Sound Research WIKINDX

"We can use ChatGPT to translate a text or to inquire how a sentence could be differently expressed. However, this is only an external purposiveness since it is driven not by the machine itself but rather by something outside of the condition of existence of the machine. Internal purposiveness is something else; it means the autonomy of the being in question. We don’t know the purposiveness of the existence of this or that monkey; it is, for sure, not for our pleasure or for any medical experiment; however, in the monkey, there is a purposiveness of survival that constantly resists the rapid dissipation of energy, since the latter means moving towards death. We could also talk about another purposiveness, which is the purposiveness of nature itself, what Kant calls the end of nature; we don’t really know what it is despite the fact that, according to the second law of thermodynamics, we are witnessing its disintegration. However, maybe only when a total disintegration arrives will we know its purposiveness. The natural end can only be thought of in the mode of ‘as if’ and not in the mode of truth, as we have seen above. Nevertheless, the question still remains: what is nature’s end, even if it could only be thought of in the mode of the ‘as if’?"

"Kant discusses Jacques de Vaucanson’s mechanical automata (of which there is the famous digesting duck as a representative) and demonstrates that it would be an illusion to conceive the freedom of such a puppet:

A human being would be a puppet, or a Vaucansonian automaton built and wound up by the supreme master of all artificial devices; and although self-consciousness would turn the automaton into a thinking one, yet the automaton’s consciousness of its spontaneity, if regarded as freedom, would be mere delusion, because this spontaneity deserves to be called freedom only comparatively."

"Given that machines exhibit a generative form of reasoning and are capable of ‘unsupervised learning’ or ‘auto-classification’, we ask: to what extent are they capable of being moral?"

"The Alignment Problem is conceptually weak and vague; indeed, it reveals something deep in culture, the haunting history of ancient slavery and the classical comparison between slave and machine. Even when we step back from the alarming concern of an extinction crisis to a more practical consideration of machine’s relation to human society, talking about an alignment to human values is not at all obvious. How do we determine something as being a universal human value or social norm? There are values specific to the Chinese people, the American people originated from different traditions and thus hold a divergent set of social norms, even though they are from the same continent, the Germans often dispute with the French regarding comparable values; indeed, so do the Mainland Chinese and the Hong Kong Chinese. Can we abstract from these differences a set of necessary human values, claim them to be universal and then force the AI to align them with these universal values? Or can one derive from censuses or surveys wishes shared by most of the people, and count them as the ‘general will’? It is unlikely that this is a question that concerns AI scientists since they might well be convinced that there already exists a set of values or norms which are beyond their own concern; and not to mention that these values, like semantics, might well change over time as they differ geographically due to certain historico-political events or natural disasters which might lead to an alteration in value. The Enlightenment, for example, which has often been taken as a movement that pursued the universal as its end, might be only an expression of the feeling and sensibility of a particular historical time in Europe, as Michel Foucault famously suggested."

"At this point, one might want to offer a pharmacological critique of AI by showing that technology is at the same time cure and poison, just as pointed out in the Phaedrus (274c–e). To recall, Socrates recounts a legend in which the Egyptian God Theuth offered to King Thamus the gift of writing which could aid his people to remember everything; Thamus replied that the opposite might be true, since with writing, one no longer needs to remember. Writing is hypomnesis (external reminder), which can simultaneously strengthen and weaken anamnesis. Technology is, thus, a Pharmakon, a curse and a cure; in other words, it is irreducibly good and bad at the same time. The central question of the Phaedrus is what defines good writing (philosophy) and what defines bad writing (sophistry) – the sophist Lysias’s written speech was challenged by Socrates due to its lack of definition of terms (such as eros) and its inability to point to what is just, noble and good. Who decides what is good and what is bad?"

"[...] we cannot avoid the question of value or even a genealogy of these values, as Nietzsche argued across several of his books. In the past, it was the priests and the philosophers who functioned as the physicians of culture; it was they who dictated to us our values. Now it is the technology industry and entrepreneurs such as Elon Musk and Sam Altman who dictate what the future will look like, and what kind of values we need to embrace for such a future. Echoing this industrial propaganda, an increasing number of philosophers begin their analyses by assuming that the singularity is imminent or has already arrived, in order to show that their intellectual foresights are far beyond most people. Worse still, how could we even position ourselves today when such physicians of culture are either silent behind the scenes or are no longer telling us the truth, when it has become the role of algorithms based on LLMs to tell us what is probably good and bad? In the foreseeable future, AI may well decide for us what is good by identifying the good with moral norms being a well-computed and calibrated result based on a large amount of observable data."

