{
  "kind": "route",
  "format": "kosplora-route/0.2",
  "question": "What could brain science offer to making AI safe?",
  "move": [
    "understand",
    "decide"
  ],
  "audience": "Some reading in neuroscience or psychology helps. No laboratory or programming needed.",
  "time": {
    "solo": "3h35m",
    "together": "45m"
  },
  "authors": [
    {
      "name": "Bentley Moon",
      "date": "2026-10-21"
    }
  ],
  "would_change": "If a published result showed one of the roadmap's paths making a model measurably safer, station 2 would start from that result instead of the roadmap's proposals.",
  "standard": null,
  "stations": [
    {
      "type": "orient",
      "time": "30m",
      "task": "Read the roadmap's summary and introduction. Write down the paths it proposes, such as copying the brain's representations, fine-tuning models on brain data, and using neuroscience methods for interpretability, and why its authors think people are a useful model for safety.",
      "sources": [
        {
          "title": "NeuroAI for AI Safety",
          "author": "Patrick Mineault, Niccolò Zanichelli, Joanne Zichen Peng, Anton Arkhipov, Eli Bingham and others",
          "date": "2024-11-27",
          "url": "https://arxiv.org/abs/2411.18526",
          "origin": "published"
        }
      ]
    },
    {
      "type": "source",
      "time": "1h",
      "task": "Choose one path and read its chapter in the roadmap. Note the claim, the evidence offered for it, and what the authors say could go wrong, including their own warning that some of this work could increase risk rather than reduce it."
    },
    {
      "type": "counterpoint",
      "time": "40m",
      "task": "Read a study that applied standard neuroscience analyses to a simple microprocessor whose workings are fully known, and found they did not reveal how it processes information. Write down what that result says about the path you chose.",
      "sources": [
        {
          "title": "Could a Neuroscientist Understand a Microprocessor?",
          "author": "Eric Jonas and Konrad Paul Kording",
          "date": "2017-01-12",
          "url": "https://doi.org/10.1371/journal.pcbi.1005268",
          "origin": "published"
        }
      ]
    },
    {
      "type": "make",
      "time": "1h",
      "task": "Write a one-page note turning your path into one claim that could be tested: the experiment, the result that would show it makes a model safer, and the result that would show it does not. Add the counterpoint and how the experiment would answer it."
    },
    {
      "type": "person",
      "time": "15m",
      "task": "Give the note to someone who studies brains or behaviour and ask where the experiment would fail. Write down their objection beside your claim."
    },
    {
      "type": "reflect",
      "time": "10m",
      "task": "Record whether you still think the path is worth pursuing, and why. If you want to go further, the fellowship below takes researchers from neuroscience, biology and other sciences into AI safety research.",
      "sources": [
        {
          "title": "PIBBSS Fellowship",
          "author": "Principles of Intelligence",
          "date": "",
          "url": "https://www.princint.ai/programs/fellowship/",
          "origin": "published"
        }
      ]
    }
  ],
  "artifact": {
    "keeps": "A one-page note turning one neuroscience path to AI safety into a testable claim, with the experiment, the result that would count against it, a published counterpoint and a scientist's objection."
  },
  "exit": [
    "keep",
    "share",
    "fork",
    "correct-a-source",
    "close"
  ],
  "revisions": [
    {
      "date": "2026-10-21",
      "change": "First version.",
      "by": "Bentley Moon"
    }
  ]
}