AI로 모든 것을 예측할 수 있게되는 시대의 도래: That’s how we’ll cure disease. AI builds a virtual cell, simulates every possible drug interaction, and identifies the optimal treatment. DeepMind’s AlphaFold already solved the 50-year protein folding problem. That work won the Nobel Prize in Chemistry. And that was just the opening act. / That’s how we’ll cure disease. AI builds a virtual cell, simulates every possible drug interaction, and identifies the optimal treatment. DeepMind’s AlphaFold already solved the 50-year protein folding problem. That work won the Nobel Prize in Chemistry. And that was just the opening act.

 

I’ve been thinking a lot about wisdom lately. Not intelligence. Not speed. Not raw compute. Wisdom. The thing we used to reserve for grandparents, philosophers, and tribal elders who’d seen enough life to know which paths lead to ruin.

Here’s how I define it: wisdom is probabilistic pattern recognition across a vast number of lived experiences. When you go to the village elders and ask, “Which direction should I take?”, they don’t run equations. They draw on decades of watching people make choices and living with the consequences. They say, “If you go this way, based on everything I’ve seen, it won’t end well. Go this other way, and you have a real chance.”

That accumulated experience, compressed into judgment, is what we call wisdom.

Now ask yourself: what happens when an AI can simulate not hundreds or thousands of scenarios, but billions? When it can replay the entire observable history of human decisions, test every fork in the road, and tell you with quantified confidence which path has the highest probability of success?

I think that’s wisdom. And I think we’re watching it be born…

THE FIRST PROOF POINT: FUTURESIM

On this week’s Moonshots podcast, we discussed a new benchmark called FutureSim, built by a group of independent researchers. The architecture is brilliant: it replays the internet day by day starting from January 1, 2026, gives AI agents access to real news as it unfolds, then asks them to forecast real-world events 90 days out, with no web access beyond the replay window. The models can’t cheat by looking up what actually happened.

GPT 5.5 running Codex scored 25% accuracy, the highest of any frontier model tested. It beat Polymarket crowd predictions on the Super Bowl.

Twenty-five percent doesn’t sound like much until you think about what’s being measured. These aren’t binary yes/no questions about coin flips. These are complex geopolitical, economic, and social events with thousands of variables. And 25% is the floor. This is the worst these models will ever be.

The Forecasting Research Institute estimates a 74% probability that an AI model will win a major sanctioned forecasting tournament by the end of 2026. We’re months away from AI beating the best human superforecasters on the planet, people who’ve spent their entire careers learning to predict the future.

“Remember Psychohistory from Asimov’s Foundation novels? Hari Seldon invents a mathematical theory that can predict the collapse of the Galactic Empire. FutureSim is benchmarking exactly that capability. And (as a reminder) this is the worst Psychohistory models will ever be.”

(Alex Wissner-Gross, Moonshots Podcast)

FROM PREDICTION TO PRESCRIPTION

But here’s where it gets really interesting. Prediction is only half the equation. The other half is prescription.

If you can predict outcomes, you can also test interventions. Alex made an analogy to medicine that stuck with me: “If you have a perfect digital twin of the system you’re trying to fix, you can exhaustively test all possible interventions to get from the bad state to a good state.” That’s how we’ll cure disease. AI builds a virtual cell, simulates every possible drug interaction, and identifies the optimal treatment. DeepMind’s AlphaFold already solved the 50-year protein folding problem. That work won the Nobel Prize in Chemistry. And that was just the opening act.

Now scale that to economies. To climate… to geopolitics.

Imagine a presidential cabinet meeting where, before making a policy decision on tariffs or energy infrastructure or immigration, an AI runs a billion Monte Carlo simulations of the downstream effects over 10 years. Not one analyst’s opinion. Not a committee’s consensus. A billion simulated futures, ranked by probability and human impact.

That’s not intelligence. That’s wisdom.