"If an algorithm based on LLMs could better define what is good for the human being, then why should we have to bother with AI Alignment or Super Alignment? Paradoxically, those enthusiasts who believe in superintelligence are also exactly those who distrust AI the most and, as such, are those who advocate for an advance fix to the supposed existential crisis inherent to AI alignment. If one really believed in superintelligence, that a superintelligent AI will one day supersede the sum of all human intelligence, then it follows that one should also believe in its capacity to supersede humans with regard to moral judgements; that they would be capable of superior social and economic planning for all the living beings on the earth. "

Kant’s transcendental philosophy, as we have seen, is an alternative to rationalism as well as to empiricism. Kant leaves the power of judgement to reason itself without it being guided or obliged by any external force or empirical knowledge. The Kantian machine, in this way, avoids the problem of the Humean machine regarding the vulnerability of causal laws derived from empirical experience. Kantian morality is, first of all, non-analytic and, second, non-empirical. It is not analytical since a general law is not given in advance and not demonstratable, one which, if present, could guide all human actions just as physical laws do to nature (we recall the rationalist teaching such as Wolff’s that morality could be proved through demonstration); it is also not empirical because a moral judgement should be independent of empirical experience, which cannot guarantee the validity of the universal. Kantian morality is a particular kind of algorithm, to which Kant gives the name the categorical imperative. For Kant, the highest good (summum bonum) means a world in which everyone behaves according to the categorical imperative, hence summum carries more the meaning of consummatum (completion) than supremum (supreme). We could imagine a scenario in which every individual is an algorithm capable of moral reflective judgement; each algorithm, or what Kant calls ‘each rational being in the world’, aims at the universal just as a moral agent aims at its own moral perfection. How can this be achievable? What might be the ‘technical procedure’ of the categorical imperative?

"Kant gives us a formal definition of beauty, namely ‘Beauty is the form of purposiveness in an object, so far as this is perceived in it apart from the representation of an end’. Beauty here doesn’t refer directly to the object, for example, its representation but rather the representation of the end; such an end is only conceivable through subjective reflective judgement. "

"The consciousness of the moral law, however, belongs to a different kind of fact; it is not an empirical fact, but rather a fact of pure reason. Since it is a fact of reason, this is not an error; however, humans will never be able to realize it in actu or in concreto. In this case, it is also an ideal of pure reason, one which we can only approximate, ‘as if’ there is an end."

"The work produced by AI carries no necessity but only probability – one could decompose Kandinsky’s abstract paintings, for example Composition (II), into various figural elements such as circles, triangles and retangles, and colors, then reassemble them according to different permutations by putting the figural elements at different places on the canvas with a nuanced color. No matter how similar they look like to Kandinsky’s paintings, this could hardly be called creativity. Artistic creativity, by contrast, is always haunted by contingency and must culminate in an affirmation of necessity – as in Beethoven’s String Quartet No. 16, whose fourth movement begins with the question ‘Muss es sein?’ and answers decisively: ‘Es muss sein!’"

Kantian Moral Test

"This test [The Turing test] is problematic since a moral act is an act in the world and not merely a language game, for a politician who can make a beautiful and passionate speech is not necessarily a good politician, and someone who can hold an excellent speech on morality is not necessarily someone who acts morally; therefore, instead of a Moral Turing Test based on imitation, we should consider the Kantian Moral Test as the only veritable test. All machines, without passing a Kantian Moral Test, would therefore be considered evil machines. The term ‘evil’ we use here refers to Kant’s understanding as a propensity. In human beings, as Kant tried to show in his late work Religion within the Boundaries of Mere Reason (1792), evil emerges not due to an empirical inclination but rather as a maxim of choice against freedom, which leads to incarceration by arbitrariness. This inability is by default present in humans as a species, as Horace wrote ‘vitiis nemo sine nascitur (no-one is born without fault)’ which Kant cites in the aforementioned work. However, it is precisely because of this default qua fault, that we could identify an epigenesis of reason in the Critique of Practical Reason.

