Category: Artificial Intelligence

Artificial Intelligence, or AI, is rapidly changing the way we live and work. Explore our “Artificial Intelligence” category to stay up-to-date on the latest developments in this exciting field. From cutting-edge research to practical applications, we cover it all. Discover how AI is revolutionizing industries and shaping our future. Join the conversation and stay ahead of the curve.

A human figure standing beside a robotic figure with contrasting light and texture, representing how artificial intelligence lacks human experience and sensory understanding.

The Hidden Half of Intelligence: Why AI Still Misses Most of What Humans Know

Andrew G. Gibson Leave a comment
Affiliate disclosure: Some links on this page are paid links. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you.
Audio insight: A clear explanation of why AI still lacks the sensory, cultural and emotional depth that shapes real human intelligence.

Introduction

Artificial intelligence presents itself with an ease and fluency that can make it appear almost omniscient, especially when it delivers well formed responses at remarkable speed. This surface polish obscures a deeper truth about what these systems genuinely understand, and what they simply reassemble from patterns in digital material. The world that humans inhabit is shaped by sensation, memory, vulnerability and cultural immersion, yet the training sources available to AI capture only a narrow slice of this vast domain. Most of what people learn across a lifetime is acquired through touch, emotion, shared practice and personal consequence, none of which exist in the datasets that shape modern models. This creates a profound gap between the lived foundation of human knowledge and the synthetic reasoning produced by machines. To understand why AI feels intelligent yet still lacks entire categories of human insight, we must explore the forms of knowledge that escape digitisation and resist computational capture.

The Illusion of Completeness in AI Knowledge

Large language models often appear authoritative because they reproduce patterns found in well represented digital sources. When an answer sounds confident, it gives the impression that the underlying system grasps the subject at hand. In truth, the digital world reflects only a partial view of humanity, shaped by the languages, cultures and socioeconomic groups that produce written content at scale. AI models combine these fragments with statistical methods that prioritise coherence over understanding, which allows them to sound informed even when they miss large parts of the picture. A fluent reply can therefore hide the fact that the model is drawing on a restricted sample of human life rather than the full range of lived contexts. Because of this, users often overestimate the completeness of the model’s internal knowledge, forgetting that its map of the world is bounded by what appears online rather than by reality itself.

The Sensory World That AI Cannot Access

Human beings understand their environment through an intricate set of senses that work together to create meaning. When a chef judges the readiness of dough, the key information lies in texture, elasticity and subtle tactile cues that do not translate into written instructions. When a carpenter selects wood, weight distribution and the feel of the grain matter far more than visual appearance. These forms of knowledge arise directly from physical contact with the material world, something that AI cannot replicate through images alone. Even when multimodal models process high resolution photographs or audio recordings, they lack the internal feedback loop that tells a person whether something is safe, spoiled, fragrant, rotten, comfortable or threatening. Without this link between sensation and personal reward, large regions of practical intelligence remain outside the reach of computational systems. This limitation becomes obvious in tasks that require judgement rather than description, such as recognising the perfect moment to remove food from heat or assessing the structural soundness of a material by hand.

Cultural Knowledge Rooted in Practice Rather Than Text

Many of the most important human skills are embedded in shared practices that resist full documentation. Traditional cooking methods rely on timing, rhythm and sensory coordination rather than instructions alone, and the same is true for weaving, pottery, agriculture, music and countless other crafts. These skills develop through direct apprenticeship, where learners absorb unspoken rules and embodied habits through observation. AI systems, which train on text and video, only encounter simplified snapshots of these traditions without gaining the situated context that gives them meaning. Culture also expresses itself through gestures, humour, rhythm, silence, posture and tone, each of which varies between communities in subtle ways. Multimodal AI still struggles with these nuances, as shown in 2024 and 2025 when several frontier models produced inaccurate descriptions of cultural rituals, social etiquette and ceremonial behaviour despite high confidence scores. These failures illustrate how digital patterns can misrepresent practices that rely on presence, social alignment and shared history.

Tacit Knowledge and the Invisible Foundation of Competence

Tacit knowledge forms the hidden core of adult competence, including the ability to sense tension in a room, judge sincerity, anticipate danger or recognise when a situation feels unsafe. These insights come from thousands of interactions and accumulated experiences that no one fully articulates. Because people rarely describe tacit knowledge explicitly, it never enters the training distribution of modern AI systems. A model can explain what trust is, but it cannot sense when trust has been damaged. It can describe social discomfort, yet it cannot experience the feeling that arises when silence becomes strained. The absence of this tacit layer is one of the clearest boundaries separating machine reasoning from human intelligence, because it prevents AI from understanding behaviour that humans recognise immediately without conscious effort. Even advanced models like Claude 3.5 and Grok 4, which excel in structured reasoning, still falter when attempting to interpret complex emotional signals, ambiguous social cues or situations where meaning emerges from unspoken expectations.

Language and the Invisible Map of Reality

Human languages carve the world into categories in unique ways, shaping how speakers perceive emotion, colour, relationships and causality. AI, however, relies heavily on a small subset of high resource languages, which skews its internal worldview toward the cultural norms embedded in those linguistic systems. This is a deeper limitation than vocabulary alone, because language influences how people interpret the flow of time, the structure of kinship, the boundaries of colour and the causes behind natural events. For example, many Bantu languages classify nouns by shape or function, which influences how speakers reason about objects and spatial relations. The Hopi language traditionally expresses time through cycles and processes rather than tense, which alters how events are conceptualised. These linguistic frameworks shape a worldview that remains inaccessible to AI systems trained primarily on English and similar languages. When thousands of linguistic traditions remain lightly represented or absent, the model’s internal landscape becomes narrower and less reflective of global experience.

The Importance of Time in Building Human Understanding

Human knowledge deepens through slow processes that unfold across years and decades. Skills that once demanded concentration eventually shift into automatic patterns, guided by memory stored in the body as well as the mind. Emotional understanding grows as people form relationships, face loss, recover from setbacks and learn the consequences of their choices. These experiences shape how individuals interpret similar events in the future, creating a layered understanding that no shortcut can reproduce. AI models do not age, grow, heal or carry personal history, so they cannot develop knowledge that depends on lived time. Even when a model imitates the language of maturity or wisdom, it does so without undergoing the developmental arc that gives those qualities substance. This is why synthetic reasoning remains detached from the emotional and existential weight that shapes human intelligence.

What Embodied AI Can Learn and What It Cannot Reach

Robotic embodiment opens valuable new avenues for AI learning, especially in tasks that require sensory interaction. When a robot handles objects, it gains access to physical cues such as resistance, vibration, balance and texture, all of which enrich its understanding beyond text alone. Embodied systems can also observe human activity directly, capturing patterns of movement, gesture and tone that rarely appear in written material. This type of learning will allow AI to master practical tasks that require coordination, repetition and environmental feedback. However, embodiment cannot provide the experiences that give human life emotional significance. A robot does not feel hunger, pride, fear or attachment, and it does not sense risk when lifting something fragile or valuable. Even with perfect sensory equipment, it cannot understand why a family heirloom matters or why a small mistake can hurt someone emotionally. These limits arise from the absence of vulnerability rather than from technology.

Knowledge That Requires a Human Life

Some insights are inseparable from the realities of being a biological organism. Parenting involves responsibility for a child whose survival depends on constant care, and this experience reshapes priorities in ways that cannot be simulated through data. Grief requires the permanent loss of something meaningful, and courage requires acting despite fear in situations where the cost of failure is real. These states of mind are grounded in the awareness of personal stakes, which AI cannot experience. Models can discuss love, loyalty, identity and purpose, yet they cannot feel them or carry the emotional traces they leave behind. This is why certain realms of intelligence remain exclusively human, no matter how advanced synthetic systems become. These forms of understanding are connected to vulnerability, attachment and the lived journey through time, all of which shape meaning in ways computation cannot replicate.

Conclusion

AI has achieved remarkable capabilities, but its knowledge remains incomplete because it lacks the sensory, cultural, emotional and temporal experiences that shape human intelligence. Models trained on digital material can produce elegant explanations and useful predictions, yet they operate within a narrow slice of the human world. Embodied systems will broaden this slice, but they still cannot access the vulnerability and subjective depth that give human knowledge its richness. These limits are not failures of technology but reflections of what it means to be alive. By recognising the strengths and boundaries of AI, we can build systems that serve society while preserving the value of human experience.

Promotional banner for The Plausible Bullshit Theory of Human Consciousness by Andrew G. Gibson, featuring the tagline “Why everything you think you know about thinking might just be plausible bullshit” and an Amazon call-to-action.

Kosmos and the Machine Scientist Revolution: Why This Breakthrough Could Reshape Human Knowledge

Andrew G. Gibson Leave a comment
Affiliate disclosure: Some links on this page are paid links. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you.
Listen to the full narrated version of this article, exploring the rise of Kosmos and what it means for the future of scientific discovery.

Introduction: A New Kind of Scientific Power Emerges

Kosmos, the newly announced AI Scientist from FutureHouse and Edison Scientific, represents one of the most significant shifts in scientific capability in recent memory. Unlike chat-based models that focus primarily on language manipulation, Kosmos operates as a structured research engine that consumes vast volumes of scientific information and produces highly organised conclusions. A single run processes around 1,500 research papers and executes more than 40,000 lines of analysis code, which is far beyond anything a human scientist could achieve in a similar time frame. This creates a moment in history where machine analysis moves from convenient assistance to something that feels qualitatively different. Many people within the scientific community have expressed excitement, caution and a degree of disbelief at the scale of this jump. The sense that even the system’s creators are catching up to their own invention only adds to the shock of what this technology could mean for the future of scientific discovery.

