In spite of being tempting and easy-to-understand, Thought Experiments have a high chance to enforce preconscious biases and assumptions of the researcher/student/listener. This "injection of bias" happens because, upon hearing a story, the listener pre-consciously fills in the blanks/the unknowns with hidden/mute assumptions, which do NOT then reach the "debate" from the conscious mind. As you may expect, this phenomenon especially affects universities from Western countries with biases about non-Western cultures.

1. The deconstruction of the
Chinese Room
Thought-experiments are Breeding Grounds for Bias, Prejudice, Preconception and
Discrimination
Bogdan BOCȘE
Managing Partner Knosis.ai

2. About Bogdan
● Chief Executive Geek at Knosis.ai and
Envisage.ai
○ AAI*: artificial and augmented intelligence
● 14+ years of experience in big data and
solution architecture, startup and corporate
● Founder of DeepVISS
○ Open-source ways to make machines talk to
each other about how humans learn
○ Applied Mathematics and Interdisciplinary
Research in Education and Psychology of
Intelligence
“To an open mind there are no closed doors.”

3. 1. The Chinese Room - an overview
2. Other thoughts experiments
a. The Turing Test
b. The Babel Library
3. Hidden fallacies of thoughts experiments:
a. The fallacy of immutability over formalism
b. The fallacy of the axiom of (forced) choice
c. The fallacy of assuming all language is formal and algorithmic
d. The unthrown exceptions and undeclared unknowns unknowns
4. Paths forwards & Tools for Researchers
a. Listeners as Compilers
b. Deconstruction is NOT the same as destruction
c. Avoid confusion between <Intuition> and <Knowledge>
d. The catastrophe of boundless abstractions: loss of significance
e. Similitude (tight, strict) vs. Simile (loose, permissive)
5. Our Work at Knosis.ai
Agenda

4. What is the Chinese Room?
● Thought experiment proposed by John Searle in his paper, "Minds, Brains,
and Programs", published in Behavioral and Brain Sciences in 1980
● Assumptions:
○ Assume there is a Computer that behave as if it understand Chinese, in writing, by interpreting
a program for parsing messages and composing replies in Chinese
○ Computer performs its task so convincingly that it comfortably passes the Turing test
● Proposal:
○ What if we were to replace the Computer that is “able” to interpret Chinese with a human
person (or a team of humans) that would execute precisely the same Program (instructions)
for parsing and generating messages in Chinese?
● Questions:
○ Could we then say that the Chinese room has the ability to speak Chinese?

6. The Chinese room argument holds that a digital
computer executing a program cannot have a
"mind", "understanding" or "consciousness",
regardless of how intelligently or human-like the
program may make the computer behave.
The argument is intended to refute a position
Searle calls strong AI: "The appropriately
programmed computer with the right inputs and
outputs would thereby have a mind in exactly
the same sense human beings have minds."
The conclusion of the thought-experiments are
however often corrupted and polluted by the
assumptions allowed or enforced by the
proposer of the thought experiment.
Such assumptions include:
● The machine cannot repair itself
● The machine cannot make copies of itself
● The machine cannot autonomously decide
it physical position (no autonomous
movement).
● The machine cannot make willful or
accidental changes to its body or to its
mind (code).

7. The Turing Test and the imitation game

8. The Turing Test is actually a two-blade scissors (for cutting the “powers” of conversation):
● Linguistic reductionism
○ The entrapment of linguistic formalism -> immutable rigidity of rules
○ No new words are formed or suggested
○ None of the unknowns can be described or inspected in further detail
○ There is no means to gauge para-verbal cues regarding reaction and affect.
○ The entire purpose of the language exchanged is to achieve persuasive imitation.
● Corporal reductionism
○ The entrapment of the interlocutors
○ Lack of material exchange (eg. objects)
○ Lack of shared stimuli (eg. isolated in different rooms)
○ The speaker (writer) and the listener (reader) are forced to communicate as if they have to awareness
of their bodies, of their surroundings and of each others’
The Turing Test and the imitation game

9. …force exact same form …
A Procrustean bed is an arbitrary standard
to which exact con-formity is forced.
In Edgar Allan Poe's influential crime story
"The Purloined Letter" (1844), the private
detective Dupin uses the metaphor of a
Procrustean bed to describe the Parisian
police's overly rigid method of looking for
clues.

10. Looking at subjects of the thought experiment as if they were inmates in a prison.
(or, at least, partially devoided of free will)

11. The Library of Babel
At present it contains all possible pages of 3200
characters, about 104677
books.
Any text you find in any location of the library will
be in the same place in perpetuity.
Try it yourself
Sometimes how one asks the question is more
important than finding the answer.
(1) Choose chamber
(3) Choose shelf
(2) Choose a wall
(4) Choose book & a page

12. The Hidden Fallacies

13. The fallacy of immutability over formalism
What it is:
● immutable, adjective. unchanging over
time or unable to be changed.
● The assumption that the objects in the
scene can only perform the actions and
transitions
● Examples:
○ It is assumed that the people working in
the Chinese room cannot leave and cannot
take breaks (for the duration of the
experiment)
○ It is assumed that none of the topics
discussed have to do with the respective
bodies of the interlocutors
What it contradicts:
● In practice (at large scale), actions and
transitions of an actor or an object are NOT
bound by the linguistic definition of the script.
The actor may choose his/her free will to exit
(escape) a role or the actor may, by mistake,
perform a deviation of the original script.
● (At small scale) Heisenberg’s principle of
uncertainty is enough to show that any body
of mass diverges informationally from the
image or from the definition that any observer
has fixed.

