A Country of Geniuses in a Data Center Doesn't Just Need Your Kid to Think. It Needs Them to Lead.
We Are Near the End of the Exponential
On February 13, 2026, Dario Amodei — the CEO of Anthropic, one of the three companies building the most powerful AI systems on Earth — said something that should be front-page news in every school district in the country. He said we are “near the end of the exponential.” That within a few years, we will have what he calls “a country of geniuses in a data center” — AI systems that can perform at or beyond the level of the best human experts across virtually every cognitive domain. He put himself at 90% confidence. Not someday. Not by mid-century. Within this decade.
And then he said something even more striking: the most surprising thing to him is “the lack of public recognition of how close we are to the end of the exponential.” People are still arguing about the same tired issues (think: AI. writing school papers and writing detectors), he said, while the most transformative technology in human history barrels toward completion.
He’s not being hypothetical. The evidence landed the day before he spoke.
On February 12, Google released an upgraded Gemini 3 Deep Think — and the numbers should make every educator sit down.
84.6% on ARC-AGI-2. That benchmark isn’t a trivia test. It isn’t pattern matching. ARC-AGI-2 — the Abstraction and Reasoning Corpus for Artificial General Intelligence — was specifically designed to measure the thing AI was supposed to be bad at: fluid intelligence. Abstract reasoning. The ability to look at a novel problem you’ve never seen before, with minimal context, and figure it out. Every task in ARC-AGI-2 was verified solvable by human participants. Previous AI models struggled to break 20%. Humans average about 60%. Gemini 3 Deep Think just scored 84.6 — crushing the human baseline on a test designed to expose the gap between human and machine reasoning.
It hit a 3,455 Elo rating on Codeforces, placing it among the top competitive programmers on Earth. It earned gold-medal-level scores on the International Physics Olympiad (87.7%) and Chemistry Olympiad (82.8%). It found a logical flaw in a peer-reviewed mathematics paper that human reviewers had missed. It took a hand-drawn sketch, analyzed the geometry, and generated a 3D-printable file.
And that’s just one number. Deep Think scored 48.4% on Humanity’s Last Exam — a benchmark sourced from the hardest questions across every academic discipline — without using any tools.
This isn’t a chatbot anymore. This is a reasoning engine that outperforms most humans on tests specifically built to measure general intelligence.
And Amodei says this is not the peak — it’s a point on a curve that is still climbing, fast. The underlying exponential in AI capability, he says, has proceeded roughly as he expected since 2019: models progressing from smart high school student to smart college student to PhD-level work, and in coding, already beyond that. The frontier is uneven but relentless.
Meanwhile, all three frontier labs — Google, Anthropic, OpenAI — shipped major new model upgrades in a single week. The gap between releases has collapsed from months to days. As one commentator put it: the competition has moved from chat to cognition. The machines are no longer competing to sound intelligent.
They are competing to reason through ambiguity, explore multiple hypotheses, and sustain long chains of structured thought.
Amodei frames this as two exponentials running in parallel: one in raw model capability, one in the diffusion of that capability into the economy. AI is already writing 90% of the code at Anthropic. His company’s revenue grew from zero to $100 million, to $1 billion, to $9–10 billion in consecutive years — and added several billion more in January 2026 alone. The “country of geniuses” isn’t an abstraction. It’s a business plan being executed in real time.
That description — reasoning through ambiguity, exploring multiple hypotheses, sustaining structured thought — is not just a description of artificial intelligence. It is a description of what happens in every competitive debate round in every high school gym and college auditorium in this country.
And it raises a question we can no longer avoid: if the machines are racing to do what we train to students to do — and the CEO of one of the world’s leading AI companies says we are “near the end” of that race — what happens to the humans who never learned to do it at all?
The World Has Changed. Education (Largely) Hasn’t.
Here’s the uncomfortable truth. The median American student is being trained for a world that no longer exists. They memorize content that any AI can retrieve in milliseconds. They write five-paragraph essays that ChatGPT can produce in five seconds. They take standardized tests that measure recall in an era where recall is free.