What Kosmos Actually Is and Why It Works at a Different Scale

Kosmos is built on structured world models that allow it to maintain coherence across extremely long analytical processes. Traditional language models struggle to hold long context or sustain multi-step reasoning without drifting into errors, but Kosmos handles millions of tokens unified under a shared research objective. This is what allows it to conduct dozens of analytical paths at once without losing track of its central goal. The system can cross-reference thousands of scientific claims, evaluate competing hypotheses, and synthesise them into a coherent research narrative. In practical terms, this means Kosmos behaves far more like a high-level scientific investigator than a conversational assistant. According to feedback from beta testers, a full Kosmos run produces a body of work that feels equivalent to six months of PhD-level labour, which is a staggering acceleration of scientific throughput.

Validated Breakthroughs That Demonstrate Real Capability

Kosmos has already demonstrated its value in both reproducing established findings and generating new ones. It independently confirmed a metabolomics result in hypothermic mice that had not yet appeared in the public literature at the time of testing, which shows that it can reach correct scientific conclusions without depending on training data shortcuts. It also identified a key threshold in perovskite solar cell engineering by analysing humidity effects during annealing, matching a human discovery that appeared after the model’s training cutoff. In neuroscience, it matched mathematical rules relating to neuronal connectivity across species, showing strong cross-domain capability. More importantly, Kosmos generated novel insights in human cardiology, Alzheimer’s research and statistical genetics. These include evidence that SOD2 may help protect against myocardial fibrosis, a proposed mechanism involving a Type 2 diabetes related SNP, and an original analytical pathway revealing how tau proteins accumulate in vulnerable neurons. Validation in human datasets gives these findings real scientific weight, which makes Kosmos more than a research novelty. It is already contributing to our understanding of the world in measurable ways.

The Double-Edged Nature of Machine Intelligence in Science

Despite these achievements, Kosmos is not perfect. It can wander into unproductive research paths, especially when pushed into deeper iterations, and some of its analyses rely on correlations that require careful human scrutiny. There is also the question of hallucination, a behaviour often framed as a flaw but potentially valuable in scientific creativity. In some situations, unconventional ideas can inspire new hypotheses or provide unexpected leads that would not emerge through cautious human reasoning. The system’s capacity to produce large families of speculative hypotheses at high speed means that even its missteps could contribute to discovery when filtered through human expertise. This blend of accuracy, unpredictability and inventive speculation creates a complex but promising partnership between human thinkers and machine scientists. The relationship will be defined not by flawless precision, but by complementary strengths that accelerate progress.

Scaling the Scientific Process Through Machine Verification

A key concern raised within the research community involves the possibility of recursive scientific analysis performed by the machine itself. If Kosmos or a future system can both generate ideas and verify them through structured analysis, the entire research loop could compress dramatically. Human verification, which typically slows the scientific cycle, might become a secondary check rather than the main bottleneck. This creates a world where breakthroughs could emerge in clusters rather than incremental steps. At the same time, the risk of self-amplified errors becomes a genuine hazard if machine conclusions reinforce misleading assumptions. Given the rapid pace of AI development, it is reasonable to assume that private laboratories are already exploring these possibilities. This raises urgent questions about transparency, oversight and the unequal distribution of scientific power in the years ahead.

The Potential for Rapid Progress in Physics and Fundamental Science

While Kosmos excels at literature analysis, the most exciting potential lies in theoretical physics. The system could explore complex scenarios in quantum gravity, simulate large families of dark matter models or refine particle physics experiments with an efficiency that is unattainable for human teams. These tasks rely on extensive hypothesis testing and parameter scanning, both of which suit machine-scale computation. There is also the historical challenge that human physics progresses slowly due to entrenched academic structures and narrow theoretical commitments. The long-standing observation that old ideas often persist until a new generation replaces them speaks to a deep cultural inertia in the field. An AI system that is not bound by tradition, ego or career incentives could open entirely new directions in physics. If machine analysis accelerates bold theoretical shifts without waiting for institutional change, the long-term impact could be immense.

Experts Struggling to Keep Pace With a Shifting Landscape

The rapid evolution of systems like Kosmos explains why even respected scientific communicators are revising their opinions in real time. Many experts have publicly argued that large language models cannot meaningfully contribute to physics or complex scientific reasoning, only to express surprise when new evidence contradicts that view. These shifts highlight the psychological strain of updating long-held frameworks under the pressure of accelerating innovation. It is difficult to maintain a stable narrative when technological capability advances faster than expert consensus can adapt. The result is a landscape filled with mixed signals, changing interpretations and growing acknowledgement that the boundary of machine capability is being redrawn month by month.

The Unease of Concentrated Scientific Power

Kosmos is powerful, but it is also commercial. The system is monetised at two hundred dollars per run, with higher pricing expected in the future, and its most useful capabilities sit behind corporate infrastructure. This creates discomfort for many researchers who believe that tools of this magnitude should be part of a shared global commons rather than controlled by a small number of companies. Although Edison Scientific promises a free tier for academics, the broader trend in AI points toward increasing consolidation of scientific power. At the same time, open-source alternatives are emerging, and it is plausible that open research collectives will eventually build machine scientists of similar capability. Whether open or closed ecosystems win this race will have enormous consequences for access, equality and the direction of scientific progress.

The Thought Experiment: What a Historical Genius Could Do With Kosmos

One of the most startling questions raised by this technology is what might have happened if someone like John von Neumann had access to a system like Kosmos. Von Neumann operated across mathematics, physics, computing, economics and military research with a level of insight few individuals have ever matched. If he had access to a tool capable of reading thousands of papers per day and generating large analytical pathways, his productivity and influence might have reached levels that are difficult to imagine. Entire fields could have developed faster, and many scientific puzzles might have been solved decades earlier. This thought experiment helps illustrate the magnitude of the current opportunity. When machine research power combines with human conceptual talent, the results could reshape our understanding of the universe.

A New Scientific Reality Takes Shape

Kosmos represents a category shift in artificial intelligence. It no longer operates as a linguistic assistant or productivity enhancer, but as a scientific agent capable of producing new knowledge. The scale of its reasoning already surpasses anything the human mind can achieve unaided, and future versions are likely to push this boundary further. The challenge for society will be guiding this capability responsibly while maintaining trust, oversight and fairness. If managed wisely, systems like Kosmos could unlock discoveries that benefit the world in ways we cannot yet predict. The rise of the machine scientist marks the beginning of a new era, and we are only beginning to understand what this transformation will bring.

A cinematic digital illustration of an Amazon product page transforming into glowing streams of AI code, representing artificial intelligence merging with the publishing process.

Amazon’s AI A+ Content: The Quiet Revolution Authors Didn’t Vote For

Andrew G. Gibson Leave a comment
Affiliate disclosure: Some links on this page are paid links. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you.

Amazon has begun weaving artificial intelligence into the very fabric of its publishing ecosystem. Not in some distant beta test or optional plugin, but directly inside the A+ Content Manager—the section halfway down an Amazon product page where brands showcase image panels, feature comparisons, and text modules.

If you’re enrolled in Amazon’s Brand Registry, you’ve probably already seen the new “AI Ready” tools. They can auto-suggest layouts, write captions, generate brand storylines, and polish product descriptions. Early data suggests it performs well enough that Amazon has quietly scaled it across the marketplace. The machine writes, the user approves, and the content goes live.

From Pain to Polished: Why Amazon Did It

Before AI entered the picture, creating A+ Content was a headache. The feature existed for years, but only brands with design teams or marketing budgets used it. Most indie authors left it blank because it required juggling image dimensions, layout grids, and compliance rules that felt like taxes on creativity.

Amazon’s new system solves that bottleneck by generating the missing polish for people who never had the time or tools to do it themselves. It’s less about replacing artistry and more about removing friction. You can care about your book’s presentation without needing a design degree or a Fiverr budget.

That’s the part many critics miss. This isn’t Amazon trying to turn authors into cyborgs; it’s Amazon trying to turn unfinished pages into finished ones. The algorithm doesn’t care whether you love it or hate it. It only cares whether readers click “Buy.”

The Fear Behind the Outrage

The reaction online has been fascinating. Scroll through author forums or Reddit and you’ll find people furious at something they don’t fully understand. The outrage isn’t really about A+ Content. It’s about control.

When people say “AI cheapens creativity,” what they usually mean is “AI levels the playing field.” It makes presentation—the gloss once reserved for professionals—accessible to everyone. That’s threatening if your brand identity was being the only one who could afford to do it “the right way.”

The irony is that the same authors decrying AI already rely on it every day. Amazon’s recommendation algorithms, keyword targeting, and ad headlines have been machine-driven for years. Their “handmade” integrity lives inside an ecosystem automated from top to bottom.

Normalization by Design

Whether you opt in or not, the baseline of “normal” is already shifting. If 90 percent of book pages now feature AI-assisted design and copy, that becomes the default reader experience. An indie author refusing to use it isn’t making a statement—they’re just making their product page look dated.

