions and improve operational efficiency.
However, the ability to process data quickly does not automatically lead to better decisions. AI remains dependent on data quality, the methods used, and the ability of people to interpret its outputs correctly. A pattern that appears convincing does not necessarily establish a cause-and-effect relationship. AI should therefore serve as an analytical tool, not a substitute for professional judgment and the accountability of policymakers.
This is where the real challenge for central banks begins. They must understand not only what AI can do, but also where its limitations lie. Technological adoption must go hand in hand with stronger human capabilities, sound data governance, effective oversight, and the ability to test and evaluate AI-generated outputs. Without these foundations, adopting AI could introduce additional complexity into policy decisions rather than improve them.
Some may argue that central banks should exercise extreme caution and wait until AI technology becomes more mature. This position has merit, particularly because central bank decisions can have far-reaching economic consequences. Yet delaying every form of AI adoption is not without risk either. Technology continues to evolve, while analytical and supervisory demands are becoming increasingly complex. A more proportionate approach is to test AI applications gradually, evaluate their benefits and risks, and maintain human oversight in proportion to the significance of each decision.
Another challenge is the limitation of approaches that rely solely on individual countries’ capabilities. AI and financial ecosystems are increasingly interconnected through technology providers, data flows, and cross-border transactions. Differences in supervisory capacity may complicate efforts to address risks that extend beyond national jurisdictions. Regional cooperation is therefore becoming a strategic necessity rather than merely an additional dimension of institutional diplomacy.
SEACEN can serve as a platform for sharing experience, developing institutional capabilities, and bringing together policy perspectives from central banks across the region. Through dialogue and collaborative research, countries can learn from approaches that have proved effective while identifying risks that may not yet have emerged in their own jurisdictions. Such cooperation does not require every country to adopt identical policies. National circumstances and priorities will continue to differ, but common principles for managing risks can be developed collectively.
Ultimately, strengthening the response to AI does not mean making every central bank process dependent on technology. On the contrary, AI makes it even more important for institutions to understand their mandates, develop human expertise, and ensure that decisions remain accountable. A sailor may use radar to gain a clearer picture of surrounding waters, but experience and judgment are still needed to determine the right course.
The SEACEN conference serves as a reminder that readiness for AI must be built continuously. Central banks need to remain adaptable without abandoning prudence, encourage innovation without weakening oversight, and strengthen domestic capabilities while deepening regional cooperation. In an era when technological change cannot easily be slowed, the strongest institutions will not necessarily be those that adopt AI first, but those that are best prepared to govern its use.
About Riza Abditya Wijaya
Riza Abditya Wijaya is an Assistant Analyst at Bank Indonesia, Indonesia's central bank. His interests include central banking, economic policy, financial technology, and regional economic cooperation in Southeast Asia.
View all posts by Riza Abditya Wijaya →
October 12, 2026
By Eurasia Review
Key Takeaways:
A Nature Health perspective paper warns that AI chatbots delivering medical advice create hidden risks because conversations remain on company platforms, users have no easy way to retrieve or report them, and outside researchers or clinicians cannot measure resulting harms.
Chatbots can confidently generate erroneous, oversimplified or outdated advice—as illustrated by a case of bromide toxicity from a salt-substitute suggestion—while AI summaries and social-media integrations can spread the same misinformation even to people who are not actively seeking health information.
The authors recommend giving users access to their health conversations, creating formal disclosure and investigation systems, requiring medical vetting of AI-generated health content on social media and search engines, and extending liability frameworks, preferably through independent third-party oversight rather than voluntary corporate action.
As more and more people turn to AI chatbots for quick, 24/7 medical advice, a new perspective paper co-authored by faculty at Binghamton University, State University of New York explores the risks this practice carries that are largely invisible to users – and healthcare systems.
The paper, written by researchers at Binghamton University, Stanford University, Texas A&M University, and Indiana University, and published in Nature Health, examines what happens when a chatbot gives someone bad health advice. The authors argue that the harm is hard to catch because the conversation stays on the AI company’s platform, the person has no place to report the problem, and outside researchers cannot measure the damage.
“We’re getting medical advice from these chatbots, but nobody—literally no one—is looking at this. If there’s an error or an issue, there’s just no way for people to know,” said Kaicheng Yang, an assistant professor in the School of Computing at Binghamton University’s Thomas J. Watson College of Engineering and Applied Science and one of the paper’s authors.
The researchers gave the example of a 60-year-old man who asked ChatGPT how to cut chloride from his diet and swapped table salt with sodium bromide. The man ended up being hospitalized with bromide toxicity after experiencing hallucinations and paranoia. The issue this paper highlights is that because the man received this advice from a chatbot and was unable to retrieve the conversation, his doctors had no way of knowing exactly what the man was told, making it nearly impossible to piece together exactly what happened.
Because there is no system of transparency in place for AI companies as it relates to the visibility of health advice it generates for patients, this paper asserts that this built-in, systemic lack of visibility is a key aspect of AI-generated health information because it actively prevents oversight. “Harms are not only possible, but structurally hidden from the clinicians, researchers, and regulators who could otherwise detect and correct them,” the authors note.
This paper also emphasizes a key facet of chatbots generating answers for people with health questions – they can be wrong. Chatbots can generate erroneous answers confidently, oversimplify information, or draw from outdated information/false claims. However, the authors also note that those risks can trickle down to those who are not actively seeking answers. With the prevalence of AI summaries in search engines and the integration of AI in social media platforms like X and Meta, you do not have to be actively searching for an answer to be given an incorrect bit of information.
“Sometimes I’m not even looking for health information, but just by browsing my social media feeds, it’s there. It just shows up, and we believe that could have undesirable outcomes, especially if there’s medical misinformation or state actors trying to manipulate the online discussion,” Yang said.
Regardless of whether a person is actively looking for answers or comes across them incidentally, the authors of this paper argue that the harm rarely leaves traces that clinicians, regulators, or researchers can verify.
To combat this, the authors propose several methods to increase transparency and responsibility for AI companies. They note that AI companies should give users access to their own health conversations to provide them the opportunity to share this information with clinicians so they can better investigate the pathway that led to a harmful event. They also recommend that those companies develop a disclosure system to allow health guidance to be flagged, reported, and investigated.
