Cognitive Dialectics and Artificial Intelligence

Temps de lecture : 9 minutes

Over the long term, the only indisputable and cumulative progress is that of knowledge (of techno-science). Of course, this does not mean that everyone has access to it, nor that there will not be setbacks, knowledge that is forgotten or suppressed in the short term before being rediscovered. However, despite what some may claim, this objective progress is largely independent of us, imposed by experience that most of the time contradicts our beliefs. Far from achieving absolute knowledge, however, this accumulation of knowledge reveals new unknown territories each time, so much so that we can say that ignorance grows as our knowledge destroys our old certainties and prejudices.

Indeed, science only progresses through the discovery of facts that escape current theories, forcing them to be reformulated. Thus, its step-by-step advances demonstrate that we do not have direct access to reality as a definitive, revealed truth that has become transparent, but that there is a cognitive dialectic at work, each time correcting the previous position and complicating it through partial negation or a paradigm shift that cannot be anticipated in advance. This cognitive dialectic based on the state of knowledge clearly illustrates that there is no access to being, as Montaigne already said, there are only approximations, approaches, like a blind man's cane moving from left to right in order to locate its object and test its limits (Fichte defines knowledge as the encounter of a free self that collides with the non-self that resists it). There is an irreconcilable dualism between knowledge and reality, to which it must adjust by trial and error. Knowledge is not original, immediate, direct, or instinctive, but is developed little by little over time, becoming more precise, complete, and nuanced.

In science, even the most established theories are therefore not immune to total reinterpretation, as relativity reinterpreted Newtonian physics, without this constituting a definitive conclusion (the problem of quantum gravity, etc.). This is not an argument in favor of the most far-fetched or spiritualist theories that challenge science, but rather an argument for not trusting our convictions and suspending our judgment by remaining open to questioning our representations through experience (rather than the equations themselves, which remain verified within their domain of validity). We can see that what the history of science implies is its temporality, just like the evolution of life and our own learning, between the legacy of the past and the discoveries of the future that are impossible to predict, passing like a child through a series of stages of development and becoming richer over time. There is therefore no reason why we should stop progressing and why time should stop. If Hegel had made this assumption, it was only for the sake of access to “absolute knowledge,” which was in reality nothing more than the overcoming of religion by self-consciousness as a product of history, and not a total omniscience that is unthinkable for finite beings (even artificial ones).

This reminds us that it is doubtful that the situation could be completely different with Artificial Intelligence, even if it surpasses us in everything, being able to exploit all available data and significantly accelerate the progress of knowledge. They undoubtedly constitute a new cognitive stage with access to knowledge for all (after Wikipedia), but we cannot expect a sudden revelation of (hidden) truth, a “singularity” beyond which there would be no more, as if time had stopped. Of course, let us repeat, AI should produce a great deal of new knowledge or equations by finding unnoticed correlations between distant fields, and could even lead to revolutionary theories that go beyond our preconceived ideas. However, no more than humans, it would be unable to invent effective laws and do without experience—as if there were no longer any external world, uncertainties, or questions. Similarly, despite its universal skills, artificial intelligence, even if fed by the best sources, cannot play the role of an omniscient God and ultimate guarantor of truth. At best, it can take on the role of a judge of first instance, one might say, who is fairly reliable and very useful but who can be wrong and overruled by higher courts.

The limitations of generative AI shed light on our own limitations, or rather on the objective laws of learning and an unavoidable cognitive dialectic that cannot skip steps. We must admit that this is not due to insufficient capabilities, as one might believe, but rather to the exteriority of the world to knowledge (its transcendence), and to the need for feedback for all perception (fine-tuning, adjustment, complexification), the test of reality revealing only after the fact the “perverse effects” or blind spots of a previous state and requiring constant error correction (the principle of cybernetics). As Aristotle and Kant knew, reason cannot deduce being; knowledge needs perception and verification. The recognition of this fundamental dualism, of the division between thought and Being, suggests that instead of an impossible end to history, we should rather witness an accelerated co-evolution of AI with the material and social world, exposed to immense catastrophes but also, as in nature, to an arms race between hostile AI. This is indeed the beginning of a new era rather than a definitive end: there is still much to be done...

Following on from the previous article, it is important to emphasize the need to differentiate between the various dialectical processes according to their object, which cannot be reduced to a simple homogeneous negativity and must be placed in their particular context, such as cognitive dialectics in this case. This should not be confused with other historical dialectics, particularly political ones, which swing from one extreme to the other according to different laws. However, they are not unrelated, as they also have a cognitive dimension, and we may wonder what influence these AIs, which far surpass our limited minds, will have on politics and democracy. Given that our congenital stupidity is at the root of the worst excesses, it would be extraordinary if we could use superior intelligence to avoid demagoguery and identity politics, but we shouldn't get our hopes up.

Obscurantism, not only in religion, is not going to disappear as if by magic, especially since AI tends to pick up our worst tendencies because it is trained on our all-too-human controversies. Much is also expected of these enhanced cognitive abilities in the ecological sphere, both in terms of reducing climate skepticism and determining the most appropriate measures to take. However, this is already what the IPCC was supposed to do, and it has not prevented the widespread resistance and denial of the coalition of all interests and various conspiracy theorists. Similarly, while fake news should not, in theory, survive the fast-checking of well-informed AI, we can unfortunately expect other malicious AI to multiply it, making it increasingly difficult to detect deception!

Finally, while I have argued that wars can be considered a cognitive error—overestimating one's own strength and underestimating that of one's opponent—could they not be avoided by AI that assesses the situation more objectively? No doubt in a number of cases, but it is not certain that we can have blind faith in rational predictions of the outcome of battle, which remain probabilities. Here again, it is impossible to do without the test of reality. It must be decided by force of arms, a kind of “judgment of God” between material powers.

Despite the extraordinary cognitive revolution of large language models that are disrupting almost every field, once the initial surprise and wonder have passed, we can therefore judge, from a historical perspective, that in the end there will be nothing new under the sun, at least in terms of global unification (economic, ecological, technological, demographic, pandemic, and media), which seems to be yet another material process with consequences for geopolitics. Fundamentally, we will never be rid of ignorance and stupidity, just as we will never be rid of the entropy that eats away at us and that we must always learn to overcome but which will never disappear—no matter what means are deployed, which are increasingly considerable in an increasingly threatening environment...

4 vues