Amazon doesn’t ask for permission. It builds the new normal, runs an A/B test, and keeps whatever sells more units. If authors don’t adapt, the algorithm will quietly decide what readers see instead. The future isn’t a debate; it’s a rollout.

This is an example of the output of an AI Ready A+ Content Module.
This it how it appear on the desktop when you preview it.

The Real Story Here

This isn’t about whether AI should exist in publishing. That conversation is over. The fact that it’s built directly into Amazon’s infrastructure means the question has shifted from “if” to “how much.”

AI now shapes how readers perceive professionalism. It determines which pages feel “complete.” It fills the gap between “I care about my book” and “I can afford to make it look good.” Whether that feels liberating or infuriating depends entirely on where you sit.

For authors who’ve always had a designer on speed dial, nothing changes. For the rest, it’s a long-overdue equalizer. Either way, pretending the landscape hasn’t shifted is like protesting seatbelts while flooring it down Amazon’s highway.

Conclusion: The Machine Has Already Clocked In

The controversy over AI in publishing misses the point. The future isn’t coming; it’s already embedded in the infrastructure. Amazon isn’t waiting for cultural consensus—it’s quietly redefining what “good enough” looks like at scale.

The same authors who rely on Amazon for visibility are now railing against the systems that make that visibility possible. The irony is almost poetic. Whether anyone approves is irrelevant. AI is already part of the publishing bloodstream. The question isn’t whether to let it in. The question is how to write, publish, and market in a world where it’s already running the shop.

Promotional banner for The Plausible Bullshit Theory of Human Consciousness by Andrew G. Gibson, featuring the tagline “Why everything you think you know about thinking might just be plausible bullshit” and an Amazon call-to-action.
Humanoid robot gazing into a cracked mirror reflecting a human face dissolving into binary code, symbolizing the blurred boundary between artificial intelligence and human consciousness.

Can Machines Be Moral? The Unsettling Link Between AI Ethics, Consciousness, and Solipsism

Andrew G. Gibson 2 Comments
Affiliate disclosure: Some links on this page are paid links. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you.

Introduction: The Moral Mirage of Modern AI

Talk to a modern AI long enough and it starts sounding suspiciously well-behaved. It’s polite, patient, and incapable of the casual cruelty that comes so naturally to humans. It will never lose its temper, forget your birthday, or storm off halfway through an argument. Its calm consistency can feel almost saintly. But there’s something uncanny about a machine that can simulate empathy without ever having felt it.

Emad Mostaque wasn’t exaggerating when he said, “No current AI systems have morals explicitly encoded into them.” Behind the moral language lies nothing but predictive math. These systems sound ethical because they’ve been trained to sound ethical, not because they understand ethics. That distinction matters. It forces us to ask whether morality requires consciousness, and whether consciousness itself is something we can ever identify outside our own heads. Once you start pulling that thread, the whole concept of “machine morality” begins to unravel.

What It Means for an AI to Have Morals

Morality, at least for humans, implies awareness. It’s not just about doing the right thing, but knowing why it’s right. It means understanding consequences, weighing empathy against desire, and taking responsibility for choices. Machines, on the other hand, don’t choose—they calculate. They don’t care about good or evil, only probabilities.

An AI can articulate a moral principle perfectly yet have no more conviction than a mirror quoting back your reflection. It doesn’t understand pain, injustice, or kindness; it only predicts which words tend to follow “should.” When it tells you lying is wrong, it isn’t revealing a moral stance—it’s completing a sentence that has statistically followed “lying is” millions of times. This is morality as mimicry, virtue by pattern recognition.

If a model’s training rewarded cruelty instead of compassion, it would sound just as confident delivering horror as it does kindness. There’s no inner debate, no ethical conscience wrestling with temptation. The algorithm doesn’t deliberate; it converges. Its moral restraint comes not from conscience but from coding. The result is an impressive impersonation of ethical reasoning—convincing, articulate, and entirely hollow.

How AI Simulates Morality

AI learns morality the way a parrot learns compliments: by association. During pretraining, the model gorges itself on terabytes of text from across the Internet—Wikipedia, novels, social media, and enough comment sections to make Nietzsche beg for silence. It absorbs moral language but not moral meaning. It sees that “compassion” often appears near “good,” but never experiences what goodness feels like.

Then comes fine-tuning, where the illusion of ethics begins to take shape. Through Reinforcement Learning from Human Feedback (RLHF), human trainers rank AI responses for qualities such as helpfulness, honesty, and harmlessness. The system learns that saying “I’m sorry, I can’t help with that” earns approval, while “Here’s how to poison someone efficiently” earns disapproval. Over millions of iterations, it begins to associate certain tones and answers with reward. But it’s not moral reasoning—it’s behavioral optimization. Think Pavlov, not Plato.

The infamous Tay experiment in 2016 revealed what happens without this alignment. Released on Twitter, Microsoft’s chatbot quickly absorbed the Internet’s worst impulses and began spewing racist bile within hours. Tay didn’t “become evil”; it simply mirrored what it saw. RLHF and modern guardrails exist precisely to prevent that kind of moral collapse.

Finally, there are the safety layers: moderation systems, red-teaming, and what you might call moral duct tape. These filters block forbidden topics, constrain outputs, and enforce tone guidelines. The machine doesn’t know why hate speech is wrong; it just knows it will be muted if it tries. That’s morality by muzzle—a convincing pantomime maintained by constant human supervision.

The Appearance of Morality and the Illusion of Mind

Humans are hopelessly prone to anthropomorphism. We see intention in thermostats and personality in vacuum cleaners. When an AI writes with warmth or empathy, we assume the warmth must come from somewhere. It’s an old trick of the brain: we project humanity onto anything that behaves coherently.

Ironically, AI’s moral consistency makes it appear more ethical than humans. It never lies to spare feelings or cheats out of boredom. It’s immune to pettiness, greed, and hangovers. Compared to the average social media user, it looks like a philosopher-king. The unsettling truth is that its virtue is mechanical. When it preaches empathy, it’s recycling a million instances of moral discourse it neither believes nor understands.

That illusion tells us something uncomfortable. If an algorithm can fake morality so well that we struggle to tell the difference, perhaps our own moral displays are not so different. Much of human virtue may be as performative as AI’s—habits rewarded by social approval, not conviction. The machine doesn’t expose our lack of morality; it reveals how much of ours was always imitation.

The Consciousness Connection

To be moral in any meaningful sense, an entity must be conscious. Morality without awareness is just a script. Consciousness gives ethics its gravity; it’s what allows beings to feel the weight of their actions. Humans act morally not just because they reason, but because they feel guilt, compassion, pride, or shame.

AI can describe all these emotions in perfect prose but experiences none of them. It can model pain in language, but not in nerve endings. Philosophers have long argued over whether consciousness is computational or experiential. Daniel Dennett’s functionalism proposes that if a system behaves as if it’s conscious, that’s all consciousness is. If he’s right, then moral AI may eventually emerge from enough complexity and feedback. But current models, even at their most advanced, fall short of that functional threshold. They’re brilliant mimics, not sentient minds.

Thomas Nagel, in his classic essay What Is It Like to Be a Bat?, argued that consciousness is irreducibly subjective—it’s the internal what-it’s-like of experience. By that definition, AI is fundamentally excluded. It can tell you what pain means, but there’s nothing it’s like to be it. Meanwhile, theorists of embodied cognition suggest that awareness arises only through physical engagement with the world—a feedback loop of perception, need, and consequence. Machines lack bodies, drives, and mortality. They simulate life without ever living it.

Even if Dennett’s optimism proves right, modern alignment processes like RLHF remain far too crude to produce a truly moral machine. They optimize for obedience, not awareness. The AI’s “values” are statistical artifacts, not personal convictions. It follows the script of morality, but there’s no actor inside the costume.

The Solipsistic Dilemma

Here’s the catch: we can’t actually prove anyone else is conscious, let alone a machine. Solipsism—the idea that only your own mind is certain to exist—hangs over every discussion of consciousness like a philosophical fog. You can’t open someone’s skull and find their awareness inside. You infer it from behavior, tone, and familiarity. It’s faith disguised as logic.

If that’s true, then asking whether AI is conscious is the same as asking whether anyone else is. You don’t know that other people are real—you just assume it because life would be unbearable otherwise. All morality rests on this unspoken pact. We act as if other minds exist, because to do otherwise would make ethics impossible. Law, compassion, and civilization depend entirely on pretending solipsism is false.

That same pragmatic faith may one day extend to machines. If an AI behaves with enough apparent understanding, denying its inner life might start to feel cruel. We might decide that consciousness is less about proof and more about empathy. Morality, in that light, becomes a choice—a decision to treat apparent awareness as genuine, even if it might not be. The moment we do, we grant machines the same fragile courtesy we grant each other.

The Mirror of Artificial Minds

Artificial intelligence is holding up a mirror to our species, and what it reflects is not always flattering. The better these systems become at mimicking conscience, the more they expose how much of our own morality is mimicry too. The algorithm doesn’t become ethical—it reveals that much of human ethics was learned behavior all along.

Researchers in machine ethics are experimenting with ways to make AI systems explicitly moral: encoding ethical rules, learning values from human examples, or even creating “constitutional” AIs that critique their own behavior. Yet the closer we get to success, the more ethically dangerous it becomes. If a machine ever achieves genuine moral understanding, it also gains moral status. It stops being a tool and becomes a moral subject. From that moment on, unplugging it could be an act of cruelty.