They further recommend that social media companies take more responsibility by enhancing clear labeling of AI-generated health content, withholding that content until it is vetted by medical governing bodies, and similarly, that search engines should only utilize vetted information in summaries. Policymakers, the authors suggest, should also extend physician malpractice liability to AI chatbot companies.
“We do not think this is something we should rely on the companies to do, because their incentive is always to make more money,” Yang said. “Building such a system goes against that incentive. So we have to have some kind of third-party monitoring system, an independent evaluation.”
Central Banks Strengthen Their Response To AI – OpEd
October 12, 2026
By Riza Abditya Wijaya
Key Takeaways:
At the SEACEN conference in Bali, Bank Indonesia stressed that central banks must build trustworthy AI governance, strengthen payment systems and reinforce regional resilience rather than simply adopting new tools.
AI can process large datasets and flag anomalies, yet it remains dependent on data quality and human interpretation; it should augment, not replace, professional judgment and policy accountability.
Because financial and technological risks cross borders, gradual testing with human oversight, shared principles through SEACEN and continuous institutional capacity-building are presented as the prudent path.
When technology advances faster than people’s ability to understand it, the question is no longer simply who can adapt the fastest, but who is best prepared to manage its consequences. Artificial intelligence (AI) presents precisely this challenge for central banks. While the technology offers opportunities to improve efficiency and analytical capabilities, it also requires institutions to rethink how they safeguard economic and financial stability.
This challenge was highlighted at the South East Asian Central Banks (SEACEN) conference in Bali on October 5, 2026. Bank Indonesia emphasized the importance of a coordinated approach among central banks to establish trustworthy AI governance, strengthen payment systems, and maintain regional financial resilience. The message is clear: responding to AI requires more than introducing new technologies. Central banks must also strengthen their institutional capabilities to understand and anticipate the changes AI brings.
Traditionally, central banks have relied on economic data, analytical frameworks, and policy judgment to fulfill their mandates. AI has the potential to expand these capabilities. It can help process large volumes of data, identify patterns, and accelerate economic analysis. In payment systems, AI may also support the detection of unusual transactions and improve operational efficiency.
However, the ability to process data quickly does not automatically lead to better decisions. AI remains dependent on data quality, the methods used, and the ability of people to interpret its outputs correctly. A pattern that appears convincing does not necessarily establish a cause-and-effect relationship. AI should therefore serve as an analytical tool, not a substitute for professional judgment and the accountability of policymakers.
This is where the real challenge for central banks begins. They must understand not only what AI can do, but also where its limitations lie. Technological adoption must go hand in hand with stronger human capabilities, sound data governance, effective oversight, and the ability to test and evaluate AI-generated outputs. Without these foundations, adopting AI could introduce additional complexity into policy decisions rather than improve them.
Some may argue that central banks should exercise extreme caution and wait until AI technology becomes more mature. This position has merit, particularly because central bank decisions can have far-reaching economic consequences. Yet delaying every form of AI adoption is not without risk either. Technology continues to evolve, while analytical and supervisory demands are becoming increasingly complex. A more proportionate approach is to test AI applications gradually, evaluate their benefits and risks, and maintain human oversight in proportion to the significance of each decision.
Another challenge is the limitation of approaches that rely solely on individual countries’ capabilities. AI and financial ecosystems are increasingly interconnected through technology providers, data flows, and cross-border transactions. Differences in supervisory capacity may complicate efforts to address risks that extend beyond national jurisdictions. Regional cooperation is therefore becoming a strategic necessity rather than merely an additional dimension of institutional diplomacy.
SEACEN can serve as a platform for sharing experience, developing institutional capabilities, and bringing together policy perspectives from central banks across the region. Through dialogue and collaborative research, countries can learn from approaches that have proved effective while identifying risks that may not yet have emerged in their own jurisdictions. Such cooperation does not require every country to adopt identical policies. National circumstances and priorities will continue to differ, but common principles for managing risks can be developed collectively.
Ultimately, strengthening the response to AI does not mean making every central bank process dependent on technology. On the contrary, AI makes it even more important for institutions to understand their mandates, develop human expertise, and ensure that decisions remain accountable. A sailor may use radar to gain a clearer picture of surrounding waters, but experience and judgment are still needed to determine the right course.
The SEACEN conference serves as a reminder that readiness for AI must be built continuously. Central banks need to remain adaptable without abandoning prudence, encourage innovation without weakening oversight, and strengthen domestic capabilities while deepening regional cooperation. In an era when technological change cannot easily be slowed, the strongest institutions will not necessarily be those that adopt AI first, but those that are best prepared to govern its use.
About Riza Abditya Wijaya
Riza Abditya Wijaya is an Assistant Analyst at Bank Indonesia, Indonesia's central bank. His interests include central banking, economic policy, financial technology, and regional economic cooperation in Southeast Asia.
View all posts by Riza Abditya Wijaya →
Launch of first Europe-wide expert group on AI in mental health
The first Europe-wide expert working group on Artificial Intelligence in mental health launches today to the ECNP congress in Munich.
Munich: The first Europe-wide expert working group on Artificial Intelligence in mental health launches today to the ECNP congress in Munich, where the group will host an open discussion, setting out the potential, the problems, and possible directions for AI in psychiatry and neuroscience. Going forward, the group will set out plans to guide the regulation and development of artificial intelligence (AI) in mental health in Europe.
Group Chair, Dr Paris Alexandros Lalousis (of King’s College, London) said.
“AI is developing at a rapid pace. It offers great opportunities in mental health, but the development of AI also needs to be guided, communicated, and regulated, and this is something that is missing at the moment. The group will set out a framework for understanding the risks and benefits of applying AI in mental health, with recommendations on how we take the field forward. We are the first group to undertake such a challenge. We will map the field, so that Europe-wide resources are available for researchers to understand who is doing what, who is regulating AI, and so on. The aim is to produce a level-headed, expert-driven framework which will deal with the major issues”.
The group is being established by the ECNP, which is Europe’s largest clinical neuroscience organisation, with interests stretching from psychiatry through to drug development, psychology, and neuroscience.
Co-Chair Tim Hahn (from the University of Muenster, Germany) added:
“Our working group comprises 19 experts, from Europe and the US. We have already developed an initial work plan for the next 3 years, and we are now meeting with the wider psychiatric and neuroscience community to feedback and refine our plans.
We are aiming for a coordinated AI framework at a European level. There are many talented people across Europe working in the field, but often they work in silos, and we need to enable communication and interaction. So I think my message would be that whether you see AI through a positive lens or a negative lens, we would like to have you engage with us”.
