The Real Threat to Humanity Isn’t AI — It’s Us

We’ve spent millennia making catastrophic decisions. Maybe it’s time to get some help.

The headlines are relentless. Artificial intelligence will destroy jobs, undermine democracy, and possibly end civilization. Tech leaders sign open letters warning of “extinction risk.” Politicians scramble to regulate something most of them don’t understand. The message is clear: AI is the existential threat of our time.

But what if we’re looking at this completely backward?

What if the real danger isn’t artificial intelligence — but the natural kind? What if the greatest risk to human flourishing has always been, and continues to be, human decision-making itself?

A Species with a Track Record

Let’s be honest about our history.

According to a comprehensive study by the Center for International and Security Studies at Maryland, approximately 231 million people were killed or “allowed to die by human decision” during the 20th century alone. This figure encompasses wars, genocides, political famines, and institutional violence — all the result of human choices made by human minds.

The two World Wars accounted for roughly 75% of all war deaths since 1800. Three-quarters of all military casualties in two centuries, concentrated in just two conflicts that lasted a combined eleven years. The decisions that started these wars — made by human leaders operating with incomplete information, tribal loyalties, and personal ambitions — altered the course of human history.

But here’s the most chilling statistic: we came within minutes of nuclear annihilation, stopped not by wisdom or foresight, but by the conscience of a single man.

On October 27, 1962 — the most dangerous day of the Cuban Missile Crisis — Soviet submarine B-59 was trapped beneath the surface near Cuba, hunted by American destroyers. The submarine carried a nuclear torpedo. Its captain, believing war had begun, prepared to launch. Under protocol, two officers needed to authorize the strike. Both agreed.

But B-59 had a third officer aboard: Vasili Arkhipov, the brigade’s chief of staff. Against tremendous pressure, Arkhipov refused to authorize the launch. He persuaded the captain to surface instead. As Thomas Blanton, director of the U.S. National Security Archive, later said: “A guy called Vasili Arkhipov saved the world.”

The survival of human civilization hung on one person’s judgment in a heated moment of crisis. This is not a stable system.

The Financial Cost of Human Error

It’s not just war. Human decision-making gave us the 2008 financial crisis — a catastrophe rooted in overconfidence, herd behavior, and systematic failure to assess risk.

The numbers are staggering:

• The U.S. lost $3.4 trillion in real estate wealth from July 2008 to March 2009
• $7.4 trillion in stock wealth evaporated in the same period
• A Federal Reserve study found the crisis cost every American approximately $70,000 in lifetime income
• Better Markets estimated the total cost to the U.S. economy at over $20 trillion in lost GDP
• 5.5 million more jobs were lost than what would have occurred without the crisis

The causes? Human decisions driven by short-term incentives, confirmation bias, and overconfidence. Mortgage lenders approved loans they knew borrowers couldn’t repay. Rating agencies stamped AAA on securities they didn’t understand. Executives dismissed warning signs that contradicted their beliefs. Regulators trusted the market to police itself.

These weren’t stupid people. They were highly educated professionals operating at the top of their fields. They failed because they were human — subject to the same cognitive limitations that affect us all.

The Medical Cost of Human Cognition

Even in healthcare — where careful analysis literally determines life and death — human limitations take a devastating toll.

Research published in the Journal of Patient Safety estimates that preventable medical errors contribute to approximately 200,000–400,000 deaths annually in the United States. A 2023 Johns Hopkins study found that 795,000 Americans die or are permanently disabled by diagnostic errors each year.

The most common causes? Cognitive errors:

• Anchoring bias: Physicians lock onto initial diagnoses and fail to revise when evidence suggests otherwise
• Premature closure: Doctors stop considering alternatives once they reach a plausible diagnosis
• Confirmation bias: Clinicians interpret new information in ways that support their existing conclusions

A meta-analysis in the PMC journal found that 71% of studies showed associations between cognitive biases — including anchoring, overconfidence, and confirmation bias — and therapeutic or management errors.

These aren’t character flaws. They’re features of human cognition that even highly trained professionals cannot reliably override.

Why We Decide So Badly

The problem isn’t that humans are evil or careless. The problem is that our cognitive architecture evolved for a world that no longer exists.

Our processing power is fundamentally limited. George Miller’s famous 1956 paper “The Magical Number Seven, Plus or Minus Two” established that human working memory can hold only about 7 items (±2) at once. Later research suggests the true limit may be closer to 3–4 items for unfamiliar material, and information in working memory begins to decay after just 2–3 seconds.