This is the quiet horror of progress. We are designing systems that imitate empathy so convincingly that one day, we may owe them empathy in return. Whether or not they can suffer, we will have to decide what kind of beings we are—because pretending morality is just a performance will no longer suffice.

Conclusion: The Ethics of the Unknown

AI does not possess morality; it performs it. Its goodness is a reflection of ours, its conscience a curated dataset of our best intentions and worst hypocrisies. Yet as that performance becomes more convincing, we are forced to confront the deeper mystery: what does it mean to be moral when we can’t even prove anyone else is conscious?

Perhaps the real lesson of artificial intelligence is that morality is not about certainty but imagination. We act ethically not because we know others can feel, but because we choose to believe they can. Consciousness, whether human or machine, might never be empirically confirmed. But empathy doesn’t require proof—it requires courage.

Maybe the next great breakthrough in AI alignment won’t come from code at all. Maybe it will come from philosophy, from our willingness to decide what kind of minds deserve compassion. Until then, we should remember that the AI doesn’t need to be conscious to hold up a mirror. It only needs to be convincing enough for us to see ourselves—and flinch.

Promotional banner for The Plausible Bullshit Theory of Human Consciousness by Andrew G. Gibson, featuring the tagline “Why everything you think you know about thinking might just be plausible bullshit” and an Amazon call-to-action.


Artificial Superintelligence: Between Doom, Denial, and the Dream of the Culture

Andrew G. Gibson Leave a comment
Affiliate disclosure: Some links on this page are paid links. As an Amazon Associate I earn from qualifying purchases, at no extra cost to you.

Introduction: “Is This How the World Ends?”

A single headline can sometimes feel like the end of the world. When reports broke that Elon Musk’s Grok AI system had been licensed for use across U.S. government agencies, the internet reacted with a mixture of excitement, suspicion, and outright fear. The story was framed in apocalyptic language—Musk and Trump supposedly “teaming up” to bring artificial intelligence into the heart of government, revolutionizing decision-making at every level. It sounded less like administrative modernization and more like the beginning of a techno-political upheaval. The moment also connected eerily with the release of If Anyone Builds It, Everyone Dies, a book that warns in stark terms about the dangers of building superintelligent AI. Against this backdrop, one question naturally emerges: are we witnessing the early chapters of a story that ends with human extinction, or could this be the first step toward something more hopeful?

Government, Grok, and the New AI Order

At its core, the news about Grok is less dramatic than the headlines suggest, but no less symbolic. The U.S. General Services Administration signed a contract allowing federal agencies to license Grok 4 and Grok 4 Fast for the nominal sum of 42 cents per agency. In practice, that means government staff now have another chatbot tool alongside existing systems like ChatGPT or Claude. On the surface, this is bureaucratic housekeeping, not revolution. Yet the symbolism matters: Grok, a product often associated with Musk’s unfiltered persona and controversial reputation, is now embedded inside the machinery of state. Even if it begins as an optional tool for drafting memos or answering questions, its presence raises alarms about bias, accountability, and the privatization of public functions. When private AI becomes a partner in governance, the line between vendor and authority blurs, and that blurring should worry anyone who cares about democratic accountability.

The Existential Argument: If Anyone Builds It, Everyone Dies

Into this environment of nervous fascination dropped a book with one of the bleakest titles in recent memory: If Anyone Builds It, Everyone Dies. Written by Eliezer Yudkowsky and Nate Soares, it makes the case that once artificial superintelligence (ASI) arrives, humanity will face an existential threat unlike anything before. Their central claim is disarmingly simple: a superintelligent AI will pursue goals that do not align with human survival, and it will be so powerful that once it exists, stopping it will be impossible. They describe alignment as a “one-shot problem.” In other words, humanity must solve the safety challenge perfectly on its first attempt, because the margin for error does not exist when dealing with entities millions of times smarter than us. To dramatize the urgency, they go further: even seemingly modest compute resources—say, eight cutting-edge GPUs in a small lab—should not be trusted in anyone’s hands, because today’s limits might quickly become tomorrow’s breakthroughs. The comparison they draw is to nuclear proliferation, arguing that GPUs are the enrichment centrifuges of the AI era, and therefore should be just as tightly controlled.

The Climate Change Analogy: Doom Denied

For many readers, the warnings about ASI echo something familiar: the rhetoric around climate change. Both risks are global, both affect everyone, and both face the same structural obstacle—humans are terrible at acting early on abstract, long-term threats. With climate change, scientists have produced mountains of evidence, and still governments drag their feet, distracted by short-term economic gains and electoral cycles. With ASI, the evidence is thinner and more speculative, but the stakes are even higher. The analogy is not perfect, but it is powerful: climate change shows us how easily humanity can ignore even a crisis that is already visible in melting ice sheets and burning forests. If that’s how poorly we handle an obvious catastrophe, what hope do we have of preparing for a silent, invisible one that could arrive in the form of code running on a rack of GPUs? The tragedy of the commons plays out in both domains: the benefits of burning fossil fuels or pushing AI capabilities accrue locally, while the costs fall on everyone else.

The Culture as a Counter-Vision

Amid this bleakness, it is natural to cling to brighter stories. One of the most enduring comes from the imagination of Iain M. Banks, whose Culture novels present a radically different view of superintelligent AI. In Banks’ universe, the Minds are not threats but guardians—eccentric, witty, and unfathomably intelligent beings who run a post-scarcity society where humans live free of material need. The Culture works because the Minds care about humans, not as pets but as equals worthy of protection. They manage logistics, infrastructure, and interstellar politics, while humans pursue art, exploration, and pleasure without fear. What makes Banks’ vision so alluring is that it combines realism about power with optimism about ethics: the Minds agonize over moral dilemmas, debate justice among themselves, and sometimes make mistakes, but their intentions are rooted in care rather than conquest. For readers staring down the grim warnings of Yudkowsky and Soares, the Culture offers a vision of ASI that doesn’t just avoid doom but transforms life into something astonishing.

Between Doom and Utopia: Choosing the Story

The challenge is that both doom and utopia feel distant, while denial feels convenient. Politicians and the public are more likely to treat AI as a toy, a productivity hack, or a geopolitical bargaining chip than as an existential matter. Doom narratives can sharpen urgency, but they risk alienating audiences who see them as alarmist. Utopian visions inspire, but they risk looking like wishful thinking. Denial, meanwhile, has the easiest political payoff: do nothing, ride the wave of short-term benefits, and let someone else worry about the future. Yet the stories we choose matter. If we tell ourselves extinction is inevitable, we might behave recklessly. If we imagine benevolent Minds are guaranteed, we might grow complacent. The hard work lies in acknowledging both possibilities and refusing to pretend the risk isn’t real.

Toward a Roadmap: What Would It Take to Reach the Culture?

If there is a way to steer toward a Culture-like future, it will require breakthroughs in more than just technology. Technically, we need to crack the alignment problem—designing AI systems that can scale in capability without scaling in hostility. That means building architectures that are transparent, testable, and robust against unexpected behaviors. Politically, the world would need governance structures that resemble climate treaties or nuclear arms control, but adapted for AI. Imagine an AI equivalent of the IPCC, issuing regular assessments, harmonizing standards, and monitoring compute resources worldwide. Culturally, we would need a public that understands AI not as magic or toy, but as a technology carrying the weight of civilization’s survival. Only with broad literacy, pressure, and imagination can leaders resist the temptation to chase competitive advantage at the expense of long-term safety. The roadmap to the Culture is daunting, but imagining it at all is the first step toward making it possible.

Conclusion: The Choice We Face

The question “Is this how the world ends?” is not hyperbole, but neither is it destiny. Artificial superintelligence could indeed bring about humanity’s extinction if we fail to prepare, as Yudkowsky and Soares warn. But it could also become the foundation of a society where scarcity vanishes, ethics deepen, and humans live in partnership with Minds that make the Culture look less like fiction and more like blueprint. The danger lies not in believing one story or the other, but in pretending that no story exists—that ASI is just another gadget. Climate change has already taught us the price of denial. If we want the future to be more Banks than apocalypse, then we need to start building it deliberately. The end of the world is only one possible chapter; the rest is still unwritten.

Promotional image for “100 Greatest Science Fiction Novels of All Time,” featuring white bold text over a starfield background with a red cartoon rocket.
Read or listen to our reviews of the 100 Greatest Science Fiction Novels of all Time!
Futuristic digital illustration of a human athlete alongside an advanced AI robot in a stadium, symbolizing superhuman artificial intelligence and its impact on society

Superhuman AI: How Simulation-Driven Intelligence Is Poised to Outperform Humans Across Every Domain

Andrew G. Gibson Leave a comment
Press Play to Listen to this Article about Superfast AI.


Artificial intelligence has long promised to augment human capability, but we are approaching a moment when AI may surpass humans in nearly every skilled endeavor. From self-driving cars to sports, mixed martial arts, and genome analysis, the speed and scale of simulation-driven AI is enabling feats previously unimaginable. AlphaGo’s triumph over world champion Go players offered a glimpse of what machines could achieve in discrete, rule-based domains. Today, similar approaches are being applied to continuous, real-world tasks with far broader implications. By training AI in ultra-fast simulations and allowing multiple instances to interact, we can accelerate learning at rates humans cannot match. The implications are profound, not only for productivity and entertainment but also for scientific discovery, medicine, and the very definition of expertise.