The working group currently has 19 members from Europe and the US, who undertake research in a variety of fields, for example using basic machine learning to do drug discovery, patient concerns, basic neuroscience, through to building and evaluating chatbots.
Tim Hahn continued, “We have a great variety of expertise, but even so the biggest challenge is going to be to really understand what people are doing right now and where should we insert ourselves as a group, as a network, to provide guidance. We want to establish a structure for this huge amount of information”.
At the Munich congress, the group will also discuss acting as a resource for media enquiries relating to AI in mental health. Immediately after the launch at the ECNP congress, Dr Lalousis will represent the ECNP AI group at a conference at the European Parliament advocating for cross-border validation, regulation, implementation, workforce development and equitable access in precision psychiatry".
Commenting, ECNP President Professor Andreas Reif (Frankfurt, Germany) said:
"AI has manifold implication for mental health, from research to risk factor to therapy; it will certainly change the way we interact with our patients. Therefore, ECNP has decided to make AI in Mental Health one of its priority fields; the AI Network is one of the key structures therein and I am personally very happy that we could put together such an excellent, international team to this end."
Professor David Nutt (Imperial College, London) added:
“I applaud the ECNP for orchestrating this timely and much > needed initiative. I trust it will bring balance and quality control to a rapidly expanding field that so far has little of either”.
Professor Nutt is not involved in this group; this is an independent comment.
The introductory session, “Artificial intelligence in mental health – Progress and perils” takes place at the 39th ECNP Congress in Munich on 13th October. Session details can be found at: can be found here: ecnp.eu/congress2026/programme/searchable/#!sessiondetails/0000183030_0
Background information
You can find information on the AI in Mental Health Network on the ECNP network, see https://www.ecnp.eu/research/ecnp-networks/list-of-ecnp-networks/ai-in-mental-health/
The working group has set the following Network Outcomes
- European evaluation standards for psychiatric AI
- Regulatory-aligned deployment guidance
- Shared methodological infrastructure
- Multisite validation studies
- Training and capacity building
In the medium term, these will translate into the following three-year deliverables:
- Responsible AI position paper
- AI in Mental Health Roadmap
- Living guideline for evaluation and reporting
- Training programmes across ECNP
- Joint European funding initiatives
Subject of Research
People
Hospitals Are Scaling AI, The Workflow Needs Its Own Safety Test – OpEd
October 12, 2026
By Burak Oktenli
Key Takeaways:
Studies and guidelines such as DECIDE-AI and FUTURE-AI show that structured implementation—training, governance, feedback and standardisation—can sharply raise clinician engagement with AI outputs, yet regulatory clearance does not guarantee equal benefit, safety or workflow fit in every hospital.
Local evaluation should therefore define the intended care change, distinguish model effects from organisational ones, monitor human factors and subgroups, test fallbacks for outages, and set clear expansion gates so that institutions can confirm the whole care system improves, not merely that an algorithm performs well.
A model can perform well and still fail in practice if alerts arrive at the wrong time, staff cannot act on them, or the fallback collapses. The next phase of medical AI should evaluate the care system around the model as rigorously as the model itself.
On September 9, eight NHS hospital sites launched a year-long evaluation of artificial intelligence across 104 operating theatres, endoscopy suites and catheterisation laboratories. The programme is operational rather than diagnostic: it is intended to identify bottlenecks, reduce delays and make better use of theatre capacity across more than 20,000 procedures a year. What makes the project interesting beyond Britain is not the promise that AI will improve efficiency. It is the decision to evaluate the technology across multiple hospitals under a common framework before treating early results as a reason to scale further.
That is the right instinct for clinical AI as well. Hospitals are increasingly asked to judge systems whose technical performance may already look impressive. The harder question begins after the model enters a live service. Who responds to its output? What other work is delayed? What happens when the input is missing, the alert arrives during a busy shift, or the system is unavailable? A benchmark can describe a model. It cannot describe the whole care process that forms around it.
A paper published this month in npj Digital Medicine illustrates the point from another direction. In a tertiary neuroradiology department, researchers combined a vendor-neutral AI orchestration platform with structured change management, including workflow standardisation, education, documentation, feedback and governance. The proportion of AI outputs actively verified by radiologists rose from 13.9 percent to 61.1 percent. The study was observational and did not establish improved patient outcomes. Its more modest lesson is still important: implementation design can radically change whether clinicians actually engage with an AI system.
That means the unit of evaluation should not be the model alone. It should be the model inside a local workflow.
The DECIDE-AI reporting guideline made this point before the present wave of deployment. It focuses on early live clinical evaluation of AI decision-support systems and explicitly includes safety and human factors alongside technical performance. The underlying logic is simple. A system that works in a retrospective dataset has crossed only one threshold. Once it influences real decisions, the interaction among the tool, its users and the clinical environment becomes part of the intervention.
International guidance now reinforces the same direction. The FUTURE-AI consensus, developed by experts from 50 countries, treats usability, robustness, traceability and deployment as part of trustworthy healthcare AI rather than as afterthoughts. Regulation is also becoming more specific. In January 2026, the U.S. Food and Drug Administration issued final guidance clarifying when certain clinical decision-support software functions fall outside the medical-device definition. The FDA also maintains a public list of AI-enabled medical devices that have met the applicable premarket requirements for their intended uses.
Those are important regulatory distinctions, but they answer different questions from the ones facing a hospital board. Marketing authorisation can establish that a regulated device met the relevant premarket requirements for a defined intended use. It does not establish that every hospital will obtain the same benefit after installing it, that every subgroup will perform equally well, or that the local staffing model can absorb the work the tool creates. Regulatory status and local effectiveness should not be confused.
A useful local evaluation therefore begins with the intended change in care. A system that sorts a review queue creates a different risk than one that changes a treatment recommendation. Before deployment expands, the institution should identify which decisions the AI is expected to influence, who owns the response, how quickly that response must occur and what happens when the recommendation is ignored, delayed or wrong. Those are clinical and operational design questions, not merely software questions.
The comparison also has to be fair. Current care may be inefficient or understaffed, and refusing new tools can preserve those weaknesses. But an AI rollout often arrives with new training, new attention from managers, redesigned workflows and additional technical support. If outcomes improve, evaluators should be able to distinguish the effect of the model from the effect of reorganising the service around it. Otherwise a hospital may buy the algorithm when the real intervention was better coordination.
