Appendix A Appendix. A potential answer to CRA
Somewhat, in a way or another, a solution can be reached at least in terms of interpreting intelligence in a varied different way. Of such, there exists my blueprint notes of before, which outlined particularly interesting aspects and somewhat naive representation and ideas on a functional construct similar of those established above. While it is not perfect, and the line of logic leaves much to be desired, the idea is nevertheless worth of consideration.
A.1 The merits of the processing unit
In fundamental neuroscience, one of the main question, that was answered figuratively and conceptually, is the question about the merit of existence of a separated organ that eventually, will exhibit controlling behaviours and might up to the point of intelligent behaviours. That is, it raises that
Protists and the simplest multicellular animals (sponges) display ingestive, defensive, reproductive, and other behaviours without any nervous system whatsoever, raising the question: what is the adaptive value of adding a nervous system to an organism?
This point is, as it might sound, dilligently enough is very strong. We study intelligence as the essential part to search and create generally, a learnable system, adaptable system, and rational system, but to which stage, which layers and which form would this be classified into? Are our systems generally similar to the Proties and multicellular animals, or have advanced to a greater abstraction level?
A.2 The layer dilemma - or Chinese Room Argument
We pertain to the layer dilemma, or as generally recalled, the Chinese room argument [Searle, 1980]. If we choose to simplify partially our system to an IO conduct, then this argument is the first to be addressed. However, this will be done in a very problematic way.
Assumption A.1.****
Of the layer dilemma, for the set $\{L_{i}\}_{i\leq n}$ of the bottom-up categorization and measure up to $n$ layers, then $L_{n-1}$ do not ”understand” what happens in $L_{n}$, but $L_{n}$ is of itself, with sufficient realization and connection. This means the hierarchical information pathway is one-way, bottom up. [^13]
However, if is so, then there exists the fact that we can define a condition $Q$ on such that, for a measurable complexity and ‘model capability’ scale $A[p]$, then if $A[p]\geq Q$, the model escaped the potential restriction, and become recursive in nature. In principle, this such that the man in the box recursively think outside what is conceived to be in his control. For every downstream, there exists an encoding to the lower dimension. Such downstream inevitably will result in an exposure to the data, in some form or another, that can be decoded. If, the model capability for the outermost layer, $L_{n}$, satisfies the inequality, then it can, indeed, escape, and interfere directly to a certain extent of the actual ‘world’ layer $L_{0}$ it receives downstream to, which is, where we feed it the data and else, isolated from the model. This is, in principle again, totally a hypothesis, but then, we need to have a boundary to restrict this.
Conjecture A.1.****
Suppose there exists a measure $\mathcal{M}[L_{i}\in A_{n}]\in\mathbb{R}^{p}$, usually for $p=1$ as scalar for any model configuration $A_{n}$. Then, if $\mathcal{M}[L_{i}]\geq Q$ for $Q\to\mathrm{dim}(p)$, then there exists a recursive pattern $\mathrm{Re}(A)$ such that it enables the model’s ability to ’learn’ of the upper $L_{i-1}$ layer.
Figure 3: A loose illustration of the layering principle. The lower layers do not know what is on the upper part. However, they can receive potential downstream input features, and hence work accordingly. This is the basis of abstraction by layers, thus stating that the man in the box do not know, and do not understand, anything but his circumstances and his current capability. Nevertheless, it still fulfils its role in the lower level hence forth, and its operation only in such range of the layer.
Hence, the answer in such form is reached, such so to address the fact that the operating machine receives downscale information within its own embedding, hence is typically restricted to that embedding layer alone. Furthermore, it also usually satisfies, since we categorize and generalise the layering, as according to the generalisability and abstraction value of each. Therefore, conspicuously, most of the time, the layer will hold, but under configuration, then it might jumps off the layer without any re-categorization method.
The layering theory also is an answer to the Chinese room argument, in the sense that there exists levels of which the abiding room is taking into account. Hence, for a person living in a scenario of input-output interaction, it is well-conceived that the ‘I’, within such capability, within such restriction, and being categorized as ’lower’, do not possess the ability to look outside the room itself. A possible hypothesis, would be in analytical form, there exists a separated evaluating space, where such layering principle is demonstrated by dimensional increases, and their functional downstream compression function.
Hypothesis A.1.****
Given the set $\{L_{i}\}_{i\leq n}$, there exists a dimensional analytic encoding $\mathsf{Comp}^{m}$, such that there also exists certain functional downstream compression function $F_{DC}$ that maps $\{L_{i}\}\times\mathsf{Comp}^{m}\to\{L_{i}\}\times\mathsf{Comp}^{m-1}$, essentially rescaling the dimensionality.