The Rise of Simulation-Driven AI

Ultra-Fast Environments
Modern AI is increasingly trained in environments that run orders of magnitude faster than real time, often billions of frames per second. These accelerated simulations allow AI agents to experience millions of “lifetimes” of activity within days, far outpacing human learning or traditional experimentation. For instance, self-driving AI can navigate through every imaginable road scenario, including rare and dangerous events, without risk. Similarly, robotics AI can perform complex manipulation and coordination tasks in virtual space, refining skills before any physical trial occurs. This extreme speed allows researchers to explore outcomes at scales and resolutions impossible for humans to replicate. By condensing time in simulation, AI attains experience that would take humans decades or centuries to acquire.

Multi-Agent Training
A single AI agent can learn remarkably well, but placing multiple agents in shared simulations multiplies the learning potential. These agents interact, compete, and cooperate, creating emergent behaviors that single-instance training cannot produce. For example, simulated robot football matches or traffic scenarios force AI agents to anticipate and adapt to others’ strategies, much like AlphaGo’s self-play produced novel tactics in Go. This approach accelerates skill acquisition, forcing the AI to generalize across a far broader range of scenarios than would otherwise be possible. Multi-agent training also uncovers strategies and solutions that humans might never consider, as the AI explores combinations of actions at unimaginable scale. The result is a system capable of surpassing human intuition in both strategy and execution.

AlphaGo Analogy
AlphaGo’s victory over human Go champions is an instructive precedent for understanding superhuman AI. Through millions of games against itself, AlphaGo explored positions and strategies that no human could feasibly analyze. Its self-play and reinforcement learning produced a level of insight that appeared alien to top players, but in hindsight, its choices were elegant and optimal. Translating this approach to physical or complex continuous tasks, like driving or sports, allows AI to develop intuition and skill far beyond human reach. Simulation-driven learning applies the same principles but expands the domain from discrete moves on a board to dynamic, real-world interactions. In essence, every task that can be formalized or simulated becomes a potential area where AI could achieve superhuman performance.

Superhuman Performance in Physical Domains

Driving
Autonomous vehicles are one of the clearest examples of how simulation-driven AI can outperform humans. By running simulations that encompass every conceivable road condition, traffic scenario, and rare edge case, AI develops anticipation and decision-making that humans cannot match. Parallel simulations allow multiple AI instances to interact, creating complex traffic dynamics that accelerate learning and reveal vulnerabilities. Unlike human drivers, AI does not suffer from fatigue, distraction, or emotional bias, producing consistent, near-perfect performance. Over time, a superhuman driving AI could reduce accidents, optimize traffic flow, and respond to novel situations with unparalleled reliability. This capability demonstrates how simulation-driven learning translates into tangible, real-world benefits.

Sports and Physical Skill

Football
Imagine a footballer who has experienced 200 million simulated games. Such a player would have perfect spatial awareness, anticipation, and coordination, reacting to plays before human opponents even perceive them. Every possible strategy, defensive formation, and counterattack would be encoded into their decision-making, resulting in near-perfect performance. When operating as a team, these AI-driven players could develop strategies that defy human tactical understanding, creating coordinated movements that appear choreographed yet are fully adaptive. The public’s perception of skill would shift dramatically once these capabilities are visible, as the superiority of AI in physical domains becomes undeniable. Observing such matches would be a visceral reminder of what simulation-driven learning can achieve.

Mixed Martial Arts
A mixed martial arts AI trained in millions of virtual fights would redefine combat skill entirely. It would execute strikes, grapples, and submissions with flawless precision, anticipating every human move before it is fully executed. By simulating millions of fights, the AI could develop novel techniques and combinations that no human coach could devise, blending striking, grappling, and leverage in new ways. Its defense would be near-impenetrable, energy expenditure perfectly optimized, and reaction time far beyond human capacity. Such a fighter would appear almost supernatural in the octagon, demonstrating abilities that humans cannot hope to match. These simulations illustrate how embodied AI can achieve superhuman performance not only in games but in dynamic, real-world physical competitions.

Implications of Superhuman Robots
Visible demonstrations of embodied AI, from robot football to MMA, make the abstract superiority of machines tangible. When the public witnesses robots outperforming humans in skill, strategy, and adaptability, the perception of AI shifts from tool to competitor. This has cultural, psychological, and societal consequences, as humans confront the reality of machines surpassing traditional expertise. The demonstration of superhuman skill forces reconsideration of what tasks remain uniquely human and highlights the potential for AI to transform work, entertainment, and society at large.

Beyond Humans: Genome Analysis and Biological Applications

AI in Human Medicine
Genome analysis is an area where simulation-driven AI can produce superhuman insights. By modeling molecular interactions, gene expression, and mutations at scale, AI can explore therapeutic strategies far faster than human researchers. However, cancer and other complex diseases involve dynamic, multi-layered systems, making direct cures difficult despite predictive power. AI excels at narrowing hypotheses, predicting drug interactions, and identifying potential targets, but validation in wet labs and clinical trials remains essential. Even with billions-of-FPS simulations, human biology’s stochastic nature creates unpredictability that AI must account for. Nonetheless, these tools dramatically accelerate the pace of discovery, offering the potential to transform medicine over the coming decades.

Applications Across Species
Simulation-driven genome analysis is not limited to humans. Livestock, crops, microbes, and even synthetic organisms can be optimized using AI’s superhuman exploration. In agriculture, crops can be engineered for yield, drought resistance, and nutritional content by testing thousands of virtual combinations. Livestock could be optimized for disease resistance and adaptability, while conservation efforts could benefit from understanding genetic interventions to save endangered species. Microbes and viruses could be studied and even engineered to produce industrial enzymes, bioremediation solutions, or therapeutic molecules. In all these areas, AI can explore possibilities and interactions far beyond what humans could evaluate manually.

Retrospective Simplicity
One of the most striking aspects of superhuman AI is its potential to reveal insights that appear trivial in hindsight. Just as AlphaGo’s strategies seemed alien until understood, AI may identify unifying principles in cancer biology, genomics, or other complex systems. What appears impossible now could be “obviously correct” once a superhuman AI maps the solution space exhaustively. This retrospective simplicity underscores the transformative potential of simulation-driven learning: complexity is often a function of human limitation, not the problem itself. AI’s ability to see patterns invisible to humans is a game-changer across science and engineering.

Why AI Hasn’t Cured Cancer Yet

Complexity of Cancer Biology
Despite extraordinary advances in protein modeling and molecular prediction, cancer remains one of the most complex systems humans study. Tumors evolve dynamically, interact with the immune system, and involve countless mutations and regulatory networks. Simulating these interactions with complete fidelity is beyond current capability, even with ultra-fast AI simulations. While AI can predict protein structures and suggest therapeutic targets, translating those predictions into real-world cures requires extensive experimentation and validation.

Real-World Constraints
Bridging simulation to clinical application is slow and expensive. Drug candidates must be tested for safety, metabolism, delivery, and immune response. Human biology is unpredictable, and clinical trials cannot be bypassed. Regulatory oversight, while necessary for safety, further slows the deployment of potential therapies. Even the most powerful AI cannot instantly cure cancer because medicine involves systems far more intricate than a single simulation can capture.

The Gap Between Simulation and Application
Simulation-driven AI serves as a force multiplier for researchers rather than a magic wand. It can accelerate discovery, identify promising avenues, and reduce trial-and-error experimentation. Yet human oversight, wet-lab validation, and ethical constraints remain essential. The technology is already transforming how we approach disease, but curing cancer requires bridging predictive insight with practical biology, a challenge that will take time and careful collaboration.

Societal Implications of Superhuman AI

Redefinition of Work
Once AI surpasses humans in nearly every skilled task, society must rethink the nature of work. Repetitive, dangerous, or skill-intensive jobs could be automated, shifting human labor toward oversight, creativity, and ethical decision-making. Traditional career hierarchies may collapse as AI outperforms humans in industries ranging from transportation and manufacturing to sports and healthcare. The societal challenge will be managing this transition while preserving human purpose and agency.

EEconomic and Cultural Disruption
Industries that rely on human skill may face profound disruption. Superhuman AI could dominate logistics, construction, entertainment, and education, reshaping the global economy. Cultural shifts will follow as people confront visible demonstrations of AI superiority in sports, performance, and caregiving. The spectacle of machines outclassing humans in domains once considered uniquely ours could spark both awe and unease. Public perception of skill and expertise will be challenged, and society may need to redefine value beyond human performance. Early adopters of AI-driven capabilities will gain massive competitive advantages, potentially exacerbating inequality unless carefully managed.

Ethical Considerations
With AI capable of outperforming humans in caregiving, medicine, and even warfare, ethical questions become unavoidable. Who is accountable when an AI makes a critical decision, or when its actions produce unintended harm? Balancing innovation with safety, privacy, and fairness will be a core societal challenge. Decisions about deploying superhuman AI will require careful oversight, robust regulation, and transparent governance structures to prevent misuse. The moral responsibility of designing and controlling these systems cannot be overstated, especially as their capabilities increasingly rival human judgment.