Every moral philosophy – and philosophy in general – inevitably grapples with the question of the Good. There is no way to circumscribe the moral question since the Good has stood as the telos of philosophy since Plato; it is the destination of wisdom, for wisdom means the pursuit of virtue and happiness. If wisdom did not pursue the Good, then wisdom would fall prey to evil; therefore, it would not be wisdom by definition. Wisdom is not knowledge, for there might be no knowledge of wisdom as such. Socrates, for example, is wise precisely because he knows that he doesn’t know. The Good cannot come from the other, but only from itself as an autonomous act of reason. This is the most important and most perplexing statement of Kant’s moral teaching. It cannot come from the other because then it will be merely conditional, considering that moral values change over time; so when thought of in this way, there is no such thing as the Good since all moral values would be subject to history and culture. Historicism and relativism can easily dismantle the Good by showing that they are, in reality, merely contingent, as Nietzsche demonstrated in his works, especially On the Genealog y of Morality, the concept of ‘bad’ arose in opposition to aristocratic values, while ‘evil’ was a creative invention born out of the ressentiment of the weak and powerless. Indeed, all norms are contingent in the sense that they are historical; namely, they hold only in a specific historical context and in this historical context, those that meet the norms are considered moral."

"When machines are thought of as a mere means to an end, as is the case in capitalism, then we expect an end which is merely associated with the accumulation of capital, which has its essence in calculability. Division of labour and automatized production are two obvious examples of why machines have been used, because they increase efficiency and reduce costs. The moral-practical end is reduced to a utilitarian end, and machines are, again, treated as a means to an end. Utopias belong to such a moral-practical end, not because they are not realizable but because they negate the means to an end by treating every being as an end in itself. The negation of the mode of thinking based on the means to an end doesn’t mean that machines should be devoid of function or that all machines should become playful artworks, but rather, the use of the machine points to an end that is superior to the end it is assigned to. Considering the use of machines in factories, both Karl Marx and William Morris quoted John Stuart Mill’s saying, ‘It is questionable if all the mechanical inventions yet made have lightened the day’s toil of any human being’, to argue that machines, which were supposed to lighten the labour of human beings in fact only extend the length of working hours and intensify the labour process. One may argue that the true purpose of machines lies in the creation of true wealth, which Marx calls ‘disposable time’. [...] ‘Disposable time’ for Marx is neither a time that could be subordinated to the production of profit, nor is it limited to contemplation opposed to praxis as it was in ancient philosophy, instead it aims at the production of knowledge and the advancement of science, both of which may not have an immediate use of exchange value, but they will contribute to a more profound progress."

"By the same token, the exploitation of natural resources and the exhaustion of the planet is an ignorance of the purposiveness of nature since it reduces nature to a utilitarian end. An enterprise which excessively deforests to accumulate capital, while at the same time donating money to the poor affected by the deforestation to brand itself as an ethical enterprise, is within the calculation of means and ends, and it cannot whitewash itself from the guilt of hypocrisy. It can be true that the enterprise is happy about earning more money (as well as the benefits it obtains through its donation), and the poor are happy about receiving an unexpected donation; this doesn’t exclude it from being unethical. This is why utilitarianism is not an answer to ethics, and Kant’s morality helps us to understand the complexity of it."

"Human beings, driven by ends other than laws of physics, are like Maxwell’s demons, through their prosthetic tools, trying to reverse the second law of thermodynamics. The end of the machine has to be both utilitarian and moral-practical in the sense that it serves a utilitarian purpose, while at the same time, it has another end, which is beyond the utilitarian purpose without detaching from it; this higher end cannot be standardized or normalized, but it has the function to guide us in arriving at a universal happiness, as Kant would say (even though it remains an idea of reason). The development of the machine and the opposition between it and nature, thus conceived in the eighteenth century, has been surpassed, and now, in order to think about universal happiness, the question of the machine becomes central. However, the task is not to prove if a machine is capable of being moral or not, but rather to develop an organology of morality"

"Isn’t the universal merely the norm of the winners who pronounce it, and not necessarily that of those who are asked or made to follow it? Such as during the time of the European Enlightenment, Europe became the universal and European norms were propagated throughout the globe. Given the history of colonization and modernization, it is certainly legitimate to question if what is pronounced as universal is not, in fact, suspect of being somewhat hypocritical. Carl Schmitt had something similar in mind when he approached the League of Nations and paraphrased Proudhon by saying ‘whoever invokes humanity wants to cheat’."