Local evidence should also include the patients and conditions that are easiest to lose inside an average. A favourable aggregate result can coexist with weak performance in a smaller group, an unusual presentation or a less common clinical setting. Not every site will have enough cases to estimate every subgroup reliably. That uncertainty should be reported as uncertainty, not converted into reassurance because a point estimate looks acceptable.
Human factors deserve equal weight. An accurate alert that reaches a clinician who has no capacity to act may not improve care. A system that produces too many low-value interruptions can shift workload rather than reduce it. A confusing interface can create delay even when the underlying prediction is correct. Hospitals should therefore observe response time, workload, escalation patterns, overrides and unintended consequences during the evaluation, while protecting patient information and preserving professional accountability.
The fallback deserves a test of its own. Turning off an AI system does not necessarily return a hospital to the workflow it had before. Staff roles may have changed. Schedules may have been reorganised. A vendor service may have become embedded in triage, documentation or prioritisation. Before expansion, the hospital should rehearse how ordinary care continues during an outage or suspension and identify whether the supposed fallback depends on the same infrastructure that has just failed. A fallback that exists only on paper is not resilience.
This does not mean every hospital should run an endless pilot. Evaluation should be proportionate to consequence, novelty and uncertainty. A low-risk administrative tool can justify a lighter process than a system influencing urgent clinical decisions. The point is to define the expansion gate in advance: what benefit must be observed, what harms or workflow failures would pause deployment, what evidence is still missing, and what conditions would justify wider use.
The recent NHS programme offers a useful model of attitude even though its technology is operational rather than clinical decision support. Scale should generate evidence, not substitute for it. A hospital should not ask only whether the AI is impressive enough to buy. It should ask whether the whole arrangement is improving care under the conditions in which its own patients and staff actually work.
The next phase of healthcare AI will not be won by the hospital with the most algorithms. It will be won by institutions that can tell the difference between a model that performs well and a care system that performs better because the model is there. The workflow needs its own safety test.
About Burak Oktenli
Burak Oktenli holds an MBA and a Master of Professional Studies in Applied Intelligence from Georgetown University. His research addresses the governance of authority in autonomous and AI-enabled systems, and his writing has appeared at the Modern War Institute at West Point, RUSI, RealClearDefense, RealClearMarkets, and Geopolitical Monitor. He is the author of Authority Architectures for Autonomous Systems, a ten-volume series on how authority in autonomous systems is delegated, monitored and recovered, at authority-architecture.me.
View all posts by Burak Oktenli →
October 12, 2026
By Burak Oktenli
Key Takeaways:
A year-long NHS evaluation of AI across 104 operating theatres and related suites at eight hospitals prioritises assessing real-world bottlenecks and capacity use before scaling, underscoring that clinical AI must be judged inside local workflows rather than by model benchmarks alone.
Studies and guidelines such as DECIDE-AI and FUTURE-AI show that structured implementation—training, governance, feedback and standardisation—can sharply raise clinician engagement with AI outputs, yet regulatory clearance does not guarantee equal benefit, safety or workflow fit in every hospital.
Local evaluation should therefore define the intended care change, distinguish model effects from organisational ones, monitor human factors and subgroups, test fallbacks for outages, and set clear expansion gates so that institutions can confirm the whole care system improves, not merely that an algorithm performs well.
A model can perform well and still fail in practice if alerts arrive at the wrong time, staff cannot act on them, or the fallback collapses. The next phase of medical AI should evaluate the care system around the model as rigorously as the model itself.
On September 9, eight NHS hospital sites launched a year-long evaluation of artificial intelligence across 104 operating theatres, endoscopy suites and catheterisation laboratories. The programme is operational rather than diagnostic: it is intended to identify bottlenecks, reduce delays and make better use of theatre capacity across more than 20,000 procedures a year. What makes the project interesting beyond Britain is not the promise that AI will improve efficiency. It is the decision to evaluate the technology across multiple hospitals under a common framework before treating early results as a reason to scale further.
That is the right instinct for clinical AI as well. Hospitals are increasingly asked to judge systems whose technical performance may already look impressive. The harder question begins after the model enters a live service. Who responds to its output? What other work is delayed? What happens when the input is missing, the alert arrives during a busy shift, or the system is unavailable? A benchmark can describe a model. It cannot describe the whole care process that forms around it.
A paper published this month in npj Digital Medicine illustrates the point from another direction. In a tertiary neuroradiology department, researchers combined a vendor-neutral AI orchestration platform with structured change management, including workflow standardisation, education, documentation, feedback and governance. The proportion of AI outputs actively verified by radiologists rose from 13.9 percent to 61.1 percent. The study was observational and did not establish improved patient outcomes. Its more modest lesson is still important: implementation design can radically change whether clinicians actually engage with an AI system.
That means the unit of evaluation should not be the model alone. It should be the model inside a local workflow.
The DECIDE-AI reporting guideline made this point before the present wave of deployment. It focuses on early live clinical evaluation of AI decision-support systems and explicitly includes safety and human factors alongside technical performance. The underlying logic is simple. A system that works in a retrospective dataset has crossed only one threshold. Once it influences real decisions, the interaction among the tool, its users and the clinical environment becomes part of the intervention.
International guidance now reinforces the same direction. The FUTURE-AI consensus, developed by experts from 50 countries, treats usability, robustness, traceability and deployment as part of trustworthy healthcare AI rather than as afterthoughts. Regulation is also becoming more specific. In January 2026, the U.S. Food and Drug Administration issued final guidance clarifying when certain clinical decision-support software functions fall outside the medical-device definition. The FDA also maintains a public list of AI-enabled medical devices that have met the applicable premarket requirements for their intended uses.
Those are important regulatory distinctions, but they answer different questions from the ones facing a hospital board. Marketing authorisation can establish that a regulated device met the relevant premarket requirements for a defined intended use. It does not establish that every hospital will obtain the same benefit after installing it, that every subgroup will perform equally well, or that the local staffing model can absorb the work the tool creates. Regulatory status and local effectiveness should not be confused.
A useful local evaluation therefore begins with the intended change in care. A system that sorts a review queue creates a different risk than one that changes a treatment recommendation. Before deployment expands, the institution should identify which decisions the AI is expected to influence, who owns the response, how quickly that response must occur and what happens when the recommendation is ignored, delayed or wrong. Those are clinical and operational design questions, not merely software questions.