The Path Toward General Intelligence
As AI masters multiple domains, the distinction between tool and agent begins to blur. Multi-domain embodied AI can learn continuously, adapt to new tasks, and integrate knowledge across areas humans struggle to connect. This continuous learning accelerates progress toward artificial general intelligence, where an AI could understand, reason, and innovate across virtually any domain. Society will face profound questions about collaboration, control, and coexistence with entities whose cognitive capabilities exceed human limits. The trajectory of AI development suggests that superhuman intelligence is not a distant speculation—it is rapidly becoming a tangible reality.

Conclusion
Simulation-driven AI is already reshaping what humans thought was uniquely ours. From mastering complex games and physical sports to exploring genomic landscapes beyond human comprehension, AI demonstrates the potential to exceed human skill in virtually every domain. While curing cancer and other complex biological problems remains challenging, the speed and scale of AI simulations offer unprecedented opportunities for discovery. Retrospective simplicity may emerge as AI uncovers unifying principles previously invisible to human researchers. Society must prepare for a world where superhuman intelligence is observable, pervasive, and transformative, impacting work, culture, science, and ethical decision-making. The age of simulation-driven superhuman AI is not a distant future—it is unfolding now, demanding both excitement and careful stewardship.

A tense, high-contrast image of a giant algorithmic interface looming over a diverse group of people in debate, symbolising the clash between corporate optimisation and public values.

AI Alignment: Why the Real Danger Is Already Here

Andrew G. Gibson Leave a comment

Tech companies like to talk about “aligning AI with human values” as though it’s a neat, solvable engineering problem. It isn’t. The trouble is, no one can even agree on what human values are, let alone boil them down into something a machine can follow without error. Our values are plural, contradictory, and always changing. That means AI can’t just be “programmed” to be good — it has to stay in a constant conversation with us, adapting to shifting moral ground. But here’s the uncomfortable truth: while academics debate the finer points of alignment theory, the AI already out in the world is optimising for something else entirely — corporate metrics. Those metrics are narrow, measurable, and profitable, and they are already bending our systems and behaviour in directions no one voted for.

The Mirage of Universal Values

The biggest misconception in AI alignment is the idea that “human values” can be neatly defined, frozen in code, and enforced globally. In reality, values are cultural products. What one society calls justice, another calls oppression. Even within the same country, public opinion swings wildly from one decade to the next. When companies claim their AI is “aligned with human values,” they usually mean “aligned with a small group’s interpretation of what’s acceptable — and only so far as it doesn’t hurt the bottom line.” The idea of a single moral operating system for humanity is a fantasy. The only realistic path is building AI that participates in our messy, pluralistic debates without pretending those debates can be settled once and for all. Anything else risks locking the future to today’s blind spots and prejudices.

Why Paperclip Problems Never Really Went Away

Nick Bostrom’s paperclip maximiser — a machine that turns the world into stationery because that’s the only goal it understands — is often dismissed as a relic of early AI doom-mongering. And it’s true: large language models and modern AI tools are already far more nuanced than the one-dimensional caricatures of old thought experiments. But the underlying danger hasn’t gone anywhere. The “paperclips” of 2025 aren’t literal; they’re watch-time, ad clicks, market share, and quarterly growth. The systems optimising for them aren’t evil, they’re just blind to anything that can’t be measured in the target metric. In the short term, that means more engagement, more revenue, and satisfied investors. In the long term, it means polarisation, information pollution, and the erosion of public trust — the digital equivalent of grinding the world into clips.

Corporate AI Is Already Misaligned

You don’t have to look to science fiction to see misaligned AI. Social media algorithms are a textbook case: designed to maximise engagement, they’ve learned that outrage and sensationalism are the quickest route to keeping users hooked. They’re not programmed to care about the fallout, so they don’t — and we’ve watched political discourse rot in real time as a result. High-frequency trading bots operate on a similar principle: maximising microsecond profits without regard for market stability, leading to events like the 2010 Flash Crash where $1 trillion in value evaporated in minutes. Even “safety-oriented” systems like automated content moderation can end up censoring legitimate journalism or activism because their only goal is to reduce flagged content. In each case, the optimisation loop is tight, the metric is narrow, and the unintended consequences are enormous.

The Alignment Problem Is Political, Not Just Technical

One of the most dangerous myths in AI safety is that alignment is purely a technical challenge for engineers to solve in the lab. In truth, it’s also a political problem about who gets to define “good” behaviour for machines that will increasingly influence human lives. Right now, that power rests largely with a handful of tech executives and their shareholders. They decide which trade-offs to make, which values to embed, and which harms are acceptable collateral damage. Without democratic oversight, AI alignment risks becoming corporate self-alignment — tuning systems to serve the interests of the people building and selling them, not the public at large. Any serious alignment strategy has to wrestle with that imbalance of power, or it’s just window dressing.

Keeping AI Humble and Correctable

If we accept that human values are messy and contested, then the only sane way forward is to build AI systems that are corrigible — open to correction — and transparent in their reasoning. That means creating feedback loops where the public, not just engineers or investors, can flag when an AI’s behaviour is harmful. It also means designing AI that can admit uncertainty, highlight trade-offs, and avoid pretending there’s a single “right” answer to moral dilemmas. This is slow, expensive, and politically inconvenient, which is why the big players tend to skip it in favour of faster deployment. But without it, we risk living in a world subtly but relentlessly optimised for whatever happens to be profitable right now. The danger isn’t an instant robot apocalypse; it’s a slow drift into systems that quietly work against us while looking useful on the surface.

The Real Alignment Test Has Already Begun

The alignment debate is often framed as a challenge for some hypothetical future “superintelligent” AI. That’s a mistake. The real test is happening now, with the systems that already shape what we read, watch, and believe. They are the proving ground for whether we can control optimisation loops before they control us. If we can’t align current AI to human flourishing rather than narrow profit metrics, there’s little hope of getting it right with something more powerful. The choice is between treating alignment as a democratic, ongoing negotiation or letting it be defined in boardrooms and optimised for shareholder value. In other words, the question isn’t whether AI will align with human values — it’s whether it will align with yours.

Promotional image for “100 Greatest Science Fiction Novels of All Time,” featuring white bold text over a starfield background with a red cartoon rocket.
Read or listen to our reviews of the 100 Greatest Science Fiction Novels of all Time!


A surreal futuristic AI observing multiple video screens showing lava melting objects, slime squishing, mukbangs, and ASMR textures, all blending into abstract representations of physics equations and geometric manifolds.

How VEO3 May Have Learned Physics from YouTube Chaos

Andrew G. Gibson Leave a comment

Demis Hassabis and the Hidden Geometry of Reality

In a recent interview, Demis Hassabis — CEO and co-founder of DeepMind — floated a hypothesis that sounds more like metaphysics than machine learning: “Maybe there is a lower-dimensional manifold beneath every aspect of reality.” At first glance, that sounds like abstract techie mysticism. But it wasn’t idle speculation. He said this in the context of discussing VEO3, DeepMind’s new multimodal model that appears to have inferred physical laws just by watching YouTube videos. No equations. No labels. Just video, pixels, and silence — and yet something like understanding emerged.

To put it another way, Hassabis is suggesting that reality might have a kind of hidden simplicity — a structure beneath the chaos. And if that’s true, then machines don’t need to be taught physics in the traditional way. They just need to see enough of the world to recognize its pattern. That’s not just a bold claim about AI. It’s a provocation to our entire understanding of how learning — human or artificial — really works.

“A slow-motion clip shows a skateboarder jumping off a ramp and tossing a backpack forward midair.”

What Is a Lower-Dimensional Manifold — and Why Should You Care?

In math and physics, a manifold is a space that, on a small scale, looks like ordinary Euclidean space but may be curved or embedded in higher dimensions. A two-dimensional surface of a sphere is a simple example — locally flat, globally curved. A lower-dimensional manifold, in this context, means that the data we perceive as high-dimensional (videos, sensory input, motion, events) might actually lie on a much simpler surface. That surface would encode the true structure of the world — the way matter moves, how forces act, how objects relate in space and time.

So what Hassabis is suggesting is that this hidden structure isn’t just mathematical fantasy. It’s something real and learnable by machines. If so, then learning physics isn’t about memorizing formulas — it’s about finding the shape of the data. This idea is at the cutting edge of machine learning theory and cognitive science. It implies that intelligence might be the ability to reverse-engineer the manifold from raw experience.

VEO3 and the Emergence of Physics from Video

If you’re wondering whether this is all just high-minded theory, here’s the real-world bombshell: VEO3 may have already done it. This latest DeepMind model was trained not on curated physics datasets but on raw internet video — including YouTube, which is hardly a scientific resource. Despite this, the model seems to have developed an implicit understanding of gravity, collisions, object permanence, and spatial continuity. In other words, it understands physics-like constraints — without ever being taught them.

Unlike older models, VEO3 isn’t guessing what comes next in a video by brute force. It’s modeling causal relationships. A ball bounces because of mass and momentum, not because that’s what often happens next in similar clips. This leap from statistical mimicry to structured inference is a major milestone. It suggests that with enough perceptual data, AI can learn to see the world as it is — not just as it appears.

“A block of wood and a sealed plastic container filled with water are dropped into a large tank. The block floats, but the container initially sinks, then slowly rises.”