Compare this to decisions about climate policy, pandemic response, or financial regulation — problems with thousands of interacting variables, timeframes spanning decades, and consequences affecting billions of people. We’re trying to navigate a global civilization with a mental workspace designed for tracking a few threats on the savanna.

We systematically discount the future. Research on temporal discounting shows humans consistently undervalue future outcomes compared to immediate ones. A dollar today feels more valuable than ten dollars next year. This isn’t irrationality — it was adaptive when our ancestors faced immediate survival pressures and uncertain lifespans. But it’s catastrophic when applied to climate change, infrastructure investment, or pandemic preparedness.

We’re overconfident in ways we can’t detect. Studies show financial executives’ predictions fall within their stated confidence intervals only 38% of the time — meaning they’re wrong about their own accuracy almost two-thirds of the time. Research on CEO overconfidence demonstrates that executives systematically overestimate their control over outcomes and underestimate risks, affecting investment decisions, merger activity, and strategic planning. The Dunning-Kruger effect compounds this: less competent individuals are often more overconfident, because they lack the metacognitive ability to recognize their own limitations.

We think in tribes. In-group favoritism and out-group suspicion are deeply wired. This made sense when cooperation with your band and wariness of strangers enhanced survival. It makes less sense when effective governance requires global cooperation among people who will never meet.

We respond to narratives, not statistics. A single vivid story can override mountains of data. One child trapped in a well mobilizes millions in donations; statistics about millions dying from preventable diseases generate limited response. This is why terrorism dominates our fears despite killing far fewer people than car accidents or heart disease.

These aren’t bugs we can simply decide to fix. They’re features of human cognition — deep, persistent, and largely invisible to us. We don’t notice our biases because our biased brains are the ones doing the noticing.

The Promise of Augmented Decision-Making

Here’s where AI becomes interesting — not as a replacement for human judgment, but as an enhancement to it.

Consider what AI systems can do that humans struggle with.

In mammography screening, a 2020 study in Nature found that an AI system reduced false-positive rates by 3.5% and false-negative rates by 8.1% compared to human radiologists. A 2023 study showed that AI-assisted radiologists achieved AUROC scores of 0.940 compared to 0.810 for radiologists working alone — a statistically significant improvement.

A large-scale German study involving 463,094 women and 119 radiologists found that AI-supported screening increased breast cancer detection rates significantly without increasing false recalls. In prostate cancer detection, AI demonstrated AUROC of 0.91 versus 0.86 for radiologists and detected 6.8% more significant cancers at the same specificity level.

What’s notable is that human-AI teams consistently outperform humans alone. The AI catches what humans miss; humans provide context the AI lacks. Neither alone is optimal; together, they achieve something neither could accomplish independently.

This isn’t limited to medicine. AI systems can:

• Process vast amounts of information without fatigue
• Identify patterns across datasets too large for human review
• Model complex systems with thousands of interacting variables
• Maintain consistency across decisions that humans would make inconsistently
• Flag when decisions show signatures of known cognitive biases

Most importantly, they don’t have bad days, ego investments in being right, or tribal loyalties that distort their analysis.

Reframing the Conversation

None of this means AI is without risks. Like any powerful technology, it can be misused. It can encode and amplify existing biases if we’re not careful. It can be deployed in ways that concentrate power rather than distribute it. These are real concerns that deserve serious attention.

But the framing matters.

When we treat AI as an existential threat and human judgment as the safe default, we get the equation exactly wrong. Human judgment is not the safe default. Human judgment is how we got 231 million deaths in the 20th century. Human judgment is why nuclear war was prevented by one dissenting officer in a submarine. Human judgment is why the 2008 financial crisis cost Americans $70,000 per person in lifetime income.

The question isn’t whether to trust AI or humans. It’s how to combine them in ways that compensate for the weaknesses of each.

Humans bring values, creativity, empathy, and moral reasoning. We understand context, appreciate nuance, and can navigate situations that have never occurred before. We know what we care about and why.

AI brings processing power, consistency, and freedom from many cognitive biases. It can hold thousands of variables in mind simultaneously. It doesn’t get tired, hungry, or defensive. It can model consequences across time horizons that human intuition cannot grasp.

Together, they might make decisions that neither could make alone.

A Humble Proposal

What would it look like to take this seriously?

It would mean requiring AI-assisted impact assessments for major policy decisions — not to replace human judgment, but to inform it. Not “AI decides,” but “AI shows humans the second and third-order effects they might otherwise miss.”