The comparison also has to be fair. Current care may be inefficient or understaffed, and refusing new tools can preserve those weaknesses. But an AI rollout often arrives with new training, new attention from managers, redesigned workflows and additional technical support. If outcomes improve, evaluators should be able to distinguish the effect of the model from the effect of reorganising the service around it. Otherwise a hospital may buy the algorithm when the real intervention was better coordination.
Local evidence should also include the patients and conditions that are easiest to lose inside an average. A favourable aggregate result can coexist with weak performance in a smaller group, an unusual presentation or a less common clinical setting. Not every site will have enough cases to estimate every subgroup reliably. That uncertainty should be reported as uncertainty, not converted into reassurance because a point estimate looks acceptable.
Human factors deserve equal weight. An accurate alert that reaches a clinician who has no capacity to act may not improve care. A system that produces too many low-value interruptions can shift workload rather than reduce it. A confusing interface can create delay even when the underlying prediction is correct. Hospitals should therefore observe response time, workload, escalation patterns, overrides and unintended consequences during the evaluation, while protecting patient information and preserving professional accountability.
The fallback deserves a test of its own. Turning off an AI system does not necessarily return a hospital to the workflow it had before. Staff roles may have changed. Schedules may have been reorganised. A vendor service may have become embedded in triage, documentation or prioritisation. Before expansion, the hospital should rehearse how ordinary care continues during an outage or suspension and identify whether the supposed fallback depends on the same infrastructure that has just failed. A fallback that exists only on paper is not resilience.
This does not mean every hospital should run an endless pilot. Evaluation should be proportionate to consequence, novelty and uncertainty. A low-risk administrative tool can justify a lighter process than a system influencing urgent clinical decisions. The point is to define the expansion gate in advance: what benefit must be observed, what harms or workflow failures would pause deployment, what evidence is still missing, and what conditions would justify wider use.
The recent NHS programme offers a useful model of attitude even though its technology is operational rather than clinical decision support. Scale should generate evidence, not substitute for it. A hospital should not ask only whether the AI is impressive enough to buy. It should ask whether the whole arrangement is improving care under the conditions in which its own patients and staff actually work.
The next phase of healthcare AI will not be won by the hospital with the most algorithms. It will be won by institutions that can tell the difference between a model that performs well and a care system that performs better because the model is there. The workflow needs its own safety test.
About Burak Oktenli
Burak Oktenli holds an MBA and a Master of Professional Studies in Applied Intelligence from Georgetown University. His research addresses the governance of authority in autonomous and AI-enabled systems, and his writing has appeared at the Modern War Institute at West Point, RUSI, RealClearDefense, RealClearMarkets, and Geopolitical Monitor. He is the author of Authority Architectures for Autonomous Systems, a ten-volume series on how authority in autonomous systems is delegated, monitored and recovered, at authority-architecture.me.
View all posts by Burak Oktenli →
The AI Race As A Dissipative Structure – Analysis
Image: Grok
October 12, 2026
By Dr. Yashwant Singh
Key Takeaways:
Empirical markers (frontier training power doubling yearly, hyperscaler capex heading past $700 billion, data-centre demand rising 17–50 percent) show efficiency gains are reinvested into still greater throughput (Jevons-like behaviour), while grid-connection delays, turbine shortages and soaring capacity prices reveal that the “sink” is now the binding constraint.
Because such structures are robust to small shocks yet brittle to loss of flux, policy should prioritise managing energy flows, strengthening sinks and couplings, and building systems that remain organised under reduced gradients instead of merely picking winners.
The dominant metaphor for artificial intelligence is the race: lanes, leaders, laggards, a finish line called AGI (artificial general intelligence) or “dominance.” The metaphor flatters our habit of reading history as a contest between agents. It also hides what is physically happening. A race presumes a fixed track and runners who can in principle stop and be counted. What we are actually watching is closer to a boiling pot: a rapidly growing flow of energy, capital, silicon, and data, passing through a set of institutions that are reorganising themselves around that flow.
Ilya Prigogine gave a name to this kind of object. In his work on non-equilibrium thermodynamics, for which he received the 1977 Nobel Prize in Chemistry, a dissipative structure is an ordered pattern that exists only because a system is open, held far from equilibrium, and continuously exporting entropy to its surroundings. The standard examples are Bénard convection cells, oscillating chemical reactions, and living organisms. Stop the flux and the structure vanishes. The order is not a possession; it is a rate.
This essay takes that idea seriously as the organising lens for the AI buildout. The claim is not that AI companies are literally Bénard cells. It is that the formal features of dissipative structures give a sharper account of what is driving, constraining, and likely to break this system than the vocabulary of competition does.
The four signatures
Four features define a dissipative structure. Each has an AI analogue.
An open system with a gradient. Bénard cells need a temperature difference between a heated floor and a cool ceiling. The AI system needs a gradient of usable energy and cheap capital. Without a continuous supply of both, nothing on the stack runs.
Throughput that sustains order. The structure is maintained by the flow, not by stored capital. In AI, a model is not an asset in the way a bridge is. It is a standing pattern in a river of electricity, GPU-hours, and fresh data, and it decays in relevance as the river moves on.
Nonlinear feedback. Self-organisation arises when fluctuations are amplified rather than damped. In AI, better models attract users, users attract revenue and capital, capital buys compute, compute yields better models. This is autocatalysis in the chemical sense.
Entropy export. Every ordered structure must dump disorder somewhere. For AI that means heat, carbon, grid instability, and water. The sink, not only the source, bounds how large the structure can become.
The empirical signature of a far-from-equilibrium system
The data now being published look less like an industry’s growth curve and more like the measurements of a system pushed ever further from equilibrium.
The gradient is steepening. Epoch AI estimates that training compute for frontier language models has grown roughly five-fold per year since 2020, with the power draw of frontier training runs growing around two-fold (2.2x) per year and the largest runs now exceeding 100 MW. The same body of work projects that continuing this trajectory to 2030 would require clusters drawing anywhere between 4 to 16 gigawatts (GW). Capital is following: the four largest US hyperscalers spent a record of roughly $410 billion on capital expenditure in 2025, and guidance as of mid-2026 points to about $725 billion this year, with banks such as Bank of America and Evercore projecting more than $1 trillion in 2027. Hyperscaler capex projections have surged to $5.3 trillion through 2030 (Goldman Sachs), against a broader $7.6 trillion AI infrastructure investment baseline for 2026-2031.