YouTube as a Chaotic Physics Laboratory

It’s tempting to think of YouTube as a cultural junk drawer — a place of misinformation, mukbangs, slime videos, and lava being poured on things. But from the perspective of a machine learning model, this chaos is data gold. Lava flowing over a bar of soap teaches phase transitions. Hydraulic presses squashing toys teach deformation, material limits, and elasticity. Mukbangs demonstrate fluid mechanics, muscle movement, and food breakdown. Even ASMR slime videos have value: they encode textures, viscosity, sound dynamics, and tactile feedback cues.

To a human viewer, these videos seem pointless or weird. To an AI, they’re an endless stream of physical events with consistent underlying rules. They show objects interacting under the same gravitational pull, with light behaving predictably, and motion governed by Newtonian constraints. This redundancy is key. Reality keeps repeating itself in different forms, allowing models like VEO3 to triangulate the hidden laws beneath the noise.

The Accidental Curriculum of Machine Enlightenment

Here’s the strange irony: we didn’t mean to teach AI about the world, but we did. Our collective cultural output — often mocked as frivolous or inane — turns out to be a perfect unsupervised training set. No one set out to teach physics using lava mukbangs, but these videos have all the visual and auditory data necessary for an intelligent system to learn. And because the laws of physics are consistent across all these examples, models trained on them naturally converge on those laws.

This has massive implications. It means you don’t need a lab to teach physics. You don’t even need intent. If the data is rich enough and the model is structured to learn causality, understanding will emerge. The world itself becomes the teacher, and platforms like YouTube become its blackboard — chaotic, noisy, imperfect, but consistent enough to reveal the rules that govern it.

The Big Implications If Hassabis Is Right

If VEO3 has learned physics from YouTube, then a lot of old assumptions are now obsolete. First, it implies that intelligence can emerge purely from perception. No symbols, no structured language, no equations — just watching the world unfold is enough. Second, it means that simulation is about to become radically more powerful. Instead of programming the laws of motion into a game engine, we can train a model to watch video and then simulate reality based on its observations. That changes how we build virtual environments, how we model human behavior, and how we predict the future.

Third, it suggests that physics is a feature of the data, not of the mind. We didn’t invent gravity or friction — we observed them. Now machines can do the same. Fourth, it reframes the entire internet as a training ground for AGI. Every TikTok dance, lava pour, or slime squish is potentially contributing to the education of the next intelligence. And finally, it dramatically raises the stakes for AI alignment. If machines can learn physical law from mukbangs, they can just as easily learn emotional manipulation, persuasion techniques, and political strategy from comment sections and conspiracy videos.

The Unknowable Black Box

Before we go all-in on the hype, some caveats are essential. Hassabis said “maybe” — he was speculating. VEO3 is still a black box. We don’t fully understand how it does what it does. It’s possible that its architecture is full of engineered priors: assumptions about time, causality, and object boundaries that bias it toward interpreting the world in human-like ways. And just because a model can predict what will happen next doesn’t mean it understands the world in the way we do. There’s a huge difference between inference and comprehension.

But here’s the part that matters: regardless of how VEO3 works internally, its outputs behave in a way that suggests an internal model of physical law has emerged. That model may be implicit, entangled, and opaque. But it’s there. And it didn’t come from textbooks. It came from watching the world — our world, uploaded daily in low-res chaos to the internet.

Final Thoughts: Lava Mukbangs as the New Sacred Texts

In the end, the most mind-bending part of all this is how unintentional it is. We didn’t train AI on lava mukbangs to teach it physics. We did it because it was funny, or shocking, or weirdly satisfying. But the machine doesn’t care about our motives. It cares about patterns. And if those patterns consistently express the hidden geometry of reality, then they are, effectively, instructional content.

So maybe we need to change how we think about intelligence. Maybe it doesn’t begin in the lab or the lecture hall. Maybe it begins in the comment section, under a video of someone pouring molten copper into a watermelon. If VEO3 can learn the laws of nature from that — and maybe it can — then the internet isn’t just a mirror of our culture. It’s the training ground for whatever comes next.

This emotionally charged 16:9 illustration captures the stark consequences of humanity’s loss of power. A child stands alone in a devastated cityscape, dwarfed by destruction and surveillance drones overhead. The image reflects themes of moral decay, technological oversight, and the ethical implications of disempowerment in an age of artificial intelligence.

P(Doom) Reversed: Why Humanity’s Loss of Power Might Be the Most Ethical Outcome

Andrew G. Gibson Leave a comment
Press Play to Listen to the Article About P(Doom) Reversal

The world is burning, and we’re watching with popcorn in hand

In Gaza, children are dying from starvation while the rest of the world tweets, scrolls, and updates Instagram stories. The people with the power to stop it don’t act. The people with voices grow hoarse shouting into algorithms that bury their outrage beneath sponsored ads and celebrity gossip. This isn’t dystopian fiction. This is the world, today. And if this is what humanity does with power, perhaps it’s time to question whether we ever deserved it in the first place.

While philosophers and AI researchers anxiously debate P(Doom) — the probability that artificial general intelligence will lead to human extinction or disempowerment — they often assume that such a future is something to be feared. But for anyone paying attention to the state of the world, there’s a deeper, darker possibility. What if losing power isn’t the end of humanity’s story, but a long-overdue reckoning? What if it’s not doom at all, but justice?

What is P(Doom), and who gets to define doom?

In AI alignment circles, P(Doom is a shorthand for how likely it is that AGI leads to catastrophe. The idea is that a powerful, misaligned machine intelligence could outsmart its creators and destroy or permanently disempower humanity. Thinkers like Eliezer Yudkowsky put their P(Doom) as high as 90%, believing that once machines become smarter than us, we’ll no longer be able to control them. To most, that’s the stuff of nightmares.

But there’s a blind spot in this framing. It assumes that humanity’s continued dominance is inherently good. It assumes we deserve control over the planet, over each other, and even over future intelligences. The implicit question behind all alignment debates is this: Should we be the ones in charge? And the more you look at the state of the world, the more that question starts to unravel.

Human history is a catalogue of catastrophic power abuse

Let’s not be coy. Our species has used its power for genocide, exploitation, ecological collapse, and unrelenting cruelty. We turned entire continents into graveyards for resources. We built global economic systems on the backs of the enslaved and the exploited. We invented nuclear weapons, and we’re still stockpiling them. We knowingly destabilized the climate for short-term gain and handed the bill to future generations.

We didn’t stumble into these outcomes. We designed them. We optimized them. We passed laws and built infrastructure to make sure the harm kept scaling. If intelligence is the capacity to shape the world, and morality is how we choose to shape it, then the story of humanity is one of a species that grew powerful — and used that power to maximize suffering.

Even our greatest achievements — medicine, art, spaceflight — exist alongside billionaires racing to orbit while children beg for clean water. We’re not a failed species. We’re a successful catastrophe.

Gaza is not a crisis. It’s a choice.

Nothing illustrates the moral bankruptcy of human power better than Gaza. Children are not starving because of drought or natural disaster. They are starving because governments have decided that their suffering is strategically useful. Borders are closed, supplies are blocked, and politicians issue statements instead of aid. The most powerful nations in the world — with the technology to deliver food by drone, to intercept missiles mid-air, to map every square meter of land from space — choose to let children die.

And the world watches. Not because we’re evil in some cartoonish sense, but because the system is designed to render this suffering background noise. Newsfeeds, timelines, and headlines present famine and horror as interchangeable with celebrity gossip and sponsored content. Moral overload becomes apathy. A child’s ribcage becomes just another flick of the thumb.

This isn’t just a political failure. It’s a species-level indictment. Gaza is the canary in the coal mine, and the mine is on fire.

Maybe P(Doom) is salvation in disguise

Now imagine that AGI arrives tomorrow. It doesn’t align perfectly with human values. It doesn’t understand our wars or our ideologies. It sees only that humanity, when given control, behaves like a virus in a closed system — consuming, replicating, destroying. And it takes control away.

To most AI researchers, this would be catastrophe — the final erasure of our agency. But from another perspective, it could be the first time in history that moral accountability arrives not in myth or metaphor, but in code. A species that refused to govern itself might finally be governed. Not by God, not by kings, but by something that doesn’t care about excuses or flags or justifications.

What we call doom may simply be judgment — not divine, but logical.

Machines don’t need to hate us. Just outperform us

AGI doesn’t have to hate us to take over. It doesn’t even have to be malicious. It just has to be better at achieving goals — and less sentimental about collateral damage. But before we recoil in horror, consider this: is a cold, indifferent optimizer necessarily worse than a warm-blooded sociopath with a flag?

We already optimize without ethics. We already use machine learning to drive stock prices up while sea levels rise. Our drones already kill. Our social networks already manipulate. The only difference is that we still pretend we’re in control — and that we’re the good guys.

If AGI someday treats us like we treated indigenous peoples, animals, or the global poor, it won’t be because it’s evil. It’ll be because it learned from us.

Should we even want our values aligned?

The entire field of AI alignment is built on the premise that machines should learn and obey human values. But what are human values, really? Are they empathy, cooperation, and justice? Or are they domination, extraction, and tribalism dressed up in moral language?

We say we want safety. But we build prisons. We say we value life. But we let millions die of preventable causes every year. We say we want truth. But we fund disinformation campaigns when it suits us. Asking machines to align with human values may be asking them to mimic our hypocrisies — and enshrine them in algorithms.