Of course, this is all conceptual. We still need to define the measure $\mathcal{M}$, the supposed inequality, the supposed construction, and furthermore, the complete test implementation that can be perhaps focused on and verify the principle. Nevertheless, this might prove useful, or at least in a direction of analysis that will either falsify entirely this approach, or partially, thus makes use of this to formalize new framework on understanding such problem[^14]. Nonetheless, such notion makes sense, for Searle-in-the-box to only be ‘aware’, of the arbitrary sense of the word ‘aware’ attributed of the analogous human awareness of what of his own domain, but not of any external or higher concept. It is apparent of human, too, as we can only semi-imagine the higher-level being, in philosophical inquiry, but nothing else — however when it comes to lower subjects, then it is surprisingly rich of certain interpretations. Such inquiry will remain valid, as for all of such layering scheme we can create. Such answer leads to two development. We can identify the layer of human, and try to build up layer-by-layer up there. Or, we might want to believe, there exists this layer, $L_{\epsilon}$, of which such restriction of top-down interpretability becomes redundant, in which the subject leaps out of its layer to the above. What of the two is more probable? We do not know.
In one way or another, CRA infers with the option of an input-output procedure. In such case, CRA clearly tells us that a simple, mundane notion of an input-output machine which ’does its tasks’ would not be sufficient of receiving the clarification and qualification for being intelligent. Which, by all means and purposes, are true. The construct if given in such circumstances, of the thought experiments, represents, if we took out the part where the argument said Searl is supposed to be everything a computer can be, then it’s true that such constructs are false in its claim that it can ’understand’. In fact, relatively simple, a given input-output mechanism, from what was observed from the outside, do not exhibit anything, and do not have the ability to even think, regardless that is valid of such. If we are to stand by Descartes arguments, then it is even more of the truth - the system in which the thought experiment was conducted provides it with no capabilities of any such.
Then what would be of the Chinese Room Argument that is worth it to dissect? Well, firstly, the claims of, at least in the acute interpretation - the man in the box is supposed to be everything a computer can be is false. IF we are to stand by our construction of facilities, then we are innately arguing about such facilities, and not the processes, and the underlying operations itself. Rather, we are complaining that the machine is not capable enough, which is true. But we also, to a given reference point, pointing to the existential facilities instead of the arbitrary, yet reasonable operational facilities instead for the comparison. And in fact, if we think of it in the layer construction, it makes more sense - each layer is classified given its arbitrary for now, an interpretation thereof. Such interpretation is contained for such layer, and hence cannot be thoroughly or at least in a glance, interpreted by lower-layer components. This construct offers a one-way restriction on the property of interpretability. In such case, the System Reply is partially right - the man inside can not be, by all means and purposes, understand what does it mean by even ’English’, or ’Chinese’. And even if such English ’understanding’ is embedded in the reasonable interpretable space of Searl in the room, then Chinese would appear to be entirely unknown, unless there exists a helper tool to resolve the situation of undefined operation. Indirectly, this prevents a Descartes situation from happens - an undefined situation with particular way to resolve.
But more than that, what constitute the notion of understanding? Taken only from the setting of the thought experiment, we cannot do but deceptively assume that understanding a language is similar to giving it certain ruleset to transfer from this word to others word only. By that, converting from base-2 to base-10 works the same - you know how to do it, yet there are things that constitute the philosophical, higher-level notion of ’understanding’ in such that allows you to actually understand the conversion - otherwise, the conversion is a blind matching. In fact, given the setting, would a conversion rulebook exists for such language? Language is, by itself, a very high-level concept. Translating from text to texts requires it not only to provide the definition matching, which could be reasonably identified by such notion like the rulebook mention, but the interpretation of the string is dictated by the logic of the language - the context in which it appears, the logical conformation that it contributes, the grammatical structure that makes sense of what is said and what is transferred, and else. Language itself, is a medium of information exchange, by one of its definition. A conversion does not constitute an understanding. And if the argument going back and forth is that Searl can somehow figure it out the patterns mean something, then it violates what I called the principle of externality - the ’lower components’ up to a given point in which the law of recursive immergence does (not) apply, cannot implement its higher constructions. Then, the Chinese Room Argument can be interpreted, in somewhat meagre form, the argument against telling the current conformation as able to understand, while it is not.