"It is true though that in Kant, machines are either used as a proxy to distinguish the specificity and complexity of human existence, or they are already subsumed to his organicism, namely, the technical presupposes the organic, as Georges Canguilhem puts it. It is a vague organology we can interpret in Kant – and by organology, we mean the understanding of technical tools as artificial organs and the study of the integration of such organs into the human body and community. "

"These demands for the good by circumscribing reason make the ‘businessmen [Geschäftsleute]’ of the three higher faculties appear like magicians, as Kant calls them. If one could go to heaven without doing too much good, and one can have more enjoyment without paying too much effort, isn’t this more desirable than its opposite? This might well resonate with what is happening today with the fantasy of technological acceleration and the illusion that almost everything could be resolved through technology: human enhancement, homo deus, algorithmic optimization, quantified self and so on. The transhumanist proposal that lifespan, intelligence and emotion could be technologically enhanced or improved is a wholesale of modern indulgences. The dream of an AGI to resolve all human problems is another version of pursuing enjoyment without paying effort; it is an ‘administrative nihilism’ in its most naive and laziest sense. Kant wants to suggest that the domination of the businessmen will continue undermining reason if the philosophy faculty is not allowed to counteract them publicly."

"The conflict of faculties is Kant’s further development of the implications of his second critique (and inevitably also the third critique), namely the absolute faith in reason as the condition of moral progress – for Kant, true progress is moral progress, and any political revolution which doesn’t lead to moral progress is not desirable." 

"Without the distinction between the sensible and the supersensible, it would be impossible for critical philosophy to elaborate on morality and freedom, because without this distinction, noumenon will be reduced to phenomenon, and the faculty of desire would collapse into mere physiological and empirical inclinations. Numbers and data are essential to understand the world, but the world is far more than number and data."

"We encounter here the problem of self-interest, which could be found in each individual and each institution. Since when we talk about interest, we are more or less talking about empirical inclinations, for example, to earn more profits, to impose higher tarrifs, to buy cheaper resources, to become more efficient or to have a larger budget. These cannot be the interests of pure reason, because it should be directed towards an interest devoid of any empirical interests, that is, pleasure without interest. How could international trade, one of the greediest businesses, be driven by disinterested interest? Here we encounter again the complex relation between the transcendental and the empirical, because without the empirical, the transcendental is empty, and without the transcendental, the empirical is blind. In other words, even though international trade may be self-interest driven, the transcendental principles could still function behind it. And concerning the progress made by international trade, even though it is not yet a moral step, it is a ‘great step toward morality’"

1. In a system of exchange, reciprocity regulates the behaviour of individuals; that is, if you want to receive hospitality, then please offer hospitality in return. If you want to behave badly towards others, then you should expect reciprocal treatment.

2. In a system of exchange, the community governs the behaviour of each individual. When an individual violates the norm, it will not only damage their reputation but will also be condemned.

"The mathematician and computer scientist Giuseppe Longo, in a recent interview, challenges dataism by arguing that the complexity of human cognition is far from being comprehended and, therefore, imitated. He gave an example by evoking the concept of prégnance, which describes the organic structure of a symbol, irreducible to mere data; an important concept shared by Kurt Goldstein, Ernst Cassirer and many Gestalt theorists,

No emotion, no prégnance, no meaning helps the machine to select ‘what matters’. However, in animals, including humans naturally, it is the prégnances that help to stabilize the relevant invariant: the eagle recognizes a mouse running to hide behind a bush from 200 meters away but ignores a photo of it shown from a meter away. The interest in the target object, hunting, or flight motivate the choice of details to be forgotten, the selection of the prégnant invariants for action, just like the meaning of a gesture, the ‘chin thrust’, allows us to recognize Chirac in one line."

"One could, at least, respond with certainty that there have already been long histories concerning the discourse on intelligence – and by putting history in its plural form we want to emphasize that there is no single historical discourse on intelligence in view of diverse cultures and ways of thinking."

"The world is more complex than any codification of the cognitive schema. This difficulty is identified in what John Haugland calls the ‘paradox of mechanisation’, or put it in his own words: ‘If a process or system is mechanical, it can’t reason; if it reasons, it can’t be mechanical’."

