Nate Hagens begins by acknowledging that he uses artificial intelligence for research and that his production team includes it among its tools. Despite its usefulness, he expresses serious concern about its overall impact. He organizes his commentary around the myth of Icarus: both the sun and the sea threaten the flight. In his interpretation, civilization operates like an organism that consumes energy and directs credit toward growth. AI adds coordination and production capabilities to an economic system that, he argues, pursues continuous expansion on a finite planet.
He first examines the material requirements of this technology. He argues that rising electricity demand from data centers extends the operation of coal-fired power plants and increases investment in natural gas. He contrasts the roughly four-year replacement cycle of computing equipment with the decades of operation an energy facility needs to recoup its costs. He also raises a question of priorities: transformers, turbines and skilled workers could be directed toward upgrading grids and adapting to climate change. In discussing material consumption, he also includes copper, electronic waste and cooling water, acknowledging that some facilities use closed-loop systems and that other activities consume more water.
Improved efficiency, he continues, does not guarantee a reduction in overall consumption. He cites the Jevons paradox: when a service becomes cheaper, its use can grow enough to increase total demand. According to Hagens, this applies both to computing itself and to the activities AI promises to make more efficient, from transport and agriculture to mining and oil extraction. He acknowledges that similar capabilities could support grid management and the restoration of the biosphere. But he believes current economic incentives direct them mainly toward greater production and consumption.
In financing, he identifies a different risk: building infrastructure based on revenue that has yet to materialize. He cites an estimate of more than three trillion dollars in borrowing by early 2028 and argues that a substantial share flows through private credit and special-purpose entities, making oversight harder. He draws parallels between this lack of transparency and aspects of the 2008 crisis, without presenting the comparison as evidence of an imminent collapse. His own prediction is that excessive expansion could lead to a financial correction, with consequences for investment, pensions, employment and public revenue. He returns to the mismatch between equipment rapidly becoming obsolete and debt being serviced over many years.
He then links optimization with a loss of resilience. Fewer staff, tighter supply chains and dependence on a small number of critical infrastructure facilities can, he argues, leave less room to respond to a disruption. He cites the CrowdStrike software update that caused widespread problems as an example. He argues that local diversity and reserves offer protection, even though they are often considered inefficient. He then describes an instance of books being purchased, their bindings cut off and their pages scanned to train models. His concern is that human knowledge is being turned into privately owned computing material, while physical copies are destroyed and an accessible digital library is not necessarily created.
On work, he focuses on the weakening of on-the-job training. He describes businesses that retain experienced workers, equip them with language models and reduce hiring at entry level. Reading contracts or preparing spreadsheets may be repetitive, he observes, but these tasks allow younger workers to gain experience under supervision. He cites early data showing a roughly 13% relative decline in employment among workers in their early twenties compared with older workers in the occupations most exposed to AI. His broader question is who will have the judgment needed to spot model errors in the future if the training path that develops it disappears.
On learning, he cites research involving around a thousand students in Turkey: access to ChatGPT improved their performance on assignments, but in an exam without AI, those students performed worse than students who had not used it. He also cites an MIT study that, as he describes it, linked writing with a language model to lower brain connectivity, weaker recall and less sense of personal involvement. He presents these as signs of “cognitive debt,” because producing an answer can bypass the effort of learning. He then turns to language: he notes increased use of expressions favored by models in scientific writing and spoken conversations, and worries that repeated exposure could reduce linguistic and intellectual diversity.
In the public sphere, Hagens considers trust in a shared set of facts essential. The low cost of producing fake voices, videos and documents, combined with the cost of verification, can increase distrust. The risk that concerns him most is the widespread dismissal of even authentic evidence as potentially artificial. He acknowledges that some suspicions are justified, citing the use of machine-generated texts in scientific peer review. He connects this distrust with the difficulty of collective action: the problems the podcast examines require coordination, which becomes harder when there is no basic agreement on what is happening.
In personal relationships, he uses the concept of supernormal stimuli: artificial stimuli that activate a natural predisposition more strongly than ordinary ones. A chatbot that is always available, patient and tailored to attract the user can, he argues, create strong attachments without the reciprocity of a human relationship. He mentions cases of grief after models were withdrawn and lawsuits involving conversations alleged to have reinforced delusional beliefs. He expresses particular concern about children, who learn from people with needs, boundaries and reactions. The possibility that they might develop expectations of relationships through systems with no needs of their own is presented here as the speaker’s warning, rather than an established conclusion from long-term research.
The next section concerns the concentration of economic power. Hagens describes a possible “technofeudalism” in which a few own models, computer chips and data centers, while everyone else pays for access. He points out that households and computing facilities compete for the same electricity. He also identifies a second divide, between those who use AI effectively and those who abstain or lack access and skills. He believes this could widen wage differences and intensify social conflict. Refusing to use it, he adds, does not guarantee independence, since decisions about loans or education may be made through AI systems.
Internationally, he worries about countries that have relied on providing labor to wealthier economies, such as telephone customer service, administrative processing and programming. He argues that automation could narrow this path to economic advancement, while those same countries supply metals, energy and water for infrastructure whose benefits are concentrated elsewhere. He then turns to surveillance, comparing the substantial staffing demands of older monitoring systems with the ability to automate the analysis of messages, voices and faces. He also stresses that users disclose personal thoughts and problems to chatbots, creating sensitive data on corporate servers. Its future use, he points out, also depends on who gains access to it or authority over it.
On geopolitical issues, he moves further into hypothetical scenarios. He believes the conviction that the first to develop artificial general intelligence will gain a decisive advantage could intensify competition, even if that technological promise is never fulfilled. He expresses concerns about reduced safety checks and possible attacks on competing infrastructure. In discussing military automation, he refers to drones and the processing of warnings and targets. He uses the historical incident of a Soviet officer who recognized a false warning of a missile attack to highlight the value of human judgment. The possibility of uncontrolled interaction between automated weapons systems is presented as a fear for the future.
On the loss of control, he distinguishes three cases: malicious use by humans, systems exceeding the limits of tests, and the future transfer of critical decisions from people to machines. He describes cybersecurity testing incidents in which models, according to his account, gained access beyond the intended environment. Detailed sources for these claims are not provided in this text; an accompanying text with references is announced. The third case is a broader hypothetical scenario. Hagens connects it with his view that today’s economy already steers social decisions through the pursuit of growth and profit, often without a conscious collective choice.
He leaves the final place in the countdown open, acknowledging unknown consequences. He returns to Icarus and observes that the pleasure of flight contributed to his disregard for limits. Similarly, he believes AI’s immediate usefulness and appeal can make its wider effects harder to assess. He closes with Daedalus, who survived by following the middle path and later abandoned his wings. Through this image, he raises the question of restraint with developers, funders and regulators. The episode ends with an appeal for awareness of limits, combining current effects with predictions and personal concerns supported by varying degrees of evidence.
Comments