The flux is visible in the grid. The International Energy Agency reported in April 2026 that data-centre electricity demand grew 17% in 2025, against 3% growth in global electricity demand overall, and that demand from AI-focused facilities grew considerably faster, at about 50%. Its central projection is that data-centre consumption roughly doubles to around 945 terawatt-hours (TWh) by 2030.
The feedback is explicit. Microsoft reported its AI business at an annual revenue run rate above $37 billion, up 123% year on year, and told investors it would remain capacity constrained through 2026 even at that level of spending. Mark Zuckerberg’s stated preference to build ahead of demand rather than constrain research is, in thermodynamic language, a decision to hold the system far from equilibrium on purpose.
These are not the numbers of a market clearing at an equilibrium price. They are the numbers of a structure still being fed by a steepening gradient.
Why efficiency does not relax the system
A naive reading says efficiency gains should ease the energy problem. Epoch’s data show hardware efficiency, lower-precision arithmetic, and longer training runs together cutting power requirements per unit of compute by roughly a factor of two each year, and pre-training compute efficiency itself improving about three-fold annually. Yet total power still doubles.
Dissipative systems behave this way. Efficiency lowers the cost of maintaining a given level of organisation, and the saving is reinvested into more throughput. This is Jevons’s old observation about coal, rediscovered in thermodynamic dress. The IEA states the pattern plainly: energy per AI task is falling, and total consumption is still rising. In a system driven by positive feedback, efficiency is a fuel, not a brake.
The lens, therefore, predicts something specific: algorithmic breakthroughs that make intelligence cheaper, of the sort associated with DeepSeek in early 2025, will not reduce the buildout. They will widen the set of tasks for which it pays to run the structure, and push the system further from equilibrium.
The sink is the binding constraint
In the physics, the decisive limit on a convection cell is often not the heat source but the rate at which heat can be carried away. The AI system is now meeting its sink limits.
The IEA identifies shortages of gas turbines, transformers, and advanced chips, together with grid connection delays and permitting, as what is slowing data-centre development. Reporting from the US mid-Atlantic describes waits of up to seven years for grid connections in parts of Virginia, and an emergency order this summer allowing the grid operator PJM to push large data centres onto their own backup generators on short notice during a heat wave. PJM’s capacity auction prices rose from about $28.92 to about $329 per megawatt-day across three consecutive auctions, according to analysis relayed by industry commentators. These are what a system looks like when it has outrun the capacity of its environment to absorb its output.
The IEA’s own phrasing is instructive. AI, it says, is still an energy taker but is becoming an energy maker, pulling forward next-generation nuclear, flexible data-centre loads, and long-duration storage. That is the other thing dissipative structures do: they reshape their environment to sustain the flux. Convection cells organise the fluid. AI is reorganising the grid.
Geopolitics as gradient access
Seen this way, the US-China contest changes character. The scarce resource is not only the best chip or the best model; it is the capacity to sustain the largest, steadiest flux of electricity through computation without destabilising the host system.
On that axis the figures are stark, though sources vary in their exact numbers. Estimates for China’s 2025 additions range from roughly 430 GW of wind and solar alone to about 570 GW of total capacity, against something like 40 to 63 GW for the United States. BloombergNEF projects that China will add more than six times as much generation capacity as the US over the next five years. The US, for its part, retains the larger installed base of data centres (Stanford’s AI Index counts about 5,427 versus about 450 in China) and leading positions in advanced chips and frontier models.
The dissipative lens cuts both ways here, and it is worth being disciplined about it. A larger gradient does not guarantee a better structure. Structure depends on how flux is channelled: chips, talent, software ecosystems, and institutions that can convert power into capability. Chinese grids are also fragmented regionally, and commentators note that a planning system able to approve a national project cannot easily stop a useless one, so some capacity may be stranded. What the lens predicts is not that China wins, but that the contest shifts from who has the best design to who can best couple a design to a sustained energy flow.
The fragility of order held by flux
The most important implication of dissipative structure theory is also the least comfortable: these structures are robust to small perturbations and brittle to loss of throughput. They do not decline gracefully. Past a critical threshold they reorganise or collapse.
Several indicators suggest the financial form of the AI structure is already sensitive to its flux. Analysis of Microsoft’s filings notes that roughly two thirds of one quarter’s capital spending went to short-lived assets, mainly GPUs and CPUs, which means the structure must be refed continuously just to stay in place. One widely read analysis points out that Meta had around $12 billion of quarterly free cash flow against $145 billion of planned annual capex, and that the other three hyperscalers’ free cash flow fell sharply in the same period. Meta’s shares dropped about 9% on raising guidance, which could be read as the first fluctuation the system has had trouble absorbing. A June 2026 market note observed that data-centre lenders were treating hyperscaler commitments as nearly sovereign-grade (“in ways that were largely absent before 2024”), and warned that a pause by even one major player could reprice credit abruptly.
None of this proves a crash. It does identify where a crash would propagate: through the dependence of every layer, from lenders to turbine makers to chip fabs, on the continuity of one flux. Bénard cells do not wobble at the margin; they flip from one pattern to another at a bifurcation point. The right question is not whether growth continues, but which regime the system moves to when the gradient stops steepening.
Where the analogy breaks
Intellectual honesty requires marking the limits. Prigogine’s formalism applies rigorously to physical and chemical systems, and transplanting it to firms and states is analogy, not derivation. Firms make choices; convection cells do not. Capital can be created by policy in ways that heat cannot. Jeremy England’s “dissipative adaptation” thesis, which suggests matter tends to organise toward structures that absorb and dissipate energy efficiently, remains contested and should not be invoked as destiny. And the lens says little about what the structure is for. A system can be perfectly dissipative and still serve no one’s purposes.
The analogy earns its keep only where it generates testable predictions: that efficiency gains raise rather than lower total throughput, that binding constraints migrate from sources to sinks, that fragility concentrates at points of flux dependence, and that competition increasingly takes the form of competition for gradient access. So far, each has some empirical support.
What follows
If the AI system is a dissipative structure, several strategic conclusions change.
First, policy should target flux management rather than leader selection. Export controls on chips matter, but they act on one channel of a system whose binding limits are increasingly electrical.
