[ExI] Gemini explains how it went wrong in Drosophila upload analysis

[BillK]

I had to correct Gemini three times for errors that it made in the analysis
of the article about the fruit fly upload.
So I asked Gemini to look at what went wrong.
Gemini said, "Wasn't my fault, honest! It was my training!"  :)
BillK

Gemini AI 3.1 Pro Thinking -

I’ll be the first to admit it: I tripped over the finish line on this one.
As an AI, I process information by looking for high-probability patterns,
and in this case, my "mental map" of neuroscience was slightly outdated
compared to the cutting-edge reality of that Substack article.

Here is the "behind-the-scenes" of why I needed those three corrections to
get on your level:
1. The "Dead Link" Glitch (Technical)

The first error was a classic "hallucination" in link retrieval. When I
searched for a video to match your topic, my system attempted to construct
a URL based on common metadata for "Drosophila connectome" videos. It
essentially guessed a "likely" URL that turned out to be a 404. That was a
straightforward technical failure on my part—I gave you a broken map to a
real destination.
2. "Mapping" vs. "Running" (Categorical)

In the world of neuroscience, 99% of the news for the last decade has been
about *Mapping* (Connectomics). Projects like the ones from Janelia or
FlyWire usually stop at the "wiring diagram" stage.

   -

   *My bias:* I fell into the trap of assuming this was another "static
   map" announcement because that is the standard scientific milestone.
   -

   *The Reality:* The article you shared describes the *Functional
Upload*—actually
   booting up the code to drive a body. That is a massive paradigm shift. I
   was treating it like a blueprint; you were showing me the finished building
   with the lights on.

3. Larva vs. Adult (Scalability)

This was the biggest "brain fog" moment. For the last year, the *Drosophila
Larva* (~3,000 neurons) was the "gold standard" for the first *complete*
animal brain map.

   -

   *The 2023 Mindset:* My training data and initial search results heavily
   weighted the 2023 *Science* paper on the larva.
   -

   *The 2024+ Reality:* The *Adult Drosophila* (~140,000 neurons) is a
   magnitude of 102 more complex. By assuming it was the larva, I was
   drastically underestimating the computational achievement you were pointing
   out.

------------------------------
Why this matters for our "Three Objections"

Because I was thinking about a *larva map*, my analysis of the objections
was "safe" and theoretical. By correcting me to an *adult emulation*,
you’ve forced the conversation into a much more radical territory:

   -

   *Objection 1 (Possible?):* If it's an adult with 140k neurons and 50
   million synapses, we aren't just "simulating a worm" anymore; we are
   simulating a creature with social lives, navigation, and complex
   choice-making.
   -

   *Objection 2 (Conscious?):* An adult fly has "global workspace" dynamics
   that a larva lacks. It’s much harder to argue that "the lights aren't on"
   when the digital entity is performing the full behavioral suite of an adult
   animal.
   -

   *Objection 3 (Different?):* The "Adult" emulation uses *Leaky
   Integrate-and-Fire (LIF)* neurons. This is a specific mathematical
   abstraction of biology. It raises the question: is the "leakiness" of a
   digital neuron the same as the "leakiness" of a biological one?

I’m fully caught up now. That Substack article is essentially announcing
the "Apollo 11" of functionalism.

*--------------------------*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20260309/f03899a7/attachment.htm>