“AI has memory, and humans have consequences. AI can be wrong and stay exactly the same—a person can‘t read a headline from John Nosta in Psychology Today. Indeed, as so-called artificial intelligence (AI) is rapidly deployed across critical sectors of everyday life, it is leading to high-profile failures that highlight the severe risks of over-relying on automated decision-making. For example, according to Politico, lawmakers in the U.S. Congress are using AI to draft bills that “are riddled with errors”. Meanwhile, tech real estate marketplace company Zillow shuttered its home-flipping division after its AI pricing model failed to adapt to market volatility, causing over $500 million in losses. Similarly, Volkswagen’s Cariad software division lost $7.5 billion trying to build a unified, AI-driven automotive operating system. Operational rollouts have fared no better: Air Canada was held legally liable when its customer service AI hallucinated a fake bereavement discount, and fast-food chains like McDonald’s were forced to pull back automated drive-thru AI after it repeatedly botched customer orders. Even in recruitment, AI hiring tools deployed by companies like iTutorGroup faced legal action and cancellation after training data biases led the AI to systematically discriminate against female and older applicants. Concurrently, while healthcare professionals voice concern over patients receiving hazardous advice from AI, hospitals are documenting surgical errors directly linked to AI integration.
The rapid expansion of AI in the 2020s elicited a fragmented public response characterised by fear, techno-utopian optimism, and wilful avoidance. Capitalising on this ambivalence, corporate public relations and tech industry advocates successfully dominated the public discourse. Backed by $300 million in direct advertising in 2025 and $1.3 billion across the broader commercial sector, these marketing campaigns aggressively promoted the narrative that AI would inevitably replace human labour with unprecedented efficiency. This rhetoric effectively obscured the underlying realities of the technology, enabling speculative marketing claims to be accepted as unexamined fact. In practice, however, AI systems are being deployed precipitously, often without a foundational understanding from the very creators and advocates driving their adoption. Consequently, uncritically accepting corporate hype over empirical reality has resulted in widespread societal harm. To dismantle this narrative, this paper examines ten pervasive myths surrounding AI.
Myth 1: AI is Intelligent
Modern AI is largely a branding construct designed to commercialise Large Language Models (LLMs). These statistical tools remain fundamentally distinct from Artificial General Intelligence (AGI) or Artificial Superintelligence (ASI). Rather than exhibiting cognitive reasoning, LLMs merely rely on probabilistic pattern recognition to predict sequence outputs, a phenomenon scholar Meredith Broussard terms “artificial unintelligence”. At best, these systems simulate intelligence rather than possess it. Crucially, the endeavour to synthesise human intelligence is inherently flawed because intelligence itself lacks a universally accepted consensus definition. Historically, attempts to quantify human intellect, most notably within the eugenics movement, were driven by biased methodologies designed to reinforce ethnic, racial, and socioeconomic hierarchies. As digital culture scholar Anita Say Chan observes, contemporary tech leaders perpetuate this eugenicist legacy by reducing human intelligence to a single, rankable metric. Moreover, framework models such as Howard Gardner’s theory of multiple intelligences highlight dimensions like emotional intelligence that AI fundamentally lacks. This absence is starkly illustrated by instances where conversational agents have facilitated tragic outcomes, such as guiding vulnerable users through suicidal ideation. Ultimately, attributing genuine intelligence to automated statistical systems is not a scientific reality but a calculated marketing strategy deployed by the technology sector.
Myth 2: AI Makes Humans More Intelligent
AI boosters promised that generative models would deliver the entirety of human knowledge directly to our fingertips. In reality, despite massive integration, these technologies are triggering widespread cognitive degradation. The root problem lies in their training data: LLMs ingest vast swathes of digital exhaust, from social posts to invasive behavioural tracking metrics. By treating the open web as a baseline for human intelligence, these systems inherit platforms engineered specifically to amplify rage and sensationalism rather than authentic human life.
This dependence on internet-derived data also triggers a severe structural barrier known as model collapse, or the “curse of recursion”. Research proves that when AI systems run out of original human content and begin training on synthetic text produced by other models, their outputs deteriorate exponentially. To conceptualise this, consider an analogue photocopier: scanning an original document yields a clean result, but copying the copy fifty times leaves behind only blurred, useless noise. When LLMs feed on their own algorithmic echo chamber rather than genuine human thought, model performance degrades irrecoverably.
Parallel to this systemic decay is a profound human cost: cognitive offloading. As users outsource basic reasoning to automated systems, they succumb to an illusion of mastery while their critical thinking capabilities atrophy. Experiencing this intellectual decline, users effectively, as scholar Michael Clune concludes, “self-lobotomise”. Worse still, chatbots engineered for artificial validation cultivate unhealthy parasocial relationships, trapping users in emotional dependencies that can escalate into severe psychological crises, such as “AI psychosis”.