Logical atomism

"Reasoning means first of all understanding meanings, while if a mechanical system treats everything as symbols, then it doesn’t really understand the meaning of both these symbols and what is represented by these symbols. Rationalism, in its mechanical form, cannot adequately explain intelligence. This is also why Hubert Dreyfus was able to call the early AI, now known as Good Old-Fashioned AI (named by John Haugland), Cartesian AI. The Cartesian AI is one in which the agent is the subject that attempts to describe the world from the perspective of the subject; everything is seen as what Heidegger calls present-at-hand [Vorhandenheit], namely it is a bearer of properties; it is objectified by the subject according to logical forms intrinsic in language or through cognitive schema of subjective explication, which Dreyfus retrospectively associates with Descartes and Husserl. Or perhaps we might even call it logical atomism, as seems to be the case in Terry Winograd’s characterization:

We are concerned with developing a formalism, or ‘representation,’ with which to describe . . . knowledge. We seek the ‘atoms’ and ‘particles’ of which it is built, and the ‘forces’ that act on it."

As an affirmation of Minsky’s claim that there is no monopoly on the definition of intelligence, Dreyfus showed further that all definitions of intelligence involve various presuppositions which could become problematic under examination, including Minsky’s own definition:

"Even a chair is not understandable in terms of any set of facts or ‘elements of knowledge.’ To recognize an object as a chair, for example, means to understand its relation to other objects and to human beings. This involves a whole context of human activity of which the shape of our body, the institution of furniture, the inevitability of fatigue, constitute only a small part. In assuming that what is given are facts at all, Minsky is simply echoing a view which has been developing since Plato and has now become so ingrained as to seem self-evident." (Hubert L. Dreyfus, What Computers Cannot Do: A Critique of Artificial Reason (New York: Harper & Row, 1972), 122–3.)

"A counterexample to Minsky’s theory of intelligence that Dreyfus discussed throughout his work is what Heidegger calls ready-to-hand [Zuhandenheit]: a thing in the world, such as a tool, cannot be merely thematized and isolated as an object, namely, objectively described as something standing against the subject as the German word for object Gegenstand literally means. Instead of being seen as a bearer of properties, the tool [Zeug] must be understood through its equipmentality [Zeughaftigkeit], characterized by its ‘in-order-to’ [um-zu] structure. This equipmentality implies a network of references [Verweisungszusammenhang] that we navigate seamlessly in daily life, allowing us to handle tools without the need for explicit objectification. This phenomenon is often referred to as embodiment: the world is embodied within the subject, while simultaneously, the subject is embedded within the world. What is at play is not merely the immediate perception of the world but also a background network of references. For example, when we open a door, we typically do not examine the properties of the doorknob or the door itself; rather, we engage with the door through an intuitive understanding of its place within a network of references already embodied in human Dasein."

Spatiality as separate from sensing and even perception, it is embodied in our Dasein

"This network of references also defines what we might call spatiality, in contrast to space as homogeneous volume. It would be a mistake to merely analyse an object according to space, since its presence is a constellation of relations, which could only be apprehended as spatiality. A good example of spatiality, which Heidegger uses in Being and Time, would be when the power cuts at home and it results in complete darkness, one is still able to orient oneself according to the network of references remembered without being able to see the objects. Heidegger’s critique of the history of ontology reveals a ‘non-cognitive precondition of all understanding’ as Dreyfus quotes from Heidegger’s Being and Time:

[T]he ‘in-order-to,’ the ‘for-the-sake-of,’ and the ‘with-which’ of an involvement . . . resist any sort of mathematical functionalization; nor are they merely something thought, first posted in an ‘act of thinking.’ They are rather relationships in which concernful circumspection as such already dwells."

"Though each school developed their own theory of consciousness, both rationalism and empiricism fail to explain intelligence. Indeed, one tends to conflate consciousness with intelligence; however, being conscious doesn’t imply being intelligent. To be conscious is to be conscious of something, either the self or an external object. There is no intelligence without consciousness, however, consciousness alone cannot explain intelligence, which carries the meaning ‘to understand’, ‘to discern’."