But here’s the part that should truly unsettle you: it’s not just the rote skills that AI has eaten. The machines can now construct arguments. They can anticipate objections. They can weigh competing values, reason under uncertainty, and generate persuasive prose on demand.
The things we told ourselves were “higher-order thinking” — the skills that were supposed to be safe from automation — are falling too.
So what’s left?
Humans.
Not human thinking in the abstract — the machines are closing that gap fast. What’s left is human connection. The ability to look another person in the eye and build trust. To walk into a room where people are angry, confused, or afraid, and change the temperature. To listen — not process, not parse, not generate a response token — but genuinely listen to another human being and make them feel heard. To negotiate not just with logic but with empathy, timing, and presence. To persuade not through optimization but through relationship.
AI can draft the argument. It cannot deliver it to a skeptical school board at 9 PM on a Tuesday, reading the exhaustion in the room, adjusting its tone, and earning trust from people who came in hostile. It cannot sit across from a grieving client and know when to push and when to stay silent. It cannot stand beside a colleague who just got terrible news and say the right thing — or say nothing at all — in a way that matters.
And even when it can — even when the AI’s words are technically flawless, its tone pitch-perfect, its logic airtight — it will still not move people the way another human does. This is not a limitation that engineering will solve. It is a feature of being human. Squirrels are moved by squirrels. Humans are moved by humans. We are wired — by millions of years of evolution, by every childhood bond, by every loss and reconciliation that shaped us — to respond to the presence of another person in ways we will never respond to a machine, no matter how sophisticated. The quiver in a voice matters. The courage it takes to stand up matters. The fact that the person across from you chose to show up, chose to be vulnerable, chose to fight for something in front of you — that matters in a way that no generated output ever will.
This is not sentimentality. It is biology. It is sociology. It is the foundation of every institution humans have ever built — from families to nations, from courtrooms to coalitions. Leadership, persuasion, and trust are not information problems. They are presence problems. And presence requires a human being.
These are not soft skills. In a world where cognitive labor is increasingly automated, they are the hardest skills. They are the skills that will determine who leads, who negotiates peace, who holds communities together, who closes the deal, who earns the vote, who keeps the team from falling apart.
Speech and debate is one of the only activities in American education that systematically trains all of them. It doesn’t just teach students to think — the machines can do that now. It teaches them to think with and in front of other people. To construct arguments, yes — but to deliver them while reading a judge’s skepticism. To anticipate objections — but from a live opponent who is adapting in real time, not a static dataset. To weigh competing values — but in a format where you must convince an actual human being, not just generate a high-scoring output.
The rooms where it counts — the boardroom, the courtroom, the negotiation table, the policy hearing, the crisis meeting — are rooms full of people. People with emotions, biases, histories, and fears. The person who thrives in those rooms won’t be the one who can out-reason a machine. It will be the one who can out-relate to every other human in it. And increasingly, those rooms are where the stakes of AI itself will be decided.
The Rifts Are Coming. Debate Is How We Survive Them.
Here’s what almost nobody in education is talking about: AGI doesn’t just disrupt jobs. It fractures societies.
We’re already seeing the fissures. The AI industry is bifurcating into two tiers: fast, general-purpose models for everyday productivity, and slower, deeply capable reasoning engines for science, engineering, and high-stakes domains. Google is explicitly building Deep Think not for speed but for discovery — gated behind Ultra subscriptions and early-access API programs. Two tiers of intelligence. Two markets. Two price points.
Now project that onto society. One tier of people will use AI as a passive tool — prompting it, consuming its outputs, trusting its conclusions. The other tier will use AI as a partner — challenging its reasoning, identifying its blind spots, directing its capabilities toward problems that matter. The gap between those two groups will widen into a chasm, and that chasm will generate conflict the likes of which we haven’t seen since industrialization.
Think about what’s already tearing at the seams. Workers displaced by automation who feel invisible. Communities divided over whether AI is a liberation or a threat. Parents who don’t understand what their children’s screens are doing to them. Nations racing toward superintelligence while their citizens can’t agree on basic facts. The societal rifts that AGI will deepen — economic inequality, political polarization, generational divides, urban-rural splits, global power imbalances — are not technical problems. They are human problems. They are problems of persuasion, negotiation, empathy, and the ability to find common ground when everything incentivizes division.