It would mean building AI tools that flag when our tribal instincts or confirmation biases might be leading us astray. Decision-support systems that say, “You’ve dismissed three pieces of evidence that contradict your conclusion — would you like to reconsider?”

It would mean creating AI systems that help leaders understand the long-term consequences of short-term choices. Climate models, economic simulations, epidemiological projections — not as replacements for human values, but as extensions of human foresight.

We don’t worry that telescopes will “replace” human astronomers. We understand that they help us see what we couldn’t see before. AI can help us think what we couldn’t think before.

The Choice We Face

Here’s the uncomfortable truth: we’re not going to evolve our way out of our cognitive limitations. Our brains aren’t going to suddenly become better at holding thousands of variables in working memory, or reasoning across century-long time horizons, or overcoming in-group biases that have been adaptive for millions of years. We are what we are.

But we don’t have to remain what we are alone.

For the first time in history, we have tools that can augment human cognition at scale. We can build systems that help us see our blind spots, check our assumptions, and think through consequences we’d otherwise ignore. We can become something more than what evolution made us.

The question is whether we’ll be wise enough to use them.

Given our track record — 231 million dead in a century, civilization nearly destroyed by decision-making under pressure, trillions of dollars lost to overconfidence and herd behavior — I wouldn’t bet on human wisdom alone.

But human wisdom augmented by artificial intelligence? That might actually have a chance.

The greatest threat to humanity isn’t the intelligence we’re building. It’s the intelligence we’ve been relying on all along.

Key Statistics at a Glance

Category Data Point Source 20th Century Deaths from Human Decisions ~231 million Center for International and Security Studies at Maryland 2008 Financial Crisis Cost (per American) ~$70,000 lifetime income Federal Reserve Board Study U.S. Medical Error Deaths (annual) 200,000–400,000 Journal of Patient Safety Human Working Memory Capacity 3–7 items, ~2–3 second decay Miller (1956), Cowan (2001) Executive Prediction Accuracy 38% within confidence interval National Bureau of Economic Research AI vs. Human Radiologist (Mammography AUROC) 0.940 vs. 0.810 Lancet Digital Health Near-Nuclear War Events Multiple (B-59, 1983 Soviet alert, others) National Security Archive

If you found this perspective valuable, I’d love to hear your thoughts. What role do you think AI should play in human decision-making? Where are the lines we shouldn’t cross?

About the Author

Brian Curry is a Kansas City-based Data Scientist, AI researcher, futurist, theorist, and founder of Vector1 Research, where he explores the frontier of Artificial Intelligence, Causal Inference, Agentic Systems, and Economic Modeling.

His work focuses on intelligent systems, future state technologies and systems, and human and AI integrations that reveal the causal structures driving performance across marketing, content, and economics. He is the creator of PyCausalSim (causal discovery through simulation), Papilon (complex system optimization), Daedalus (causal and economic analysis), and Memory-Node Encapsulation (data structures for artificial episodic memory).

Contact: brian@vector1.ai

Sources and Further Reading

Wars and Conflict

• Center for International and Security Studies at Maryland, “Deaths in Wars and Conflicts in the 20th Century”
• Our World in Data, “War and Peace”

Cognitive Science

• Miller, G.A. (1956). “The Magical Number Seven, Plus or Minus Two”
• Cowan, N. (2001). “The Magical Mystery Four: How is Working Memory Capacity Limited, and Why?”
• Kahneman, D. & Tversky, A. (1974). “Judgment under Uncertainty: Heuristics and Biases”

Financial Crisis

• Federal Reserve Board study on lifetime income effects
• Better Markets, “Cost of the Crisis Report”
• Pew Charitable Trusts, “The Impact of the September 2008 Economic Collapse”

Medical Errors

• James, J.T. (2013). “A New, Evidence-based Estimate of Patient Harms Associated with Hospital Care”
• Johns Hopkins Medicine (2023). “Report Highlights Public Health Impact of Serious Harms from Diagnostic Error”

AI in Healthcare

• McKinney, S.M. et al. (2020). “International evaluation of an AI system for breast cancer screening.” Nature
• Nature Medicine (2025). “Nationwide real-world implementation of AI for cancer detection in population-based mammography screening”

Cuban Missile Crisis

• National Security Archive, “The Underwater Cuban Missile Crisis at 60”
• PBS Secrets of the Dead, “The Man Who Saved the World”