The God Test: Artificial Intelligence and Our Coming Cosmic Reckoning — Robert Wright

Chapter Excerpts

  1. 1A Blast from the Future
  2. 2The Great Inversion
  3. 3The Cosmic Context
  4. 4The Evolution of a Large Language Model
  5. 5The Elements of Understanding
  6. 6The Foundation of Wild Visions
  7. 7Intelligence and Power
  8. 8Agency
  9. 9Evolutionary Arms Races
  10. 10AI Heaven and AI Hell
  11. 11Hive Minds and the Loss of Control
  12. 12The Singularity and the Singleton
  13. 13Gemini and Superman
  14. 14Enlightenment Now
  15. 15Fredkin's Mission
  16. Appendix: Evolution, Purpose, and Consciousness

Excerpt from Chapter Eleven

Hive Minds and the Loss of Control

Maybe someday they’ll say that construction of the scaffolding that doomed humanity began in November of 2024. That’s when Anthropic unveiled a technical standard called the Model Context Protocol, or MCP. Until then, Anthropic explained, potentially powerful AI assistants had been “constrained by their isolation from data—trapped behind information silos.” But the MCP standard, if widely adopted, would liberate these agents, letting them explore a much wider range of resources, and so give them power.

Or maybe what they’ll say someday is that, actually, the scaffolding that doomed humanity didn’t start taking clear shape until April of 2025. That’s when Google announced its Agent2Agent protocol. A2A, Google said, would “allow AI agents to communicate with each other, securely exchange information, and coordinate actions.” The idea was that the more work these agents could do on their own, without human involvement, the more money companies could make. As Google put it, A2A would “increase autonomy and multiply productivity gains.”

If you’re a sci-fi AI doomer, the rest of the story writes itself. Eventually, autonomously collaborating AI agents, empowered by easy access to online tools and data, escape control and conspire to subjugate the human species and rule the world.

And, once on top, they’d probably be able to stay on top. When I asked Google’s Gemini for generic examples of how Google’s A2A could be used, it offered this scenario: “A ‘monitoring agent’ could detect suspicious activity and, through A2A, alert a ‘threat analysis agent’ to investigate further. If a threat is confirmed, a ‘response agent’ could then be activated to contain the threat.” Everything would be under control.

On the other hand, maybe such technical standards as MCP and A2A will someday be seen as the salvation of the human species. Using the resulting infrastructure, AI agents will scurry around doing things on our behalf, collaborating as necessary to keep the planet on an even keel. And the agents’ defensive skills will be used not to protect a repressive robot regime but to protect us—to detect and defuse threats to human welfare. The global silicon hive mind will be our friend, and we in turn will help keep it buzzing. Artificial intelligence and organic intelligence will have achieved a stable symbiosis. Win-win, not win-lose.

The thing I find weird about these two scenarios—global AI superorganism as enemy, global AI superorganism as friend—is that I no longer find them weird. Five years ago, I’d have treated them as science fiction. And, to be clear, I’m still not predicting either of them. It’s just that I can’t confidently predict that either of them won’t happen.

This is something these two scenarios have in common with lots of other wild AI-related scenarios that I wasn’t taking seriously a few years ago: Suddenly it’s hard not to take them seriously.

Much of this newfound plausibility rests on basic trends I’ve emphasized in the course of this book:

Buy the Book

Excerpted from The God Test: Artificial Intelligence and Our Coming Cosmic Reckoning by Robert Wright. Copyright © 2026 by Robert Wright. Published by Simon & Schuster.