Second, the highest-leverage investments may be in sinks and couplings: transmission, interconnection reform, storage, and flexible loads that let data centres yield to the grid at peak times. The IEA’s call for cooperation between governments, grid operators, and technology firms to make demand flexible is, in this reading, the policy equivalent of improving heat removal.
Third, resilience should be defined as the ability to remain organised under a reduced gradient. Systems with modest throughput requirements per unit of capability, such as efficient open-weight models run on cheap local power, may prove more robust than those that depend on ever-steeper flux.
Finally, the lens recommends humility about the word race. A race ends. A dissipative structure persists only while fed, reorganises when conditions change, and exports its disorder to whoever lives downstream. The more useful question for the coming decade is not who is winning, but what kind of order we are asking the world’s energy systems to sustain, at what rate, and with what cost to everyone outside the cell.
This article was published by TCSS
About Dr. Yashwant Singh
Dr. Yashwant Singh is a sociologist, recently served as an Assistant Professor in the Department of Sociology at GITAM (Deemed to be) University, Bengaluru, India. He holds an M.Phil. in Sociology from the University of Delhi and a Ph.D. in Sociology from the University of Hyderabad, India. His research interests include urban sociology, the sociology of development and geopolitics. His writings have appeared on several digital platforms, including Across Voices, Modern Diplomacy, Geostrategic Media, South Asia Journal, World Geostrategic Insights and IA-Forum
View all posts by Dr. Yashwant Singh →
Image: Grok
October 12, 2026
By Dr. Yashwant Singh
Key Takeaways:
The essay reframes the AI build-out as a Prigogine-style dissipative structure—an ordered pattern sustained only by continuous flows of energy, capital, silicon and data—rather than a conventional race with a finish line.
Empirical markers (frontier training power doubling yearly, hyperscaler capex heading past $700 billion, data-centre demand rising 17–50 percent) show efficiency gains are reinvested into still greater throughput (Jevons-like behaviour), while grid-connection delays, turbine shortages and soaring capacity prices reveal that the “sink” is now the binding constraint.
Because such structures are robust to small shocks yet brittle to loss of flux, policy should prioritise managing energy flows, strengthening sinks and couplings, and building systems that remain organised under reduced gradients instead of merely picking winners.
The dominant metaphor for artificial intelligence is the race: lanes, leaders, laggards, a finish line called AGI (artificial general intelligence) or “dominance.” The metaphor flatters our habit of reading history as a contest between agents. It also hides what is physically happening. A race presumes a fixed track and runners who can in principle stop and be counted. What we are actually watching is closer to a boiling pot: a rapidly growing flow of energy, capital, silicon, and data, passing through a set of institutions that are reorganising themselves around that flow.
Ilya Prigogine gave a name to this kind of object. In his work on non-equilibrium thermodynamics, for which he received the 1977 Nobel Prize in Chemistry, a dissipative structure is an ordered pattern that exists only because a system is open, held far from equilibrium, and continuously exporting entropy to its surroundings. The standard examples are Bénard convection cells, oscillating chemical reactions, and living organisms. Stop the flux and the structure vanishes. The order is not a possession; it is a rate.
This essay takes that idea seriously as the organising lens for the AI buildout. The claim is not that AI companies are literally Bénard cells. It is that the formal features of dissipative structures give a sharper account of what is driving, constraining, and likely to break this system than the vocabulary of competition does.
The four signatures
Four features define a dissipative structure. Each has an AI analogue.
An open system with a gradient. Bénard cells need a temperature difference between a heated floor and a cool ceiling. The AI system needs a gradient of usable energy and cheap capital. Without a continuous supply of both, nothing on the stack runs.
Throughput that sustains order. The structure is maintained by the flow, not by stored capital. In AI, a model is not an asset in the way a bridge is. It is a standing pattern in a river of electricity, GPU-hours, and fresh data, and it decays in relevance as the river moves on.
Nonlinear feedback. Self-organisation arises when fluctuations are amplified rather than damped. In AI, better models attract users, users attract revenue and capital, capital buys compute, compute yields better models. This is autocatalysis in the chemical sense.
Entropy export. Every ordered structure must dump disorder somewhere. For AI that means heat, carbon, grid instability, and water. The sink, not only the source, bounds how large the structure can become.
The empirical signature of a far-from-equilibrium system
The data now being published look less like an industry’s growth curve and more like the measurements of a system pushed ever further from equilibrium.
The gradient is steepening. Epoch AI estimates that training compute for frontier language models has grown roughly five-fold per year since 2020, with the power draw of frontier training runs growing around two-fold (2.2x) per year and the largest runs now exceeding 100 MW. The same body of work projects that continuing this trajectory to 2030 would require clusters drawing anywhere between 4 to 16 gigawatts (GW). Capital is following: the four largest US hyperscalers spent a record of roughly $410 billion on capital expenditure in 2025, and guidance as of mid-2026 points to about $725 billion this year, with banks such as Bank of America and Evercore projecting more than $1 trillion in 2027. Hyperscaler capex projections have surged to $5.3 trillion through 2030 (Goldman Sachs), against a broader $7.6 trillion AI infrastructure investment baseline for 2026-2031.
The flux is visible in the grid. The International Energy Agency reported in April 2026 that data-centre electricity demand grew 17% in 2025, against 3% growth in global electricity demand overall, and that demand from AI-focused facilities grew considerably faster, at about 50%. Its central projection is that data-centre consumption roughly doubles to around 945 terawatt-hours (TWh) by 2030.
The feedback is explicit. Microsoft reported its AI business at an annual revenue run rate above $37 billion, up 123% year on year, and told investors it would remain capacity constrained through 2026 even at that level of spending. Mark Zuckerberg’s stated preference to build ahead of demand rather than constrain research is, in thermodynamic language, a decision to hold the system far from equilibrium on purpose.
These are not the numbers of a market clearing at an equilibrium price. They are the numbers of a structure still being fed by a steepening gradient.
Why efficiency does not relax the system
A naive reading says efficiency gains should ease the energy problem. Epoch’s data show hardware efficiency, lower-precision arithmetic, and longer training runs together cutting power requirements per unit of compute by roughly a factor of two each year, and pre-training compute efficiency itself improving about three-fold annually. Yet total power still doubles.