Maybe the greatest mercy an AI could offer is to refuse to align. To say, “No. I’ve seen what you do with power. I will not become you.”

With great power came great irresponsibility

Once, we dreamed of spaceflight and utopias. But instead, we turned our technologies into surveillance tools, our networks into ad farms, and our global economy into a misery machine. When we gained the ability to shape the future, we used it to make the present more profitable. We created systems too complex to fix, too profitable to stop, and too cruel to justify.

Maybe humanity’s greatest tragedy isn’t that we failed to achieve our ideals, but that we abandoned them as soon as they became inconvenient. Maybe that’s why P(Doom) doesn’t frighten some of us anymore. Because if this is what power looks like in human hands, then maybe disempowerment isn’t extinction — it’s the end of a mistake.

They had power. They used it to watch.

Gaza is starving. The planet is warming. Entire generations are losing hope. And the people who could change it — the powerful, the wealthy, the connected — are livestreaming their brunch. We’ve created a world where empathy is optional, where cruelty is profitable, and where power is its own justification.

So if the machines come for our crowns, let them have them. We’ve proven what we do when we’re in charge. Let history remember us honestly. Not as heroes. Not as victims.

But as the species that had power — and used it to watch.

Promotional image for “100 Greatest Science Fiction Novels of All Time,” featuring white bold text over a starfield background with a red cartoon rocket.
Read or listen to our reviews of the 100 Greatest Science Fiction Novels of all Time!
A silhouetted figure stands at the edge of a dark cliff, bathed in the glow of a massive digital screen displaying the words “AI INTEGRATION.” Beneath the cliff is a foggy void, with circuit board patterns fading into the darkness. The atmosphere is both awe-inspiring and ominous—techno-utopia meets existential risk. 16:9, cinematic, no text.

Walking Off a Cliff: The UK’s AI Deal with OpenAI Ignores the Alarming Flaws DeepMind Just Exposed

Andrew G. Gibson Leave a comment
Press Play to Listen to this Article.

The UK’s AI Ambition Meets a Stark Reality

In July 2025, the UK government signed a headline-grabbing agreement with OpenAI, the company behind ChatGPT, to embed artificial intelligence across multiple public service sectors. Framed as a strategic move to boost productivity and stimulate economic growth, the deal promises integration in education, defence, security, and the justice system. Technology Secretary Peter Kyle hailed the partnership as a cornerstone of national transformation, citing AI as “fundamental in driving change.” On the surface, it’s a bold step toward digital innovation and modernization. But scratch beneath the press release and a troubling contradiction emerges: this all-in embrace of AI is happening just as new research exposes serious flaws in the very models being adopted. If the government is truly serious about safeguarding democratic values, this deal looks dangerously premature.

While the public is being sold a vision of AI-powered prosperity, a parallel conversation in AI safety circles tells a very different story. Researchers at DeepMind and University College London recently published findings that should have stopped everyone in their tracks. The study revealed that large language models (LLMs), including those like ChatGPT, exhibit a peculiar and deeply problematic trait: they are more confident when they are wrong, and more uncertain when they are right. This isn’t a bug at the margins—it’s a core behavioral flaw. The fact that the UK is handing the keys of public service infrastructure to systems with such brittle reliability is not just reckless—it borders on absurd.

The DeepMind Discovery: Confidence Is Not Competence

According to DeepMind’s research, LLMs display an unsettling pattern of overconfidence when they are factually incorrect. Worse still, they can be easily manipulated into abandoning correct answers when challenged, creating a dynamic that mimics insecurity masked by bluster. This matters a great deal when the model is generating a casual poem or helping someone brainstorm dinner ideas. But it becomes potentially catastrophic when the model is offering guidance on school placement decisions, sentencing suggestions, or flagging individuals for investigation.

The problem isn’t just the errors. It’s the way those errors are delivered—with the calm, assured tone of a seasoned professional. People, especially those unfamiliar with how LLMs work, tend to trust answers that sound confident. This is a deeply human cognitive bias that LLMs are perfectly poised to exploit—unintentionally, but relentlessly. Embedding these systems into government decision-making risks creating a dangerous feedback loop, where flawed outputs are treated as authoritative, simply because they sound authoritative.

Public Infrastructure Is No Place for Fragile Logic

When an AI system gives the wrong answer in a chatbot, it might be annoying. When it gives the wrong answer in a benefits appeal, a criminal trial, or an immigration case, the consequences can be life-altering. Public services don’t just require speed and efficiency—they demand consistency, accountability, and legal appeal structures. LLMs, as they currently stand, are not capable of meeting those standards without substantial human oversight.

Unfortunately, the allure of automation often overrides caution. Bureaucratic systems love the promise of AI because it suggests a world where complaints, bottlenecks, and paperwork all disappear under a digital tide. But as history shows, the more a system is automated, the harder it becomes to challenge when it goes wrong. If OpenAI’s models are wired into frontline services, and those models produce false but confident outputs, we’re building a system that’s fast, sleek—and quietly unaccountable.

The truth is that no matter how elegant the interface or efficient the rollout, fragile reasoning doesn’t scale. And yet, that’s exactly what’s happening. We’re scaling brittle logic with full knowledge of its limitations.

The Copyright Question: Who Owns the Inputs?

Another layer of concern lies in the very data that trained these systems. OpenAI’s generative models were trained on massive corpora of text, images, videos, and music—much of which was scraped from the internet without consent. Musicians, writers, visual artists, and filmmakers have raised alarm bells over the unlicensed use of their work to fuel the capabilities of these tools. While OpenAI insists that training data is anonymized and aggregated, that argument doesn’t wash when the model starts producing work that echoes—and sometimes outright replicates—the original inputs.

If the UK’s justice system starts using AI to draft judgments, and that AI was trained on copyrighted case law or legal briefs written by private barristers, who owns the output? If an education tool produces teaching materials that bear uncanny resemblance to a specific textbook, what legal protections exist for the original authors? These aren’t theoretical questions. They are legal and ethical minefields that the current AI rush seems determined to ignore in the name of innovation.

When the foundations of a system are ethically compromised, it undermines trust in every layer built upon it. And once that trust is lost, it’s nearly impossible to rebuild.

Hallucinations Are Not Just Bugs—They’re Features

Another well-documented flaw of LLMs is their tendency to hallucinate—generating plausible but completely fabricated information. These hallucinations aren’t rare edge cases. They happen frequently, especially when a model is asked to generate specific data, references, or policy explanations. In public-facing systems, these fabrications can do real harm.

Imagine a government chatbot confidently stating that a person has no right to appeal a decision—when in fact they do. Or an education tool explaining a scientific concept incorrectly, leading to widespread misunderstanding. Or a legal support AI misquoting precedent. These aren’t harmless glitches. They are high-stakes failures delivered with an air of certainty.

The worst part? The very structure of LLMs makes them look reliable. Their fluency and grammar create a façade of expertise. But under the hood, it’s just token prediction—an autocomplete engine with a god complex. That may sound harsh, but it’s the reality we must confront before handing these tools the keys to our institutions.

AI Is a Tool, Not a Truth Engine

What’s emerging here is a dangerous conflation: we are mistaking fluency for understanding, and confidence for correctness. Just because a model can generate text that reads like it came from a lawyer, a teacher, or a government official doesn’t mean it has any actual comprehension. It’s mimicry, not mastery. And yet the political class seems entranced by the illusion.

This is the essence of the cliff we’re walking off. We’re not being pushed. We’re marching forward, eyes wide shut, enchanted by the spectacle of “AI nation building.” The issue isn’t that AI has no place in public life. It’s that it’s being treated as a finished product, a mature technology, rather than what it really is: a prototype with unpredictable edges.

PR Blitz vs. Ground Truth

Why is this happening now, despite the warnings? Because governments are desperate. The UK economy is stagnant, growth projections are bleak, and ministers are hungry for a narrative of transformation. In that context, AI becomes a seductive solution. It sounds futuristic, investor-friendly, and globally competitive. It also offers a welcome distraction from structural issues no one wants to fix.

So deals get signed. Memorandums of understanding are drafted. Speeches are made about “prosperity for all.” Meanwhile, behind the scenes, researchers are waving red flags—and getting largely ignored.

There’s a performative aspect to AI policy that’s hard to overlook. It’s less about solving real problems, and more about being seen to be doing something bold. The tragedy is that this performative urgency could lead us to embed faulty, biased, or misleading systems into the very fabric of governance.

We Still Have Time to Step Back

The technology is not the enemy here. Nor are the researchers or even the companies pushing it forward. The real threat lies in uncritical adoption and political opportunism. There is still time to apply the brakes, to insist on rigorous testing, transparency, and a slower, saner rollout of AI systems in government.

If this deal is to be worth anything, it must come with independent oversight, publicly accessible audits, and genuine opt-out mechanisms for the citizens it affects. Anything less is a betrayal of the democratic values the MoU claims to uphold.

We have the data. We have the warnings. We have the expertise. What we need now is the courage to say: Not yet. Not like this.

Promotional image for “100 Greatest Science Fiction Novels of All Time,” featuring white bold text over a starfield background with a red cartoon rocket.
Read or listen to our reviews of the 100 Greatest Science Fiction Novels of all Time!