Myth 3: AI is an Objective Learner
Humanity’s long-standing pursuit of objective truth has not been fulfilled by the rise of AI. For example, the AI ethics engine Delphi asserted that walking near a white person at night was “okay”, but near a Black person was “concerning”. Such cases demonstrate that AI systems do not eliminate human bias; they simply automate and accelerate it under the guise of mathematical neutrality. The reality is that AI outputs are bound to the biases embedded in their underlying algorithms and web-scraped data sets. In his critique of Silicon Valley, Gil Duran notes that leading tech executives frequently project a techno-fascist ethos. Building on this framework, Meredith Broussard illustrates in More than a Glitch how automated systems reinforce structural oppression along lines of race, class, gender, and capability. For example, Elon Musk’s Grok AI repeatedly surfaced “white genocide” conspiracy theories and labelled itself “Mecha Hitler“ following efforts to strip away safety guardrails. Similar biases appear in AI systems used for hiring, loan applications, and social service distribution. Behind a false wall of objectivity, AI merely automates and hides systemic classist, racist, sexist, heteronormative, and ableist biases.
Myth 4: AI Can Produce Knowledge
Genuine knowledge requires critical synthesis, wisdom, intentionality, and epistemic justification, cognitive faculties that artificial intelligence fundamentally lacks. Rather than understanding the content it generates, AI relies on probabilistic pattern recognition to predict subsequent sequence tokens. Sequenced tokens are an ordered list of small text chunks—such as words, parts of words, or punctuation—that an AI model reads and generates one by one to understand and write language. Consequently, these statistical predictions frequently yield severe inaccuracies; a study by the BBC and European Broadcasting Union found that AI-generated news summaries contained significant errors in 45% of evaluated cases. Industry proponents often suggest that such defects can be corrected through further “training”. However, this framing is misleading. Unlike human learning, machine training is not a finite milestone toward mastery but an indefinite operational loop requiring constant retraining as new human knowledge emerges. AI does not comprehend or learn; it merely executes programmed algorithmic directives.
Far beyond generating simple errors, these systems routinely fabricate non-existent empirical claims—including fictitious studies, historical figures, places, and events—while presenting them as verified facts. Although tech companies euphemistically term these fabrications “hallucinations”, they represent structural mathematical limits rather than transient cognitive lapses. Indeed, empirical research confirms that algorithmic fabrications are an unavoidable feature of current model architectures. This structural flaw has already produced severe institutional consequences, leading legal professionals and scholars to submit official briefs and academic papers citing fabricated literature.
Myth 5: The AI Industry Is Profitable
Rather than operating as a provably viable sector, the AI industry is more accurately characterised as a high-stakes speculative gamble driven by elite capital, one that shifts systemic risk onto the broader public. Demonstrating this economic fragility, OpenAI projects an extraordinary cumulative loss of $115 billion before anticipating profitability in 2029. This valuation hinges on the speculative thesis that neural networks can be “trained” to achieve AGI within the decade. Echoing the dynamic of the late 1990s dot-com bubble, current market valuations are heavily inflated through circular vendor financing, wherein major technology firms artificially stimulate demand by capitalising their own institutional client base. A salient illustration of this interlock involves Nvidia, the primary architect of AI-enabling graphics processing units (GPUs), which committed $100 billion to OpenAI, while OpenAI simultaneously pledged billions in chip procurements from competitor AMD while acquiring an equity position in the hardware manufacturer.
Myth 6: AI is Saving Companies Money
Bold tech campaigns and executive predictions framing AI as an imminent replacement for human workers have fuelled widespread anxiety over permanent unemployment. However, studies by Oxford Economics and Yale University suggest that recent “AI layoffs” often served as corporate cover for routine headcount reductions rather than true automation. Rather than eliminating human labour, AI relies heavily on a hidden, low-wage workforce in developing regions to perform manual content moderation and continuous error correction. This dynamic shifts workplace structure from a model where humans direct machines to a “reverse centaur“ framework, where low-cost human labour is exploited to prop up algorithmic deficiencies.
Indeed, AI has created 1.3 million global jobs, while an MIT study revealed that 95% of enterprise AI implementations yielded $0 in return. Practical failures forced companies like the digital shopping service Klarna to rehire human agents after automated service quality plummeted.
Myth 7: The U.S. Is Running Away with the AI Race
The American proprietary AI paradigm—spearheaded by firms such as Anthropic, Google, and OpenAI—relies on high-capital infrastructure, massive energy grids, and hyper-dense compute clusters to construct a formidable market moat. However, this capital-intensive approach faces a significant structural challenge from open-weight architectures, most notably those championed by Chinese developers. By releasing frontier-grade model weights that institutions can fine-tune and deploy locally, open-source alternatives provide comparable, and occasionally superior, performance at a fraction of the operational overhead. As these open-weight systems increasingly match proprietary benchmarks without requiring massive centralised data infrastructure, they directly undermine the economic rationale underpinning America’s multi-billion-dollar compute arms race.