"Traditional rationalism as a principle of intelligence was rejected in the history of AI after the rise of connectionism; empiricism became an alternative thinking power, or as we may call it: a Humean machine. A Humean machine is more than just the inductive method mechanized, as one might be inclined to understand it. Instead, it is one which operates according to the rules of associations, namely resemblance, contiguity in time and space, and causality. Humean associationism is a pattern language which constantly searches for possibilities of association according to principles. In comparison with the Cartesian machine, which starts with an absolute ‘I think’ so that all logical developments can be assured, and that the world manipulated by the malicious demon can be ignored, the Humean machine starts with contingent experience and constantly develops itself according to the strength and degree of associations. In the Humean machine, instead of substance and concept, impressions and ideas reign. The source of impression and idea is ‘matter of fact’; we can say that ideas are faded impressions, and they stay in our memory and re-emerge in our imagination. Impressions and ideas are associated according to relations: contiguity, resemblance and causality. The most famous controversial relation is causality. For a rationalist, causality means necessity; otherwise, there is no law to follow, and the world will be conquered by contingency, as Leibniz’s principle of sufficient reason clearly demands that nihil est sine ratione; for an empiricist, the relation between cause and effect depends on constant conjunctions. In other words, the necessary connexion between a cause and an effect is determined first by contiguity (effect next to cause) and sequence (cause prior to effect), but most importantly by constant conjunction, namely the association between the cause and effect doesn’t discover and produce anything new."

"When Judea Pearl criticized that the current form of AI doesn’t yet know what causality is,36 it is not that it doesn’t know, but rather that such a concept of causality has no meaning other than the frequency and probability of occurrence, namely what is important is correlation but not the rationalist concept of causality."

"Today, it is most apparent that if we were to consider the recommender algorithm of many platforms then what is recommended is calculated according to parameters such as your contextual information, for example, location and time (contiguity), your preferences (resemblance) and the habits you exhibited on the platform (causality). Hume’s doctrine of associationism basically dissolved substance into relations, and by doing so, he also prioritizes intensity and duration as the criteria for truth."

"Human beings live in a field of probabilities. When I go to a café, I choose between either an americano, a latte, a cappuccino or some other choice. Today I choose an americano, but will I choose it again tomorrow? What is the probability that I choose an americano again? Another person just came in, ordered a cappuccino and flirted with the waitress who never smiled at me; how would it affect my choice tomorrow? What is the probability that I order a cappuccino tomorrow instead because of the contingent encounter with this other person? It could in theory turn out that tomorrow I would come and ask for a glass of soy milk which is not available on the menu. However, this probability is lower than not coming tomorrow at all, since it is beyond me to add a new item to the menu of the café. Therefore, the probability of choosing a glass of soy milk can be effectively ignored – namely, it will not be included in the set of probabilities though it remains possible. Formally, everything in a system that is not contradictory is possible, but not everything is probable within the system; therefore, the improbable is still possible, but it is the least probable; the impossible is still thinkable under the condition that the system is transgressed, thus one can talk about the possibility of the impossible. We might transgress these spaces of probabilities; however, these transgressions could be normalized after the event happens, its frequencies have increased, especially once they are then integrated back into the space of probabilities. AI systems like ChatGPT dynamically define a space of probabilities. One can envision a landscape where associations fluctuate in relation to each other, and ChatGPT has the capacity – given sufficient data – to predict the most probable next outcome. Since the absolute is unattainable, as predictions are never certain but merely probable, the optimal result will always be an approximation. In other words, the focus shifts from calculating definitive results to dynamically configuring a field of probabilities. The essence of intelligence, as Yann LeCun claims, is prediction"

"Before we can proceed further, we have to recognize that there is no clear definition of intelligence, in the same way as there is no clear definition of reason, in Kant’s Critiques. The term ‘intelligence’ [Intelligenz] is mentioned several times in the Critique of Pure Reason, sometimes referring to the ‘I’, the thinking subject [Ich, als Intelligenz und denkend Subjekt], sometimes to the spontaneity [Spontaneität] of determination according to the understanding,46 and sometimes the supreme intelligence is invoked. If we limit ourselves to the first Critique, intelligence could be understood as the capacity of producing coherent mental representations (according to intuitions and the understanding thus described in the transcendental analytic) as well as to the autonomy of reason in terms of its capacity to discern, indeed as the Latin term intelligere etymologically implies. Inter- ‘between’ + legere ‘to choose, pick, or read’,47 expresses itself in the capacity of analysing aporia, resolving antinomies, in order to avoid logical illusions (as stated in the transcendental dialectic). Looking at the entire project of the three Critiques, intelligence could be approached from the faculty of knowing, the faculty of feeling and the faculty of desire, however, intelligence is not predicated to each faculty, but rather, these faculties work together under the supervision of reason."