RAND researchers are already publishing urgent analyses of how the United States and China should navigate the path to superintelligence. One proposed framework, MAIM (Mutual Assured AI Malfunction), essentially institutionalizes the threat of preventive strikes against the other side’s AI infrastructure — destroying data centers and chip fabrication plants to prevent a technological lead. An alternative, what RAND’s Michael Mazarr calls “Resilient Equilibrium,” argues for mutual restraint, assurance, and resilience. Instead of threatening to attack, nations would invest in understanding, transparency, and the societal capacity to absorb shocks.
The difference between these two frameworks is fundamentally a debate about how to argue, persuade, and build trust under conditions of radical uncertainty. MAIM says: threaten and coerce. Resilient Equilibrium says: communicate, reassure, and build shared frameworks.
That’s not just geopolitics. That’s the template for every societal rift AGI will generate. In every community, every workplace, every family — people will face wrenching disagreements about how to live alongside increasingly powerful machines. The people who can navigate those disagreements without resorting to tribalism, demonization, or paralysis will be the ones who hold society together.
Speech and debate trains exactly this. Debaters learn to argue positions they disagree with — not as an academic exercise, but as a discipline for understanding why the other side believes what they believe. They learn to find the strongest version of an opposing argument before attacking it. They learn that persuasion requires meeting people where they are, not shouting from where you stand.
In a society fracturing under the pressure of technological change, those aren’t just nice skills. They’re load-bearing.
The Techlash Is Here. Human Connection Is the Counterweight.
There’s another dimension to this that the AI conversation almost entirely ignores: the human one.
This week, a researcher resigned from Anthropic warning that “the world is in peril” from interconnected crises, and that he’d seen how hard it is to let values govern actions. Days later, another researcher left OpenAI over the company’s decision to introduce advertising into ChatGPT, warning that ads built on users’ most private thoughts create potential for manipulation we can’t yet comprehend. The backlash against technology’s intrusion into every corner of human life is accelerating, and it’s not irrational. People sense — correctly — that something is being lost.
What’s being lost is human-to-human skill. The ability to read a room. To hear what someone means beneath what they say. To hold eye contact during a difficult conversation. To modulate your voice, adjust your argument, respond to emotion in real time. To stand in front of people who disagree with you and hold your ground — or change your mind — with grace.
AI doesn’t build these skills. It atrophies them. Every hour spent prompting a chatbot instead of persuading a person, every meeting replaced by an AI summary, every difficult conversation avoided because a bot can draft the email — these are reps we’re no longer doing in the gym of human interaction. And like any muscle, persuasion and presence atrophy without use.
Speech and debate is the most powerful counterweight to this erosion in American education. It is one of the last structured activities where students must stand up, face other human beings, and perform cognition live. Not through a screen. Not mediated by an algorithm. Not with the ability to edit, delete, and regenerate.
In a debate round, you cannot hit “undo.” You cannot ask the AI to rewrite your rebuttal. You must listen — actively, intensely — to what the other person is saying, process it in real time, and respond with arguments that are both logically sound and humanly persuasive. You must read the judge’s body language. You must manage your own nerves. You must think and speak and connect, simultaneously, under pressure.
These are not quaint skills from a pre-digital era. In a world where every mediated interaction can be automated, the ability to be compelling in unmediated human interaction becomes the scarcest and most valuable capacity a person can have. The techlash isn’t going away. The pendulum will swing back toward valuing human presence, human judgment, and human connection. The students who have trained those muscles — thousands of hours of rounds, drills, cross-examinations, and impromptu speeches — will be the ones who thrive when it does.
The Ethics Crisis Is a Human Relations Crisis
The ethical challenges surrounding AI are not going to be resolved by smarter algorithms or better technical guardrails. They will be resolved — or not — by human beings sitting across from other human beings, navigating disagreements about values, power, and the future.