Dissipative systems behave this way. Efficiency lowers the cost of maintaining a given level of organisation, and the saving is reinvested into more throughput. This is Jevons’s old observation about coal, rediscovered in thermodynamic dress. The IEA states the pattern plainly: energy per AI task is falling, and total consumption is still rising. In a system driven by positive feedback, efficiency is a fuel, not a brake.
The lens, therefore, predicts something specific: algorithmic breakthroughs that make intelligence cheaper, of the sort associated with DeepSeek in early 2025, will not reduce the buildout. They will widen the set of tasks for which it pays to run the structure, and push the system further from equilibrium.
The sink is the binding constraint
In the physics, the decisive limit on a convection cell is often not the heat source but the rate at which heat can be carried away. The AI system is now meeting its sink limits.
The IEA identifies shortages of gas turbines, transformers, and advanced chips, together with grid connection delays and permitting, as what is slowing data-centre development. Reporting from the US mid-Atlantic describes waits of up to seven years for grid connections in parts of Virginia, and an emergency order this summer allowing the grid operator PJM to push large data centres onto their own backup generators on short notice during a heat wave. PJM’s capacity auction prices rose from about $28.92 to about $329 per megawatt-day across three consecutive auctions, according to analysis relayed by industry commentators. These are what a system looks like when it has outrun the capacity of its environment to absorb its output.
The IEA’s own phrasing is instructive. AI, it says, is still an energy taker but is becoming an energy maker, pulling forward next-generation nuclear, flexible data-centre loads, and long-duration storage. That is the other thing dissipative structures do: they reshape their environment to sustain the flux. Convection cells organise the fluid. AI is reorganising the grid.
Geopolitics as gradient access
Seen this way, the US-China contest changes character. The scarce resource is not only the best chip or the best model; it is the capacity to sustain the largest, steadiest flux of electricity through computation without destabilising the host system.
On that axis the figures are stark, though sources vary in their exact numbers. Estimates for China’s 2025 additions range from roughly 430 GW of wind and solar alone to about 570 GW of total capacity, against something like 40 to 63 GW for the United States. BloombergNEF projects that China will add more than six times as much generation capacity as the US over the next five years. The US, for its part, retains the larger installed base of data centres (Stanford’s AI Index counts about 5,427 versus about 450 in China) and leading positions in advanced chips and frontier models.
The dissipative lens cuts both ways here, and it is worth being disciplined about it. A larger gradient does not guarantee a better structure. Structure depends on how flux is channelled: chips, talent, software ecosystems, and institutions that can convert power into capability. Chinese grids are also fragmented regionally, and commentators note that a planning system able to approve a national project cannot easily stop a useless one, so some capacity may be stranded. What the lens predicts is not that China wins, but that the contest shifts from who has the best design to who can best couple a design to a sustained energy flow.
The fragility of order held by flux
The most important implication of dissipative structure theory is also the least comfortable: these structures are robust to small perturbations and brittle to loss of throughput. They do not decline gracefully. Past a critical threshold they reorganise or collapse.
Several indicators suggest the financial form of the AI structure is already sensitive to its flux. Analysis of Microsoft’s filings notes that roughly two thirds of one quarter’s capital spending went to short-lived assets, mainly GPUs and CPUs, which means the structure must be refed continuously just to stay in place. One widely read analysis points out that Meta had around $12 billion of quarterly free cash flow against $145 billion of planned annual capex, and that the other three hyperscalers’ free cash flow fell sharply in the same period. Meta’s shares dropped about 9% on raising guidance, which could be read as the first fluctuation the system has had trouble absorbing. A June 2026 market note observed that data-centre lenders were treating hyperscaler commitments as nearly sovereign-grade (“in ways that were largely absent before 2024”), and warned that a pause by even one major player could reprice credit abruptly.
None of this proves a crash. It does identify where a crash would propagate: through the dependence of every layer, from lenders to turbine makers to chip fabs, on the continuity of one flux. Bénard cells do not wobble at the margin; they flip from one pattern to another at a bifurcation point. The right question is not whether growth continues, but which regime the system moves to when the gradient stops steepening.
Where the analogy breaks
Intellectual honesty requires marking the limits. Prigogine’s formalism applies rigorously to physical and chemical systems, and transplanting it to firms and states is analogy, not derivation. Firms make choices; convection cells do not. Capital can be created by policy in ways that heat cannot. Jeremy England’s “dissipative adaptation” thesis, which suggests matter tends to organise toward structures that absorb and dissipate energy efficiently, remains contested and should not be invoked as destiny. And the lens says little about what the structure is for. A system can be perfectly dissipative and still serve no one’s purposes.
The analogy earns its keep only where it generates testable predictions: that efficiency gains raise rather than lower total throughput, that binding constraints migrate from sources to sinks, that fragility concentrates at points of flux dependence, and that competition increasingly takes the form of competition for gradient access. So far, each has some empirical support.
What follows
If the AI system is a dissipative structure, several strategic conclusions change.
First, policy should target flux management rather than leader selection. Export controls on chips matter, but they act on one channel of a system whose binding limits are increasingly electrical.
Second, the highest-leverage investments may be in sinks and couplings: transmission, interconnection reform, storage, and flexible loads that let data centres yield to the grid at peak times. The IEA’s call for cooperation between governments, grid operators, and technology firms to make demand flexible is, in this reading, the policy equivalent of improving heat removal.
Third, resilience should be defined as the ability to remain organised under a reduced gradient. Systems with modest throughput requirements per unit of capability, such as efficient open-weight models run on cheap local power, may prove more robust than those that depend on ever-steeper flux.
Finally, the lens recommends humility about the word race. A race ends. A dissipative structure persists only while fed, reorganises when conditions change, and exports its disorder to whoever lives downstream. The more useful question for the coming decade is not who is winning, but what kind of order we are asking the world’s energy systems to sustain, at what rate, and with what cost to everyone outside the cell.
This article was published by TCSS
About Dr. Yashwant Singh
Dr. Yashwant Singh is a sociologist, recently served as an Assistant Professor in the Department of Sociology at GITAM (Deemed to be) University, Bengaluru, India. He holds an M.Phil. in Sociology from the University of Delhi and a Ph.D. in Sociology from the University of Hyderabad, India. His research interests include urban sociology, the sociology of development and geopolitics. His writings have appeared on several digital platforms, including Across Voices, Modern Diplomacy, Geostrategic Media, South Asia Journal, World Geostrategic Insights and IA-Forum
View all posts by Dr. Yashwant Singh →