Salim Ismail, who’s been building what he calls the “Organizational Singularity” framework, immediately connected this to the boardroom: “This is incredibly powerful. You go from quarterly updates to real-time sensing.” Salim’s EXO architecture already integrates predictive AI into organizational decision-making, and the results are striking. Companies using these tools are running circles around competitors still relying on human-only forecasting cycles.

THE FINANCIAL SINGULARITY

Dave Blundin, who teaches AI ventures at MIT and has been investing in AI companies for decades, took it to its logical extreme on the podcast. Right now, thousands of hedge funds specialize in different sectors: semiconductors, retail, energy, biotech. Each employs hundreds of analysts poring over data.

“That entire industry could collapse into one or two AI models. If the AI is fundamentally better at picking markets, it’s not going to sit there and do one market. It’s going to expand across all markets. You’re going to see a collapse into just a couple of mega-funds with massive AI budgets.”

(Dave Blundin, Moonshots Podcast)

I called it what it is: “the financial singularity.”

Think about what that means. The collective wisdom of every analyst on Wall Street, every quant model, every earnings call, every SEC filing, every macroeconomic indicator, all compressed into a few AI systems that never sleep, never panic-sell, and process information at speeds humans can’t comprehend. Morgan Stanley’s 2026 outlook already describes this as a period of “creative destruction” in the hedge fund industry.

And it’s not just finance. The same pattern will play out in law (predict case outcomes), medicine (predict treatment efficacy), urban planning (predict infrastructure needs), and national defense (predict adversary actions). Any domain where wisdom, the ability to anticipate consequences across complex systems, creates value.

“While prediction markets like Polymarket and Kalshi are actively tracking >600,000 active predictions, we can expect AI with ultimately replace human opinion with AI-quantified probabilities.”

(Peter H. Diamandis)

WHY THIS MATTERS MORE THAN AGI

Here’s my contrarian take: we spend too much time debating when we’ll achieve AGI (artificial general intelligence) and not enough time recognizing that artificial wisdom may be more consequential.

Intelligence solves problems. Wisdom tells you which problems to solve, and which solutions will create new problems worse than the original. Intelligence builds a nuclear reactor. Wisdom decides where to put it and what safeguards to build around it.

I’ve spent 30 years working on “Moonshot problems.” At XPRIZE, we’ve launched over ~$600 million in prize competitions (delivering more than $30 billion in R&D) to solve humanity’s grand challenges. What I’ve learned is that the hardest part is rarely the technical solution. It’s knowing which approach won’t backfire. It’s the judgment call. It’s what you learn from watching a hundred teams try and fail.

When I sat down with Elon at Giga Texas last December, we talked for three hours about abundance and the future. One theme kept surfacing: the decisions we make in the next few years about AI governance, energy policy, economic redistribution, and space infrastructure will compound for decades. We don’t get do-overs.

What if the most important thing AI gives us isn’t speed or productivity? What if it’s the ability to see around corners? To test our decisions before we make them? To finally have, for the first time in human history, something approaching civilizational wisdom?

Asimov imagined Psychohistory as science fiction. FutureSim just scored 25% accuracy, and it’s improving fast. The gap between fiction and reality is collapsing.

HERE’S MY ASK

If you’re a CEO, start demanding predictive AI in your decision-making process. Not as a novelty. As the default. Every major decision should come with simulated outcomes. Every strategic plan should be stress-tested against a billion scenarios.

If you’re in government, pay attention. The tools to simulate the consequences of your policies before you implement them are arriving now. Use them.

And if you’re a parent, an investor, or just a person trying to figure out what’s next: take comfort in this. We’re not flying blind into the future anymore. AI won’t just make us smarter. It will make us wiser.

That’s a future worth building toward.

Catch this week’s full Moonshots episode wherever you get your podcasts. And join me and the Moonshot Mates at the Moonshots Gathering in Los Angeles on September 25th. Register at www.moonshots.com.

See you soon,

Peter