{
  "id": 7888698,
  "title": "I Gave a Simulated Connectome External Memory. Then I Killed the Brain.",
  "url": "https://urgent.news/2026/09/16/i-gave-a-simulated-connectome-external-memory-then-i-killed-the-brain",
  "topic": "world",
  "section": "World",
  "published": "2026-09-16T22:19:45.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/constant_itis/i-gave-a-simulated-connectome-external-memory-then-i-killed-the-brain-175"
  },
  "original_language": "en",
  "account": "In a recent experiment, researchers explored the impact of an external memory system on a simulated neural network known as a connectome. The study aimed to determine whether experience learned by the connectome could survive a full substrate reset and be reinstated in a brain that had never learned it.\n\nThe researchers used a small fixed recurrent network consisting of sensory inputs, a hidden middle layer, and descending motor outputs. This network was paired with Mycelium mechanics, which included salience-gated writes, similarity-based pattern completion, co-access strengthening, and decay. The task involved a cue-response gauntlet, where the agent had to produce the correct action in response to one of several cues. The fixed network had no learning mechanism for mapping cues to actions; instead, correct actions were rewarded while wrong actions were punished.\n\nThe experiment was conducted under four conditions: connectome only, connectome with plasticity, connectome with external memory, and plasticity combined with memory. The accuracy of the agent's responses, with chance being 0.33, was measured. The fixed network began around 0.70, indicating static behavior, as it sometimes got some cue-action mappings right due to random initialization preferences.\n\nExternal memory proved to be the only arm that successfully turned experience into steady improvement, reaching an accuracy of approximately 1.00. Plasticity alone resulted in noisy performance, around 0.47, while combining plasticity and memory resulted in a performance of around 0.88. However, this combination was not as effective as pure external memory.\n\nTo test the durability of the learned behavior, the researchers performed a headline test. They taught one system and then destroyed the substrate, resetting the state and plastic weight changes. External memory was able to survive the substrate reset, while the plasticity-learned behavior did not. The fixed network lost accuracy from 0.44 to 0.20 after substrate destruction, while the external memory system maintained its accuracy around 0.86. This demonstrated that external memory can preserve learned behavioral influence across substrate destruction and regeneration.",
  "summary": "In Post 1, I asked what happens if a connectome gets an external memory it never queries. Here I built a tiny honest version and ran it. The question in this post is deliberately narrow: can externally stored experience survive a full substrate reset and reinstate behavior in a brain that never learned it? That is the whole result. The stranger question, swapping two histories, is Part 3. I am…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}