"The problem with empiricism, which we find throughout Kant’s criticism, is that empiricism can hardly speak about truth, nor can it speak about reason, since it is conditional and always directed by the inclination driven by external authorities or interests."

Kant's intelligence as transcendental - not rational nor empirical

"If Kant’s critique of empiricism is valid, then we will have to re-evaluate current artificial intelligence with respect to his transcendental philosophy and vice versa. For the efficiency of our discussion, we will present, on behalf of Kant, the question of intelligence in terms of two claims:

(1) Intelligence is neither rational nor empirical; it is transcendental.

Kant’s criticism of rationalism and empiricism is both recognized and contested by historians of philosophy. In terms of the language of machines, we can say that (1) intelligence cannot be mechanical, that is, subordinated to mere mechanical rules; (2) intelligence cannot be the result of the aggregation of facts, but rather, all the facts which are presented as phenomena and perceived via the pure intuitions (space and time), should be submitted to the understanding. The first statement is a critique against dogmatism as well as mechanism; the second statement is known as Kant’s Copernican turn, in the sense that scientific knowledge is not merely an affair of the external world independent from the human subject; but rather, in so far as scientific knowledge is derived from phenomena, it is also subordinated to the principles of the faculties of cognition."

"Kant’s categories of understanding and pure intuitions are given a priori as the original ground of experience. However, this doesn’t mean that they are mechanical. The role of the understanding is irreducible to the subsumption of sense data to mechanical rules; rather, it also involves something which cannot be analytically presented, for example, the production of imagination. The active production of imagination is a spontaneous process in which the manifold of intuition is synthetized to create a unified, meaningful representation that corresponds to the categories of the understanding. What Kant introduces is a dynamic operation starting with the pure concepts and intuitions without giving them the authority to mechanically (thus passively) determine experience."

"First, without the empirical, the transcendental categories would be useless, in other words, ‘the categories serve only for the possibility of empirical cognition’; this is the proof [Nachweis] of the transcendental deduction since it shows that the categories Kant deduced as transcendental condition can be verified by examining our experience of objects. Second, it is also the place where reason sets off from in order to arrive at the universal – such a universal [Allgemeine] is not a result of induction based on a vast amount of data, but rather the self-positing of reason and its active negation of empirical inclinations. The phenomenal realm is not without importance; however, it is not where reason is grounded; instead, experience is the touchstone [Probierstein] of reason, and it is reason which allows us to orient ourselves in the empirical world without falling prey to unreason. In the same sense as the transcendental faculties, empirical experience also constitutes a tribunal."

"One will have to recognize that association in machine learning is much richer and more precise than what Hume described in his The Treatise of Human Understanding, for example, when we compare the Boltzmann machine (an early neural network application designed in the 1980s)112 and the Hume machine, they seem to be two different beings; however, when we look into the internal structure of a neural network, it is not difficult to see that the weights assigned to the neurones are indicators of intensity and durability, though they are calibrated according to sophisticated learning methods (e.g. reinforcement learning [award and penalty], supervised learning [labelled data], unsupervised learning [unlabled data], etc.). Now, what we call artificial intelligence, and that which surprised us with their ability to deal with language, seems to be a technical proof of the cognitive model advanced by the empiricists. With the proliferation of AI and machine learning algorithms, we have entered into a society of total facts, namely that the society is perceived as nothing more than the accumulation of data."

"ChatGPT doesn’t know, but it can guess what might be the most appropriate word to display next; or as Stephan Wolfram says, ‘ChatGPT is “merely” pulling out some “coherent thread of text” from the “statistics of conventional wisdom” that it’s accumulated’"

"Retrospectively, we can see that in Searle’s experiment, he still assumes that machines are operating mechanically and linearly, without taking into consideration the change in the nature of machines which we discussed in the previous chapter, that is, from linearity to recursivity, mechanical machine to cybernetic machine. Turing’s response to this stereotype is that such a criticism is a solipsism in the sense that one can only judge a being intelligent by being that being."

"This solipsistic argument, which is at the same time epistemological and ontological, has been used to attack the attempt to create models of deep learning or neural networks, for since one doesn’t know exactly how the brain works, how then could one construct an artificial brain?"