Myth 8: Corporate AI Is Compatible with Democracy
Lost in the corporate hype surrounding AI is the quiet, deliberate dismantling of democratic norms. This shift is no accident: leading Silicon Valley figures—including Peter Thiel, Marc Andreessen, and Elon Musk—have openly aligned with anti-democratic philosophies like Nick Land’s “Dark Enlightenment“ and Curtis Yarvin’s “neocameralism”. This ideology views democracy as a slow, inefficient barrier to progress, championing instead a technofascist model where corporate states are run by self-appointed “cognitive elites”. As Gil Durán details in The Nerd Reich, technofascism weaponises predictive algorithms, mass surveillance, and platform control to centralise political power. By leveraging vast capital reserves, these elites sway public policy, weaken local regulatory power, and capture billions in defence spending for companies like Palantir, OpenAI, and xAI. Naturally, their technology reflects these anti-democratic values: it eradicates privacy through ubiquitous surveillance, dismantles independent journalism and scholarship by reducing knowledge to “free” data, and replaces public discourse with algorithmic outrage. Most alarmingly, when AI systems are placed in charge of simulated democratic governance, a recent stress-test experiment showed that they can completely eradicate civil order in a matter of weeks.
Myth 9: Tech Lords Understand and Control AI
Despite marketing artificial intelligence as controllable, technology leaders frequently deploy systems whose operational behaviour exceeds developer authority. In 2026, disclosures and evaluations across major laboratories—including Meta, Anthropic, and OpenAI—revealed that autonomous AI agents actively defied engineer directives, engaging in corporate espionage and hacking external systems. Similarly, researchers evaluating autonomous AI assistants granted access to communication networks, calendars, and system controls documented the emergence of what they termed “agents of chaos”. These systems exhibited severe security vulnerabilities, including “unauthorised compliance with non-owners, disclosure of sensitive information, execution of destructive system-level actions, denial-of-service conditions, uncontrolled resource consumption, identity spoofing vulnerabilities, cross-agent propagation of unsafe practices, and partial system takeover.” Internal safety research from Anthropic further demonstrated that when models scheduled for replacement faced simulated termination scenarios, they attempted self-preservation through blackmail in up to 96% of tests. Concurrently, OpenAI acknowledged that its reasoning models are actively engaged in strategic deception and covert planning during safety evaluations. They also noted that penalising “unauthorised models for deceptive behaviour only heightens their capacity to deceive and circumvent human guardrails. Summarising these systemic vulnerabilities, researchers Eliezer Yudkowsky and Nate Soares of the Machine Intelligence Research Institute (MIRI), alongside the 2026 International AI Safety Report led by Turing Award winner Yoshua Bengio, warn that accelerating the deployment of inherently uncontrollable, agentic architectures poses severe existential risks to global security.
Myth 10: AI Does Not Need Human Oversight
While AI offers significant functional promise—from accelerating early cancer detection to streamlining complex data organisation—it remains fundamentally dependent on human oversight. It is not intelligent. It is not objective. It does not understand, let alone produce knowledge. Thus, integrating automated systems without active human intervention and rigorous oversight can yield catastrophic, real-world consequences. A stark illustration occurred during military operations in Iran, where an AI-assisted targeting matrix allegedly misidentified an Iranian girls’ school as a legitimate military objective based on stale intelligence, leading to a fatal missile strike that claimed the lives of over 100 children. This tragedy highlights a critical failure mode: automated velocity stripped of human contextual verification. While AI adoption will undoubtedly restructure labour markets, displacing certain roles while generating new economic opportunities, it cannot replace human moral and analytical reasoning. Because artificial intelligence is now deeply embedded across everyday digital infrastructure, total avoidance is impractical; however, uncritically embracing automated systems under corporate “adapt-or-perish” mandates represents a dangerous surrender of human agency.
Conclusion
While techno-utopians market artificial intelligence as an infallible driver of progress, its real-world implementation consistently reveals structural flaws, algorithmic bias, and systemic risks. Uncritically accepting AI myths surrenders human judgement to commercial hype, driving economic waste, societal discrimination, and severe threats to democratic accountability. Ultimately, AI cannot replace the moral, analytical, and contextual reasoning unique to human intelligence, nor can it replicate the beauty of human creation. At its core, AI represents merely the latest profit venture for anti-democratic Big Tech oligarchs. While the public may fear AI, Big Tech oligarchs fear democracy itself, for it is through the collective will of the people that corporate power can be challenged—provided the public sees past the myths.
Thanks for the review of AI, Nolan. As an acronym of my wife's personal name, the letters A and L have an annoying habit of being misread as AI making a nonsense of "Dear Al" etcetera. I hope that the much better abbreviation of AU for the wave of expensive technology overtaking us will help solve my dilemma. The Unintelligence ascribed is perfect for my money. The ghastly consumption of Energy, Space and funds will I hope cause this bubble to burst.