14. The fallacy of the axiom of (forced) choice
What it is:
● Informally put, the axiom of choice says
that given any collection of bins, each
containing at least one object, it is possible
to construct a set by arbitrarily choosing
one object from each bin, even if the
collection is infinite. Formally, it states that
for every indexed family of nonempty sets,
there exists an indexed set such that for
every . The axiom of choice was
formulated in 1904 by Ernst Zermelo in
order to formalize his proof of the
well-ordering theorem.
What it contradicts:
● Only bounded (finite) entropy can be
present in a finite volume (Bekenstein
bound).
● No system can, in the observable
universe, expend an infinite amount of
work to perform, at least once, the
prescribed Supertask (eg. picking from
infinitely many non-empty baskets).

15. The fallacy of assuming all* language is formal and algorithmic
What it is:
● Saying that something* is formal is
equivalent with saying that “the vessel
which holds the rules which govern the
existence of said something* does not
change its forms/shapes/topologies”.
● A formalism has at least a core set of rules
(a kernel) which is deemed immutable
(and often impenetrable).
● Such assumptions come at the expense of
the freedoms (degrees of freedom,
entropy, free energy) of the actors involved
in using the prescribed language in a
given script.
What it contradicts:
● The ability of language to form and to adapt.
○ Forming new words for newly-observed
phenomena or newly-defined intentions
○ Forming new language by mixing, especially
during speech, one’s native language with words
from an international language.
● The ability of the interlocutors to negotiate and to
agree-upon changing rules of the conversation or
the form of the conversation.
○ Describing new phenomena using new signals or
symbols which are inspired by local sensing rather
than previous formal convention.
○ Find a name for a new, unnamed fruit after first
tasting it
● The choice to _wait_ for more information to
become available is often available to most
agents in real-life situations.

16. The unthrown exceptions and undeclared unknowns unknowns
What it is:
● ”<Unthrown> exceptions” may refer to
situations where the (en)actor of the formalism
(the person or machine from a thought
experiment) would observe _in-situ anomalies_
(considerable differences between what is
expected and what is observed) and would be
unwilling or unable to report such anomaly to the
architect of the formalism.
● Concealing evidence that contradicts expectations
only to avoid the complexities and perils involved
scrutiny of the core/ground rules (axioms, articles
of the constitution, the N commendments).
What it is:
● “Undeclared unknown unknowns” are a
form of epistemic misrepresentation, which
can be either willful (done to mislead) or
neglijent (done in alleged good faith).
● Example:
○ Pretending to be sure of something you
are not sure about.
○ Falsely pretending to hold knowledge that
you know the other cannot verify.

17. Intuition as a magic mirror.
● Thought experiments rely on intuition
and imagination in order to
mentally-enact the realization of the
proposed narrative under the proposed
assumptions (i.e. anonymous axioms).
● Intuition is the opening (i.e. availability
to expend time and mental resources)
of the reader and writer to also allow
the usage of partially-, weakly-, or
loosely-defined relationships of
similitude->simile->similarity in the
furling and unfurling of the narrative.

18. Paths forwards & Tools for Researchers

19. Listeners as Compilers
● In computing, tombstone diagrams (or
T-diagrams) consist of a set of “puzzle
pieces” representing compilers and other
related language processing programs.
They are used to illustrate and reason
about transformations from a source
language (left of T) to a target language
(right of T) realised in an implementation
language (bottom of T).
● Each listener can be thought of executing
a compiler (corresponding to known
languages, including body language) that
can output thoughts, questions, answers,
as well as descriptions for new compilers.

20. Deconstruction is NOT the same as destruction
Deconstruction ≠ Destruction
Deconstruction ≡ Disassembly
(Taking something apart to see what it is made of and how/why it works the way it does)
(Just like debugging)
(Or like reverse engineering)
( Deconstruction is the act of inventorying hidden assumptiuons)

21. Avoid confusion between <Intuition> and <Knowledge>, while allowing both types of
statements..
● Avoid confusion between <Intuition> and
<Knowledge>, while allowing both types of
statements..
● The formalism of reducing the result of evaluating
ALL statements to a bipolar, flat, uni-linear, True and
False, is incongruent with the levels of uncertainty
most observers experience for most observations in
our Known Universe.
● Recommended alternative: Subjective Logic

22. The catastrophe of boundless abstractions.
● Collapse of the virtual-to-real entropic
representation ratio: too few bits on the “map”
(symbol) representing too many possible states of
the underlying area (the signified).
● Categorical collapse of assumed dependencies
and entropic under-representation.
● See also:
○ Loss of Significance
○ Catastrophic Cancelation
(especially as applied to weights of large neural models)

23. Similitude (tight, strict) vs. Simile (loose, permissive)
A model is said to have similitude with the real
application if the two share geometric similarity,
kinematic similarity and dynamic similarity.
A simile is a figure of speech involving the
comparison of one thing with another thing of a
different kind, used to make a description more
emphatic or vivid (e.g. as brave as a lion).

24. Our Work at Knosis.ai
● Data augmentation
○ Descriptive
■ Labeling
■ Tagging
○ Adversarial
■ Fake Detection
■ Fake Injection
○ Elicitation
■ Imaginative projection
■ Conotational projection
● Hybrid, Full-Duplex Learning between Human and Machine Agents
○ H2M2H Learning
○ Recursive auto-poiesis and reflection
● The Immersion of Directed Multi-graphs in Embedding Fields. (Paper)

25. bogdan@knosis.ai
https://twitter.com/bocse
https://www.linkedin.com/in/bogdanbocse/