Consider what the AI ethics landscape actually looks like right now. It’s not a logic puzzle waiting to be solved. It’s a web of relationships under strain. Employees confronting their own executives about safety — and sometimes walking out the door. Regulators trying to understand technology that evolves faster than they can write policy, needing to build trust with industry leaders who speak a different language. Communities demanding a voice in decisions that will reshape their livelihoods, facing panels of experts who struggle to listen. Nations that must cooperate on existential risk while competing for strategic advantage. Parents trying to set boundaries with teenagers who understand the technology better than they do.
Every one of these is a human relations problem. Not a reasoning problem. Not a technical problem. A problem of trust, empathy, persuasion, and the ability to hold difficult conversations without the relationship collapsing.
An AI can draft the perfect policy memo on AI safety. It cannot look a room full of laid-off workers in the eye and explain why this transition matters and what comes next — and make them believe it. It cannot sit in a Congressional hearing and build rapport with a skeptical senator who doesn’t trust Silicon Valley. It cannot mediate between a terrified parent and a defiant teenager over screen time and chatbot use. It cannot hold the tension between two colleagues who fundamentally disagree about whether their company’s AI product is safe to ship.
Speech and debate trains students for exactly these moments. Policy debaters don’t just weigh harms and benefits in the abstract — they defend those positions to a live judge whose skepticism they must overcome through credibility, not just logic. Lincoln-Douglas debaters don’t just clash over values — they learn to articulate why a value matters to someone who doesn’t share it, building a bridge rather than scoring a point. Public forum debaters argue both sides of pressing current events — and in doing so, develop the rare capacity to genuinely understand a position they oppose, which is the foundation of every successful negotiation, mediation, and act of leadership.
In a world where an AI can generate a thousand words of confident-sounding argument in two seconds, the ability to detect hollow rhetoric is essential. But the deeper skill — the one that will actually determine whether we navigate the AI transition without tearing each other apart — is the ability to sit with another person who sees the world differently, hold the discomfort, and find a way forward together. That’s not reasoning. That’s relationship. And speech and debate is one of the only places left in American education where students practice it, under pressure, every single week.
The Case Is Urgent
Let me be direct. We are building a country of geniuses in a data center. Those geniuses will write the code, pass the exams, draft the memos, and generate the analyses. They will not — cannot — do the thing that matters most: walk into a room full of human beings and lead.
Leadership is not a cognitive task. It is a relational one. It is earning trust from people who have no reason to give it. It is holding a team together when the mission is unclear and the stakes are high. It is standing in front of a hostile audience and turning the temperature down, not through better arguments, but through presence, credibility, and the willingness to listen before you speak. It is knowing when to push and when to yield, when to fight and when to build a bridge — and making that judgment in real time, face to face, with no algorithm to fall back on.
Every year that we invest in training students to pass tests a machine can already ace, and fail to invest in training them to connect with, persuade, and lead other human beings, we fall further behind the moment we are living in.
If you are a school administrator cutting speech and debate to save money, understand what you are cutting: the one activity that puts students in front of live opponents, live judges, and live audiences — week after week, year after year — and forces them to practice the human skills that no data center will ever replicate. You are not trimming a line item. You are eliminating your students’ best preparation for the world Dario Amodei just described.
If you are a parent wondering whether debate is “worth it” compared to another AP class, ask yourself this: in a world where a machine can score gold on the Physics Olympiad, what is the value of one more test score? And what is the value of a teenager who can stand up, look another person in the eye, and make a case that moves people — not because the logic is optimized, but because the human delivering it has learned to connect?
If you are a policymaker or philanthropist, speech and debate is one of the highest-leverage investments in American education. It is inexpensive. It scales. It serves every demographic. And it is one of the only activities that systematically builds the leaders, negotiators, and communicators that a fractured, AI-transformed society will desperately need.
The machines can reason, discover, and outperform Olympic-level humans on tests designed to measure pure intelligence. The country of geniuses is arriving on schedule. The question is no longer whether AI will be smart enough. It’s whether our kids will be human enough — connected enough, courageous enough, present enough — to lead in the world those geniuses create.
Speech and debate is where that happens. It has been for a hundred years. The difference is that now, for the first time, the stakes are existential.