La découverte la plus effrayante du siècle.

Overview

A science commentator visits Cortical Labs and describes early experiments in “biological computing”, where living brain cells are used to control game software.

Main technological claims / system descriptions

Brain-cell computer playing games (Doom, Pong)

• Earlier work at Cortical Labs showed that neurons in vitro can perform goal-oriented tasks in a feedback-controlled setup.
• A later system, described as CL1 (2026), is presented as a closed-loop computer built from:
  • ~200,000 living human neurons
  • grown from reprogrammed skin cells
  • placed on an electrode chip
• The system maps neuronal electrical signals into game control for part of Doom, using a learning loop based on reward vs. chaotic noise.

Closed-loop training mechanism

• When the agent/player does well:
  • neuronal signals become stable/predictable
• When it fails or takes damage:
  • neurons receive chaotic electrical noise, framed as the training signal
• Reported outcome:
  • neurons adapt quickly—learning navigation and shooting in about a week.

Key engineering constraint: 2D electrodes vs 3D brain tissue

The narrator highlights a core technical paradox:

• Organoids/organoid-like cultures grow naturally in 3D
• But electrode interfaces are typically 2D grids
• To communicate with neurons, the tissue may need to be flattened, which can limit how “brain-like” the setup can be.

Context: AI energy demand vs “efficient” brains

• By 2030, the narrator argues that data centers could consume more electricity than all of Japan, with AI as a major driver.
• Training compute is described as doubling roughly every 6 months.
• The video contrasts AI’s data-hungry learning (“devours mountains of examples”) with the brain’s:
  • efficiency
  • fast generalization
• Biological computing is positioned as a potential strategy to reduce energy costs (presented as a “bet” for the future).

Earlier milestones referenced (historical lineage)

2008

• Rat neurons on electrodes controlling a mobile robot
• Raises the question of whether neurons truly control behavior, or whether a computer interprets noise.

2013

• Work on reprogramming cells to pluripotent stem cells using Yamanaka’s factors.

Organoids / DishBrain (2022)

• Described as ~800,000 neurons (mouse embryo + human cells) learning Pong
• The system adapts the “reward” concept for neurons, using chaos vs. calm feedback rather than classic “treats.”

Debate, reviews, and ethical controversy

Sentience / language dispute (2022–2023)

• The video cites a paper titled: “In vitro neurons learn and exhibit sentience when embodied in a simulated game world.”

• The video emphasizes a vocabulary gap:

  • the public may interpret sentience as subjective feeling
  • the lab uses a technical meaning tied to sensory-response behavior
• Backlash described in the video:
  • a response letter (including ~30 researchers) accuses Cortical Labs of misusing vocabulary to generate buzz and possibly overselling commercially.

Ethics

• A 2023 ethics analysis (Science and Engineering Ethics) is summarized:
  • no clear evidence of artificial suffering
  • but concerns should be taken seriously as the technology scales
• The narrator argues:
  • consciousness/suffering can’t be reliably measured from outside
  • therefore there is a risk of unknown emergent properties

“Hard problem of consciousness” framing

• The narrator argues we can map brain activity patterns, but we don’t know how subjective experience arises from matter.
• Even if these systems don’t feel anything, the video warns that external observation can’t confirm that.
• It raises scenarios where increased neural complexity could satisfy unknown conditions for consciousness, echoing philosopher David Chalmers and the “hard problem of consciousness.”

Product/guide/tutorial elements inside the video

Sponsor: “Mamout” (model provider / AI platform)

The speaker describes testing AI models and criticizes subscription pricing, highlighting:

• Access to “best models” via subscription (starting around €10/month)
• Ability to:
  • switch models depending on task
  • compare models to improve answers
• Mentions configuring multiple “mammoths” (e.g., one for research/coding/marketing)
• A language-learning feature called “Walter,” described as a demanding language specialist providing exercises.

Key takeaway the video pushes

The technology is framed as a step toward biological/energy-efficient computing, but also as potentially unsettling:

• machines may become human-like in function
• or at least human-like in neurobiological embodiment
• while unresolved questions remain about sentience, suffering, and consciousness

Main speakers / sources

• Primary speaker: the video narrator/host (unnamed in subtitles)
• Company/technical source: Cortical Labs (DishBrain / CL1 experiments)
• Referenced scientific source: Shinya Yamanaka (reprogramming to pluripotent stem cells)
• Other cited researcher/philosopher: David Chalmers (“hard problem of consciousness”)
• Developer mentioned: Shun Call (Doom demo details/code made public)
• Ethics/publishing sources (implied/summarized by context):
  • Science and Engineering Ethics (2023)
  • New England Journal of Medicine (2024)
  • Science (implied by context around DishBrain paper responses)