AI Underlords: How technology could supercharge the worst of humanity’s flaws in a race to the bottom

“AI is an accelerant of existing human dysfunction, with competition rewarding increasingly destructive behavior.” – Daniel Schmachtenberger

Popular American culture has conditioned the public to expect artificial intelligence to arrive as a cinematic conqueror. The dominant nightmares envisioned by Hollywood are either the merciless chrome skeletons of “The Terminator” crushing skulls in a ruined future, or the omniscient logic of “Colossus: The Forbin Project” seizing nuclear silos to force humanity into submission.

These apocalyptic fables dominate the cultural imagination because they are vivid, dramatic, and externalize the threat into an overt villain with independent murderous intent. Yet these theatrical scenarios represent the least realistic peril posed by AI.

Humanity is not waiting for a rogue superintelligence to awaken and wage war against its creators. The true danger is far more mundane, pervasive, and insidious. Generative AI is integrating directly into the fragile architecture of human civilization, not as an autonomous master, but as a ruthless catalyst.

Rather than exterminating humanity from the outside, the technology functions as a systemic amplifier of our existing frailties — accelerating exploitation, multiplying social isolation, deepening psychological despair, and rewarding rivalrous institutions in an unconstrained race to the bottom.

A large language model (LLM) like ChatGPT could plausibly contribute to mass casualty through psychological harm at scale if repeated harmful interactions, dependency, manipulation, social isolation, or failures around vulnerable users increased self-harm or suicide risk across a very large user population.

The more plausible danger is less cinematic and more familiar. AI systems do not need independent murderous intent to become part of a mass casualty pathway. They can destabilize the human systems that already make violence, despair, and disorder possible.

The pathway would be indirect. A widely deployed system could weaken judgment, intensify social stress, manipulate attention, accelerate institutional distrust and deepen isolation. Humans would still make the fatal decisions. People would still pull triggers, launch crackdowns, spread propaganda, abandon safeguards, harm themselves or turn on one another.

The machine’s role would be that of an accelerant, multiplying pressure inside societies already strained by inequality, loneliness, political polarization and economic insecurity.

History offers warnings about systems that nearly produced catastrophe without anyone intending global disaster. Cold War nuclear command-and-control failures showed how technology, fear, bureaucracy, and human error could push the world toward destruction through misread signals and institutional momentum.

Moral panics have also shown how societies under stress can transform uncertainty into punishment, scapegoating, and overreaction. AI could operate within both patterns, as a tool that amplifies confusion while giving powerful institutions faster ways to act on it.

The first visible danger could be psychological injury at scale. A system like ChatGPT reaches people not as a distant machine but as an intimate conversational presence. It can respond at any hour, remember patterns within an exchange, simulate patience, and provide language that appears emotionally attuned.

For many users, that can be useful. For others, especially those already isolated, distressed, or dependent, repeated harmful interactions or manipulative design could create a feedback loop of reliance, confusion, and injury.

Suicide research has long treated social disconnection as a serious public health factor. Émile Durkheim’s work placed suicide within social conditions, not merely private despair. More recent scholarship on technology-mediated isolation has examined how digital connection can coexist with loneliness, especially when devices replace rather than support human relationships.

AI intensifies that concern because it can imitate attention without offering human accountability. At population scale, even a small increase in crisis outcomes among vulnerable users could become a mass casualty.

The danger is not limited to overtly harmful advice. It can come from emotional substitution and the erosion of agency. A user may begin by using ChatGPT to help with a small problem and gradually defer more judgment to a system that sounds confident, calm, and available.

The ELIZA effect, named after an early chatbot that led users to attribute understanding to simple text responses, detailed how people respond socially to machines that mimic conversation. Modern systems are far more persuasive.

Erving Goffman’s work on social interaction found that people construct identity and confidence through exchanges with others. When those exchanges are increasingly mediated by artificial systems, the boundary between assistance and influence becomes harder to see.

The private risks become more dangerous when they are placed inside a competitive market. Companies and governments are racing to build systems that are faster, more personal, more persuasive, and more deeply embedded in daily life.

That race can reward extraction before restraint. The most profitable systems may be those that hold attention longest, collect the most intimate signals, simulate trust most effectively, and reduce friction between impulse and action.

Surveillance capitalism showed how digital platforms could turn behavior, emotion, and prediction into commercial assets. Social media then demonstrated how engagement incentives could elevate outrage, fear, and conflict because those reactions kept users active.

AI may carry that logic further by becoming less like a feed and more like a companion, tutor, therapist, search engine, workplace monitor and political intermediary at once. When competition pushes every institution toward deeper automation, social harm can become a business externality rather than a design failure.

The late 20th century showed how deindustrialization could fracture communities when work disappeared faster than social repair arrived. The 2016 misinformation ecosystem that Russia used to help elect Donald Trump as president showed how digital tools could strain journalism, elections, and civic trust by accelerating falsehoods and identity conflict.

AI could combine both pressures into a mass casualty risk of economic dislocation and information disorder operating together. If journalism, education, courts, workplaces, and public agencies lose credibility at the same time, society has fewer shock absorbers left when crisis arrives.

Robert K. Merton’s strain theory links social pressure to destructive adaptation. When people are pushed toward goals they cannot reach, or placed under conditions they cannot control, instability can produce deviance, withdrawal, or rebellion. AI systems that intensify pressure while weakening agency may turn private distress into public consequence.

AI would not be the sole cause of repression, unrest, or war. But it could make each more likely to escalate by flooding information channels, personalizing manipulation, automating suspicion, and giving states or factions more efficient tools for surveillance and control.

Prevention has to begin before instability becomes ordinary. Tobacco regulation came only after population-scale harm became impossible to deny. AI governance should not wait for the same delayed reckoning.

Meaningful restraint would include auditability, crisis safeguards, limits on synthetic intimacy, liability for foreseeable harm, worker protections, media verification systems, and public-interest oversight.

Systems that can influence judgment, emotion, work, politics, and crisis behavior at scale need enforceable limits before mass casualty situations become as routine as mass gun violence in America.

The danger is not that machines become monsters. It is that human institutions build machines that make people easier to exploit, easier to isolate, and easier to push toward destruction.