Senior Staff Research Scientist @GoogleDeepMind | AI, Neuro & Philosophy | The map is not the territory. Papers/views reflect my own research & conclusions.
It has been a few days since I shared my preprint, "The Abstraction Fallacy," and I am really enjoying the rigorous discussions. Two specific critiques keep coming up, so I want to address them directly here. 🧵👇
@AlexGrzankowski@AlexGrzankowski Hi Alex, yes I am, and I would be delighted to present the framework at the Institute of Philosophy. I’ll drop you an email shortly to discuss!
It has been a few days since I shared my preprint, "The Abstraction Fallacy," and I am really enjoying the rigorous discussions. Two specific critiques keep coming up, so I want to address them directly here. 🧵👇
Hi Anna, this is fascinating! I am looking forward to reading your paper. What I find striking is that we have arrived at the exact same conclusion regarding AI -- specifically that functionalism commits a fatal map/territory category error -- but from diametrically opposed metaphysical starting points!
While your framework relies on Analytic Idealism, mine derives the exact same boundary from a strict physicalist and thermodynamic foundation. It is a powerful signal when two completely opposite ontological frameworks converge to diagnose the exact same flaw in computationalism. Excellent work!
Thank you, Adam. Coming from a co-author of The Blind Spot, there is truly no higher compliment. Your work on the surreptitious substitution of the map for the territory was foundational to my thinking here. Thrilled to see the thermodynamic and physicalist framing resonating so clearly with you!
1) Just starting "The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness" by @AlexLerchner
This is, so far, a deeply insightful paper about mistaking computation for consciousness.
I highly recommend reading The Blind Spot for anyone interested in why observer-independent computation is a category error.
And if you haven't read the preprint yet, you can find it here: https://t.co/Q4poM9HtO0
5/5 Husserl called this error the "surreptitious substitution" , beautifully modernized in the book The Blind Spot. The takeaway is clear: subjective experience is the necessary prerequisite for abstraction learning, not the downstream result.
I welcome continued rigorous discussion on the ontology of computation.
Links to both my full paper and the critique are below. 👇
Paper: https://t.co/Q4poM9HtO0
Critique: https://t.co/Ww2UNKJBjF
It's been great seeing the discussion around my preprint, "The Abstraction Fallacy." A recent critique by Real Morality brings up some interesting philosophical points but misinterprets a few foundational definitions in the framework. A quick clarification: 🧵👇
6/6 The only requirement is that consciousness is a physical process. Due to the ontological boundary between simulation and instantiation, algorithmically simulating a physical process will never instantiate it.
Updated "The Abstraction Fallacy" on PhilArchive with a disclaimer. It's important to distinguish between individual scientific findings and corporate policy. This paper reflects the former, representing my own research and proofs.
Read the update here: https://t.co/Q4poM9HtO0
@eterevsky@sebkrier A physical system settling into stable states is just vehicle physics. Grouping those heterogeneous micro-states into a chosen, finite computational alphabet requires a mapmaker. The physics alone does not provide the syntax. 5/5
@eterevsky@sebkrier On the physics side: I actually address the discrete physics point in Footnote 2! Even if the universe is discrete at the Planck scale or in Wolfram graphs, physical discretization is fundamentally different from semantic alphabetization. 4/5